diff --git a/.secrets.baseline b/.secrets.baseline index 50594377b1..3755c2172b 100644 --- a/.secrets.baseline +++ b/.secrets.baseline @@ -8480,5 +8480,5 @@ } ] }, - "generated_at": "2026-04-23T21:12:19Z" + "generated_at": "2026-04-24T03:31:30Z" } diff --git a/RELEASE.md b/RELEASE.md index f18f02a6bd..49e0659cd1 100644 --- a/RELEASE.md +++ b/RELEASE.md @@ -39,9 +39,28 @@ This step also usually lasts about a week. After QA and bugfixing are complete for both OSS and Desktop: * Final releases are cut from their respective RC branches. -* Release timing is coordinated with Langflow’s DevRel team. +* Release timing is coordinated with Langflow's DevRel team. * For at least 24 hours after release, Discord, GitHub, and other support channels should be monitored for critical bug reports. +### 4. Release Artifacts + +The release workflow automatically publishes the following artifacts: + +* **PyPI Packages:** + * `langflow` - Main package with all integrations + * `langflow-base` - Core framework without integrations + * `lfx` - Lightweight executor CLI + * `langflow-sdk` - SDK for programmatic access (when updated) + +* **Docker Images:** + * `langflowai/langflow` - Full Langflow image + * `langflowai/langflow-backend` - Backend-only image (published independently) + * `langflowai/langflow-frontend` - Frontend-only image (published independently) + * `langflowai/langflow-ep` - Enterprise edition image (published independently) + * `langflowai/langflow-base` - Base image without integrations + +**Note:** Backend, frontend, and enterprise images are published separately from the main image and will be built even if the main version already exists on Docker Hub. + ## Branch Model | Branch | Purpose | Merge Policy | @@ -101,6 +120,10 @@ git merge --ff-only release-X.Y.Z # Fast-forward main to include RC changes * Follows [Semantic Versioning](https://semver.org): `MAJOR.MINOR.PATCH`. * RC tags use `-rc.N`, e.g. `v1.8.0-rc.1`. +* **All tags MUST start with `v` prefix** (e.g., `v1.9.1`, not `1.9.1`). + * The release workflow validates this format and rejects tags without the `v` prefix. + * Duplicate tags (e.g., both `1.8.3` and `v1.8.3`) cause GitHub's release notes generation to use the wrong base comparison, resulting in incomplete changelogs. + * The workflow automatically checks for and prevents duplicate tags. ## Roles diff --git a/docs/docs/Deployment/deployment-kubernetes-dev.mdx b/docs/docs/Deployment/deployment-kubernetes-dev.mdx index 4a657e6af5..7817a46e11 100644 --- a/docs/docs/Deployment/deployment-kubernetes-dev.mdx +++ b/docs/docs/Deployment/deployment-kubernetes-dev.mdx @@ -131,6 +131,8 @@ langflow: enabled: true driver: value: "postgresql" + host: + value: "postgresql-svc.langflow.svc.cluster.local" port: value: "5432" user: diff --git a/docs/versioned_docs/version-1.8.0/Deployment/deployment-kubernetes-dev.mdx b/docs/versioned_docs/version-1.8.0/Deployment/deployment-kubernetes-dev.mdx index 4a657e6af5..7817a46e11 100644 --- a/docs/versioned_docs/version-1.8.0/Deployment/deployment-kubernetes-dev.mdx +++ b/docs/versioned_docs/version-1.8.0/Deployment/deployment-kubernetes-dev.mdx @@ -131,6 +131,8 @@ langflow: enabled: true driver: value: "postgresql" + host: + value: "postgresql-svc.langflow.svc.cluster.local" port: value: "5432" user: diff --git a/docs/versioned_docs/version-1.9.0/Deployment/deployment-kubernetes-dev.mdx b/docs/versioned_docs/version-1.9.0/Deployment/deployment-kubernetes-dev.mdx index 4a657e6af5..7817a46e11 100644 --- a/docs/versioned_docs/version-1.9.0/Deployment/deployment-kubernetes-dev.mdx +++ b/docs/versioned_docs/version-1.9.0/Deployment/deployment-kubernetes-dev.mdx @@ -131,6 +131,8 @@ langflow: enabled: true driver: value: "postgresql" + host: + value: "postgresql-svc.langflow.svc.cluster.local" port: value: "5432" user: diff --git a/src/backend/base/langflow/alembic/versions/mb00a1b2c3d4_add_memory_base_schema.py b/src/backend/base/langflow/alembic/versions/mb00a1b2c3d4_add_memory_base_schema.py new file mode 100644 index 0000000000..c921370472 --- /dev/null +++ b/src/backend/base/langflow/alembic/versions/mb00a1b2c3d4_add_memory_base_schema.py @@ -0,0 +1,235 @@ +"""add_memory_base_schema + +Consolidates all Memory Base schema changes into a single migration: + - job.dedupe_key (nullable String) + ix_job_dedupe_key + - message.run_id (nullable UUID) + ix_message_run_id + - message.is_output (bool, default false) + - memory_base table + ix_memory_base_flow_id + ix_memory_base_user_id + - memory_base_session table + three indexes + - message_ingestion_record table + three indexes + - memory_base_workflow_run table + two indexes + +Phase: EXPAND + +Revision ID: mb00a1b2c3d4 +Revises: d306e5c17c41 +Create Date: 2026-04-14 00:00:00.000000 +""" + +from collections.abc import Sequence + +import sqlalchemy as sa +from alembic import op +from langflow.utils import migration + +# revision identifiers, used by Alembic. +revision: str = "mb00a1b2c3d4" # pragma: allowlist secret +down_revision: str | None = "d306e5c17c41" # pragma: allowlist secret +branch_labels: str | Sequence[str] | None = None +depends_on: str | Sequence[str] | None = None + + +def upgrade() -> None: + conn = op.get_bind() + + # ------------------------------------------------------------------ # + # job.dedupe_key # + # ------------------------------------------------------------------ # + inspector = sa.inspect(conn) + existing_job_indexes = {idx["name"] for idx in inspector.get_indexes("job")} + with op.batch_alter_table("job", schema=None) as batch_op: + if not migration.column_exists("job", "dedupe_key", conn): + batch_op.add_column(sa.Column("dedupe_key", sa.String(), nullable=True)) + if "ix_job_dedupe_key" not in existing_job_indexes: + batch_op.create_index(batch_op.f("ix_job_dedupe_key"), ["dedupe_key"], unique=False) + + # ------------------------------------------------------------------ # + # message.run_id + message.is_output # + # ------------------------------------------------------------------ # + with op.batch_alter_table("message", schema=None) as batch_op: + if not migration.column_exists("message", "run_id", conn): + batch_op.add_column(sa.Column("run_id", sa.Uuid(), nullable=True)) + if not migration.column_exists("message", "is_output", conn): + batch_op.add_column(sa.Column("is_output", sa.Boolean(), nullable=False, server_default=sa.text("false"))) + + existing_message_indexes = {idx["name"] for idx in sa.inspect(conn).get_indexes("message")} + if "ix_message_run_id" not in existing_message_indexes: + op.create_index("ix_message_run_id", "message", ["run_id"]) + + # ------------------------------------------------------------------ # + # memory_base # + # ------------------------------------------------------------------ # + if not migration.table_exists("memory_base", conn): + op.create_table( + "memory_base", + sa.Column("id", sa.Uuid(), nullable=False), + sa.Column("name", sa.String(), nullable=False), + sa.Column("flow_id", sa.Uuid(), nullable=False), + sa.Column("user_id", sa.Uuid(), nullable=False), + sa.Column("threshold", sa.Integer(), nullable=False, server_default=sa.text("50")), + sa.Column("auto_capture", sa.Boolean(), nullable=False, server_default=sa.text("true")), + sa.Column("embedding_model", sa.String(), nullable=False, server_default=sa.text("''")), + sa.Column("preprocessing", sa.Boolean(), nullable=False, server_default=sa.text("false")), + sa.Column("preproc_model", sa.String(), nullable=True), + sa.Column("preproc_instructions", sa.String(), nullable=True), + sa.Column("kb_name", sa.String(), nullable=False), + sa.Column("created_at", sa.DateTime(timezone=True), nullable=False), + sa.PrimaryKeyConstraint("id"), + sa.UniqueConstraint("user_id", "name", name="uq_memory_base_user_name"), + sa.Index("ix_memory_base_flow_id", "flow_id"), + sa.Index("ix_memory_base_user_id", "user_id"), + ) + + # ------------------------------------------------------------------ # + # memory_base_session # + # ------------------------------------------------------------------ # + if not migration.table_exists("memory_base_session", conn): + op.create_table( + "memory_base_session", + sa.Column("id", sa.Uuid(), nullable=False), + sa.Column( + "memory_base_id", + sa.Uuid(), + sa.ForeignKey("memory_base.id", ondelete="CASCADE"), + nullable=False, + ), + sa.Column("session_id", sa.String(), nullable=False), + sa.Column("cursor_id", sa.Uuid(), nullable=True), + sa.Column("total_processed", sa.Integer(), nullable=False, server_default=sa.text("0")), + sa.Column("last_sync_at", sa.DateTime(timezone=True), nullable=True), + sa.PrimaryKeyConstraint("id"), + sa.UniqueConstraint("memory_base_id", "session_id", name="uq_memory_base_session"), + ) + op.create_index("ix_memory_base_session_memory_base_id", "memory_base_session", ["memory_base_id"]) + op.create_index("ix_memory_base_session_session_id", "memory_base_session", ["session_id"]) + op.create_index( + "ix_memory_base_session_lookup", + "memory_base_session", + ["memory_base_id", "session_id"], + ) + + # ------------------------------------------------------------------ # + # message_ingestion_record # + # ------------------------------------------------------------------ # + if not migration.table_exists("message_ingestion_record", conn): + op.create_table( + "message_ingestion_record", + sa.Column("id", sa.Uuid(), nullable=False), + sa.Column( + "message_id", + sa.Uuid(), + sa.ForeignKey("message.id", ondelete="CASCADE"), + nullable=False, + ), + sa.Column( + "memory_base_id", + sa.Uuid(), + sa.ForeignKey("memory_base.id", ondelete="CASCADE"), + nullable=False, + ), + sa.Column( + "job_id", + sa.Uuid(), + sa.ForeignKey("job.job_id", ondelete="SET NULL"), + nullable=True, + ), + sa.Column("session_id", sa.String(), nullable=False), + sa.Column("ingested_at", sa.DateTime(timezone=True), nullable=False), + sa.PrimaryKeyConstraint("id"), + sa.UniqueConstraint( + "message_id", + "session_id", + "memory_base_id", + name="uq_mir_message_session_mb", + ), + ) + op.create_index("ix_mir_message_id", "message_ingestion_record", ["message_id"]) + op.create_index("ix_mir_job_id", "message_ingestion_record", ["job_id"]) + op.create_index( + "ix_mir_memory_base_session", + "message_ingestion_record", + ["memory_base_id", "session_id"], + ) + + # ------------------------------------------------------------------ # + # memory_base_workflow_run # + # ------------------------------------------------------------------ # + if not migration.table_exists("memory_base_workflow_run", conn): + op.create_table( + "memory_base_workflow_run", + sa.Column("id", sa.Uuid(), nullable=False), + sa.Column( + "memory_base_id", + sa.Uuid(), + sa.ForeignKey("memory_base.id", ondelete="CASCADE"), + nullable=False, + ), + sa.Column("session_id", sa.String(), nullable=False), + sa.Column( + "workflow_job_id", + sa.Uuid(), + sa.ForeignKey("job.job_id", ondelete="SET NULL"), + nullable=True, + ), + sa.Column( + "ingestion_job_id", + sa.Uuid(), + sa.ForeignKey("job.job_id", ondelete="SET NULL"), + nullable=True, + ), + sa.Column("recorded_at", sa.DateTime(timezone=True), nullable=False), + sa.PrimaryKeyConstraint("id"), + sa.UniqueConstraint( + "memory_base_id", + "session_id", + "workflow_job_id", + name="uq_mbwr_mb_session_wf_job", + ), + ) + op.create_index("ix_mbwr_mb_session", "memory_base_workflow_run", ["memory_base_id", "session_id"]) + op.create_index("ix_mbwr_ingestion_job_id", "memory_base_workflow_run", ["ingestion_job_id"]) + + +def downgrade() -> None: + conn = op.get_bind() + + # Children first (FK dependencies) ----------------------------------- # + if migration.table_exists("memory_base_workflow_run", conn): + op.drop_index("ix_mbwr_ingestion_job_id", table_name="memory_base_workflow_run") + op.drop_index("ix_mbwr_mb_session", table_name="memory_base_workflow_run") + op.drop_table("memory_base_workflow_run") + + if migration.table_exists("message_ingestion_record", conn): + op.drop_index("ix_mir_memory_base_session", table_name="message_ingestion_record") + op.drop_index("ix_mir_job_id", table_name="message_ingestion_record") + op.drop_index("ix_mir_message_id", table_name="message_ingestion_record") + op.drop_table("message_ingestion_record") + + if migration.table_exists("memory_base_session", conn): + op.drop_index("ix_memory_base_session_lookup", table_name="memory_base_session") + op.drop_index("ix_memory_base_session_session_id", table_name="memory_base_session") + op.drop_index("ix_memory_base_session_memory_base_id", table_name="memory_base_session") + op.drop_table("memory_base_session") + + if migration.table_exists("memory_base", conn): + op.drop_index("ix_memory_base_user_id", table_name="memory_base") + op.drop_index("ix_memory_base_flow_id", table_name="memory_base") + op.drop_table("memory_base") + + # Message column/index ----------------------------------------------- # + existing_message_indexes = {idx["name"] for idx in sa.inspect(conn).get_indexes("message")} + if "ix_message_run_id" in existing_message_indexes: + op.drop_index("ix_message_run_id", table_name="message") + with op.batch_alter_table("message", schema=None) as batch_op: + if migration.column_exists("message", "is_output", conn): + batch_op.drop_column("is_output") + if migration.column_exists("message", "run_id", conn): + batch_op.drop_column("run_id") + + # Job column/index --------------------------------------------------- # + with op.batch_alter_table("job", schema=None) as batch_op: + existing_job_indexes = {idx["name"] for idx in sa.inspect(conn).get_indexes("job")} + if "ix_job_dedupe_key" in existing_job_indexes: + batch_op.drop_index(batch_op.f("ix_job_dedupe_key")) + if migration.column_exists("job", "dedupe_key", conn): + batch_op.drop_column("dedupe_key") diff --git a/src/backend/base/langflow/api/build.py b/src/backend/base/langflow/api/build.py index 7ed4feffd6..a35c69c6d4 100644 --- a/src/backend/base/langflow/api/build.py +++ b/src/backend/base/langflow/api/build.py @@ -29,7 +29,15 @@ from langflow.exceptions.component import ComponentBuildError from langflow.schema.message import ErrorMessage from langflow.schema.schema import OutputValue from langflow.services.database.models.flow.model import Flow -from langflow.services.deps import get_chat_service, get_telemetry_service, session_scope +from langflow.services.database.models.jobs.model import JobType +from langflow.services.deps import ( + get_chat_service, + get_job_service, + get_memory_base_service, + get_task_service, + get_telemetry_service, + session_scope, +) from langflow.services.job_queue.service import JobQueueNotFoundError, JobQueueService from langflow.services.telemetry.schema import ComponentInputsPayload, ComponentPayload, PlaygroundPayload @@ -527,35 +535,81 @@ async def generate_flow_events( event_manager.on_error(data=error_message.data) raise + # Create a WORKFLOW job record so memory-base on_flow_output can track this run. + # Best-effort: failures here must never break the build path. + _build_job_svc = None + _build_run_id: uuid.UUID | None = None + try: + _build_run_id = uuid.UUID(graph.run_id) if graph.run_id else None + if _build_run_id is not None: + _build_job_svc = get_job_service() + await _build_job_svc.create_job( + job_id=_build_run_id, + flow_id=flow_id, + user_id=current_user.id, + job_type=JobType.WORKFLOW, + ) + except Exception: # noqa: BLE001 + await logger.awarning( + "Failed to create workflow job for /build — memory base tracking disabled for flow %s", + flow_id, + exc_info=True, + ) + _build_job_svc = None + event_manager.on_vertices_sorted(data={"ids": ids, "to_run": vertices_to_run}) vertex_timedeltas: list[float] = [] event_manager.on_build_start(data={}) - tasks = [] - for vertex_id in ids: - task = asyncio.create_task(build_vertices(vertex_id, graph, event_manager, vertex_timedeltas)) - tasks.append(task) - try: - await asyncio.gather(*tasks) - except asyncio.CancelledError: - background_tasks.add_task(graph.end_all_traces_in_context()) - raise - except Exception as e: - await logger.aerror(f"Error building vertices: {e}") - custom_component = graph.get_vertex(vertex_id).custom_component - trace_name = getattr(custom_component, "trace_name", None) - error_message = ErrorMessage( - flow_id=flow_id, - exception=e, - session_id=graph.session_id, - trace_name=trace_name, - ) - event_manager.on_error(data=error_message.data) - raise + + async def _run_vertex_build() -> None: + tasks = [] + for vertex_id in ids: + task = asyncio.create_task(build_vertices(vertex_id, graph, event_manager, vertex_timedeltas)) + tasks.append(task) + try: + await asyncio.gather(*tasks) + except asyncio.CancelledError: + background_tasks.add_task(graph.end_all_traces_in_context()) + raise + except Exception as e: + await logger.aerror(f"Error building vertices: {e}") + custom_component = graph.get_vertex(vertex_id).custom_component + trace_name = getattr(custom_component, "trace_name", None) + error_message = ErrorMessage( + flow_id=flow_id, + exception=e, + session_id=graph.session_id, + trace_name=trace_name, + ) + event_manager.on_error(data=error_message.data) + raise + + if _build_job_svc and _build_run_id: + await _build_job_svc.execute_with_status(_build_run_id, _run_vertex_build) + else: + await _run_vertex_build() build_duration = sum(vertex_timedeltas) event_manager.on_end(data={"build_duration": build_duration}) await graph.end_all_traces() + + # Fire memory-base auto-capture hook — non-blocking background effect. + # Must use fire_and_forget_task (not background_tasks.add_task) because + # generate_flow_events runs as an asyncio task; by the time the flow + # finishes, FastAPI has already drained the background_tasks queue and any + # tasks added after that point are silently dropped. + try: + _run_id_uuid = uuid.UUID(graph.run_id) if graph.run_id else None # type-cast only; same run_id set on graph + await get_task_service().fire_and_forget_task( + get_memory_base_service().on_flow_output, + flow_id=flow_id, + session_id=graph.session_id or str(flow_id), + job_id=_run_id_uuid, + ) + except (RuntimeError, ValueError, OSError): + await logger.awarning("Memory base hook scheduling failed for flow %s", flow_id, exc_info=True) + await event_manager.queue.put((None, None, time.time())) diff --git a/src/backend/base/langflow/api/router.py b/src/backend/base/langflow/api/router.py index c4a38d44fa..bfb8b45c97 100644 --- a/src/backend/base/langflow/api/router.py +++ b/src/backend/base/langflow/api/router.py @@ -15,6 +15,7 @@ from langflow.api.v1 import ( login_router, mcp_projects_router, mcp_router, + memories_router, model_options_router, models_router, monitor_router, @@ -68,6 +69,7 @@ router_v1.include_router(folders_router) router_v1.include_router(projects_router) router_v1.include_router(starter_projects_router) router_v1.include_router(knowledge_bases_router) +router_v1.include_router(memories_router) router_v1.include_router(mcp_router) router_v1.include_router(voice_mode_router) router_v1.include_router(mcp_projects_router) diff --git a/src/backend/base/langflow/api/utils/kb_helpers.py b/src/backend/base/langflow/api/utils/kb_helpers.py index b195777f19..63a1c0190b 100644 --- a/src/backend/base/langflow/api/utils/kb_helpers.py +++ b/src/backend/base/langflow/api/utils/kb_helpers.py @@ -445,7 +445,7 @@ class KBIngestionHelper: splitter_kwargs["separators"] = [resolved_separator] text_splitter = RecursiveCharacterTextSplitter(**splitter_kwargs) - embeddings = await KBIngestionHelper._build_embeddings(embedding_provider, embedding_model, current_user) + embeddings = await KBIngestionHelper.build_embeddings(embedding_provider, embedding_model, current_user) client = KBStorageHelper.get_fresh_chroma_client(kb_path) chroma = Chroma( @@ -462,40 +462,30 @@ class KBIngestionHelper: continue chunks = text_splitter.split_text(content) - for i in range(0, len(chunks), INGESTION_BATCH_SIZE): - if await KBIngestionHelper._is_job_cancelled(job_service, task_job_id): - raise IngestionCancelledError + docs = [ + Document( + page_content=c, + metadata={ + "source": source_name or file_name, + "file_name": file_name, + "chunk_index": i, + "total_chunks": len(chunks), + "ingested_at": datetime.now(timezone.utc).isoformat(), + "job_id": job_id_str, + }, + ) + for i, c in enumerate(chunks) + ] - batch = chunks[i : i + INGESTION_BATCH_SIZE] - docs = [ - Document( - page_content=c, - metadata={ - "source": source_name or file_name, - "file_name": file_name, - "chunk_index": i + j, - "total_chunks": len(chunks), - "ingested_at": datetime.now(timezone.utc).isoformat(), - "job_id": job_id_str, - }, - ) - for j, c in enumerate(batch) - ] - - for attempt in range(MAX_RETRY_ATTEMPTS): - if await KBIngestionHelper._is_job_cancelled(job_service, task_job_id): - raise IngestionCancelledError - try: - await chroma.aadd_documents(docs) - break - except Exception as e: - if attempt == MAX_RETRY_ATTEMPTS - 1: - raise - wait = (attempt + 1) * EXPONENTIAL_BACKOFF_MULTIPLIER - await logger.awarning("Write failed, retrying in %ds: %s", wait, e) - await asyncio.sleep(wait) - - await asyncio.sleep(0.01) + written = await KBIngestionHelper.write_documents_to_chroma( + documents=docs, + chroma=chroma, + task_job_id=task_job_id, + job_service=job_service, + ) + if written < len(docs): + # Job was cancelled mid-file + raise IngestionCancelledError total_chunks_created += len(chunks) processed_files.append(file_name) @@ -555,13 +545,69 @@ class KBIngestionHelper: KBStorageHelper.release_chroma_resources(kb_path) @staticmethod - async def _is_job_cancelled(job_service: JobService, job_id: uuid.UUID) -> bool: + async def write_documents_to_chroma( + *, + documents: list[Document], + chroma: Chroma, + task_job_id: uuid.UUID, + job_service: JobService, + ) -> int: + """Write pre-built Documents into an open Chroma collection. + + This is the shared primitive used by both file-based KB ingestion + (``perform_ingestion``) and message-based Memory Base ingestion. + + Documents must already be chunked and have their metadata populated + by the caller — this method only handles the batched write, cancellation + checking, and retry logic. + + Args: + documents: LangChain Document objects ready for embedding. + chroma: An already-constructed ``Chroma`` instance pointing at the + target collection. + task_job_id: Job ID used to poll for cancellation. + job_service: Service for checking job status. + + Returns: + Number of documents successfully written. If the job is cancelled + mid-batch this will be less than ``len(documents)``. + + Raises: + Exception: Re-raises any non-cancellation write failure after the + retry budget is exhausted. + """ + written = 0 + for i in range(0, len(documents), INGESTION_BATCH_SIZE): + if await KBIngestionHelper.is_job_cancelled(job_service, task_job_id): + return written + + batch = documents[i : i + INGESTION_BATCH_SIZE] + for attempt in range(MAX_RETRY_ATTEMPTS): + if await KBIngestionHelper.is_job_cancelled(job_service, task_job_id): + return written + try: + await chroma.aadd_documents(batch) + break + except Exception as e: + if attempt == MAX_RETRY_ATTEMPTS - 1: + raise + wait = (attempt + 1) * EXPONENTIAL_BACKOFF_MULTIPLIER + await logger.awarning("Write failed, retrying in %ds: %s", wait, e) + await asyncio.sleep(wait) + + written += len(batch) + await asyncio.sleep(0.01) + + return written + + @staticmethod + async def is_job_cancelled(job_service: JobService, job_id: uuid.UUID) -> bool: """Internal helper to check if a job has been cancelled.""" job = await job_service.get_job_by_job_id(job_id) return job is not None and job.status == JobStatus.CANCELLED @staticmethod - async def _build_embeddings(provider: str, model: str, current_user): + async def build_embeddings(provider: str, model: str, current_user): """Internal helper to build embeddings object.""" options = get_embedding_model_options(user_id=current_user.id) selected_option = next((o for o in options if o["provider"] == provider and o["name"] == model), None) diff --git a/src/backend/base/langflow/api/v1/__init__.py b/src/backend/base/langflow/api/v1/__init__.py index 177fc34df0..11dcb2816b 100644 --- a/src/backend/base/langflow/api/v1/__init__.py +++ b/src/backend/base/langflow/api/v1/__init__.py @@ -10,6 +10,7 @@ from langflow.api.v1.knowledge_bases import router as knowledge_bases_router from langflow.api.v1.login import router as login_router from langflow.api.v1.mcp import router as mcp_router from langflow.api.v1.mcp_projects import router as mcp_projects_router +from langflow.api.v1.memories import router as memories_router from langflow.api.v1.model_options import router as model_options_router from langflow.api.v1.models import router as models_router from langflow.api.v1.monitor import router as monitor_router @@ -36,6 +37,7 @@ __all__ = [ "login_router", "mcp_projects_router", "mcp_router", + "memories_router", "model_options_router", "models_router", "monitor_router", diff --git a/src/backend/base/langflow/api/v1/endpoints.py b/src/backend/base/langflow/api/v1/endpoints.py index 0a077f0504..a137c15543 100644 --- a/src/backend/base/langflow/api/v1/endpoints.py +++ b/src/backend/base/langflow/api/v1/endpoints.py @@ -6,7 +6,7 @@ import time from collections.abc import AsyncGenerator from http import HTTPStatus from typing import TYPE_CHECKING, Annotated -from uuid import uuid4 +from uuid import UUID, uuid4 import orjson import sqlalchemy as sa @@ -62,8 +62,17 @@ from langflow.services.auth.utils import ( from langflow.services.cache.utils import save_uploaded_file from langflow.services.database.models.flow.model import Flow, FlowRead from langflow.services.database.models.flow.utils import get_all_webhook_components_in_flow +from langflow.services.database.models.jobs.model import JobType from langflow.services.database.models.user.model import User, UserRead -from langflow.services.deps import get_auth_service, get_session_service, get_settings_service, get_telemetry_service +from langflow.services.deps import ( + get_auth_service, + get_job_service, + get_memory_base_service, + get_session_service, + get_settings_service, + get_task_service, + get_telemetry_service, +) from langflow.services.event_manager import create_webhook_event_manager, webhook_event_manager from langflow.services.telemetry.schema import RunPayload from langflow.utils.compression import compress_response @@ -173,8 +182,8 @@ async def simple_run_flow( graph = Graph.from_payload( graph_data, flow_id=flow_id_str, user_id=str(user_id), flow_name=flow.name, context=context ) - if run_id is None: - run_id = str(uuid4()) + run_id_uuid = uuid4() if run_id is None else UUID(run_id) + run_id = str(run_id_uuid) graph.set_run_id(run_id) inputs = None if input_request.input_value is not None: @@ -197,15 +206,57 @@ async def simple_run_flow( and (input_request.output_type == "any" or input_request.output_type in vertex.id.lower()) # type: ignore[operator] ) ] - task_result, session_id = await run_graph_internal( - graph=graph, - flow_id=flow_id_str, - session_id=input_request.session_id, - inputs=inputs, - outputs=outputs, - stream=stream, - event_manager=event_manager, - ) + + # Create a WORKFLOW job record so memory-base on_flow_output can track this run. + if user_id is None: + raise HTTPException( + status_code=status.HTTP_401_UNAUTHORIZED, + detail="Authentication required to run flows.", + ) + + try: + _job_svc = get_job_service() + await _job_svc.create_job( + job_id=run_id_uuid, + flow_id=flow.id, + user_id=user_id, + job_type=JobType.WORKFLOW, + ) + task_result, session_id = await _job_svc.execute_with_status( + run_id_uuid, + run_graph_internal, + graph=graph, + flow_id=flow_id_str, + session_id=input_request.session_id, + inputs=inputs, + outputs=outputs, + stream=stream, + event_manager=event_manager, + ) + except Exception as exc: + await logger.aerror( + "Workflow job execution failed for flow %s: %s", + flow.id, + str(exc), + exc_info=True, + ) + raise APIException( + status_code=status.HTTP_500_INTERNAL_SERVER_ERROR, + exception=exc, + flow=flow, + ) from exc + + # Fire memory-base auto-capture hook — non-blocking background effect. + try: + _run_id_uuid = UUID(graph.run_id) if graph.run_id else None # type-cast only + await get_task_service().fire_and_forget_task( + get_memory_base_service().on_flow_output, + flow_id=flow.id, + session_id=session_id, + job_id=_run_id_uuid, + ) + except (RuntimeError, ValueError, OSError): + await logger.awarning("Memory base hook scheduling failed for flow %s", flow.id, exc_info=True) return RunResponse(outputs=task_result, session_id=session_id) @@ -970,6 +1021,18 @@ async def experimental_run_flow( except Exception as exc: raise HTTPException(status_code=status.HTTP_500_INTERNAL_SERVER_ERROR, detail=str(exc)) from exc + # Fire memory-base auto-capture hook — non-blocking background effect. + try: + _run_id_uuid = UUID(graph.run_id) if graph.run_id else None # type-cast only + await get_task_service().fire_and_forget_task( + get_memory_base_service().on_flow_output, + flow_id=flow.id, + session_id=session_id, + job_id=_run_id_uuid, + ) + except (RuntimeError, ValueError, OSError): + await logger.awarning("Memory base hook scheduling failed for flow %s", flow.id, exc_info=True) + return RunResponse(outputs=task_result, session_id=session_id) diff --git a/src/backend/base/langflow/api/v1/knowledge_bases.py b/src/backend/base/langflow/api/v1/knowledge_bases.py index 62ddbf41a1..6c9490edcc 100644 --- a/src/backend/base/langflow/api/v1/knowledge_bases.py +++ b/src/backend/base/langflow/api/v1/knowledge_bases.py @@ -74,6 +74,30 @@ def _resolve_kb_path(kb_name: str, current_user: CurrentActiveUser) -> Path: return kb_path +def _is_memory_base_associated(metadata: dict[str, Any]) -> bool: + """Return True if the KB metadata indicates an association with a Memory Base.""" + source_types = metadata.get("source_types") + return isinstance(source_types, list) and "memory" in source_types + + +def _check_memory_base_association(kb_name: str, current_user: CurrentActiveUser) -> None: + """Raise 403 if the KB is associated with a Memory Base. + + Designed as a FastAPI dependency for per-KB routes — FastAPI injects + ``kb_name`` from the path parameter and ``current_user`` via its own + dependency. The list endpoint filters memory KBs inline using + ``_is_memory_base_associated`` directly. + """ + kb_path = _resolve_kb_path(kb_name, current_user) + + metadata = KBAnalysisHelper.get_metadata(kb_path, fast=True) + if _is_memory_base_associated(metadata): + raise HTTPException( + status_code=403, + detail=f"Access denied: knowledge base '{kb_name}' is managed by a Memory Base.", + ) + + @router.post("", status_code=HTTPStatus.CREATED) @router.post("/", status_code=HTTPStatus.CREATED) async def create_knowledge_base( @@ -275,7 +299,7 @@ async def preview_chunks( return {"files": file_previews} -@router.post("/{kb_name}/ingest", status_code=HTTPStatus.OK) +@router.post("/{kb_name}/ingest", status_code=HTTPStatus.OK, dependencies=[Depends(_check_memory_base_association)]) async def ingest_files_to_knowledge_base( kb_name: str, current_user: CurrentActiveUser, @@ -427,6 +451,8 @@ async def list_knowledge_bases( try: # Use deep update (fast=False) to ensure legacy KBs are migrated on first view metadata = KBAnalysisHelper.get_metadata(kb_dir, fast=False) + if _is_memory_base_associated(metadata): + continue # Skip KBs that are associated with a Memory Base # Extract KB ID from metadata (stored as string, convert to UUID) kb_id_str = metadata.get("id") @@ -503,7 +529,7 @@ async def list_knowledge_bases( return knowledge_bases -@router.get("/{kb_name}", status_code=HTTPStatus.OK) +@router.get("/{kb_name}", status_code=HTTPStatus.OK, dependencies=[Depends(_check_memory_base_association)]) async def get_knowledge_base(kb_name: str, current_user: CurrentActiveUser) -> KnowledgeBaseInfo: """Get detailed information about a specific knowledge base.""" try: @@ -543,7 +569,7 @@ async def get_knowledge_base(kb_name: str, current_user: CurrentActiveUser) -> K raise HTTPException(status_code=500, detail="Error getting knowledge base.") from e -@router.get("/{kb_name}/chunks", status_code=HTTPStatus.OK) +@router.get("/{kb_name}/chunks", status_code=HTTPStatus.OK, dependencies=[Depends(_check_memory_base_association)]) async def get_knowledge_base_chunks( kb_name: str, current_user: CurrentActiveUser, @@ -636,7 +662,7 @@ async def get_knowledge_base_chunks( KBStorageHelper.release_chroma_resources(kb_path) -@router.delete("/{kb_name}", status_code=HTTPStatus.OK) +@router.delete("/{kb_name}", status_code=HTTPStatus.OK, dependencies=[Depends(_check_memory_base_association)]) async def delete_knowledge_base(kb_name: str, current_user: CurrentActiveUser) -> dict[str, str]: """Delete a specific knowledge base.""" try: @@ -703,7 +729,7 @@ async def delete_knowledge_bases_bulk(request: BulkDeleteRequest, current_user: return result -@router.post("/{kb_name}/cancel", status_code=HTTPStatus.OK) +@router.post("/{kb_name}/cancel", status_code=HTTPStatus.OK, dependencies=[Depends(_check_memory_base_association)]) async def cancel_ingestion( kb_name: str, current_user: CurrentActiveUser, diff --git a/src/backend/base/langflow/api/v1/memories.py b/src/backend/base/langflow/api/v1/memories.py new file mode 100644 index 0000000000..176efceb7f --- /dev/null +++ b/src/backend/base/langflow/api/v1/memories.py @@ -0,0 +1,345 @@ +"""REST API for Memory Base management. + +Endpoints: + POST /memories - Create + GET /memories - List (current user, paginated) + GET /memories/{id} - Get one + GET /memories/{id}/sessions - List sessions (tracked + untracked from MessageTable) + PATCH /memories/{id} - Update (name / threshold / auto_capture / preprocessing) + DELETE /memories/{id} - Delete (cancels active tasks + removes KB from disk) + POST /memories/{id}/flush - Manual flush / trigger ingestion + POST /memories/{id}/regenerate - Regenerate from mismatch + +Edge cases enforced: + 409 Conflict - name already in use for this user (on create). + 409 Conflict - active ingestion task already running for same (mb, session). + 404 Not Found - memory base does not belong to the current user. + 422 Unprocessable - preprocessing=true but preproc_model missing. +""" + +from __future__ import annotations + +import uuid +from datetime import datetime +from http import HTTPStatus +from typing import Annotated, Any + +from fastapi import APIRouter, Body, Depends, HTTPException +from fastapi_pagination import Page, Params +from fastapi_pagination.ext.sqlmodel import apaginate +from pydantic import BaseModel +from sqlmodel import col, select + +from langflow.api.utils import CurrentActiveUser +from langflow.services.database.models.memory_base.model import ( + MemoryBase, + MemoryBaseCreate, + MemoryBaseRead, + MemoryBaseSessionRead, + MemoryBaseUpdate, +) +from langflow.services.database.models.message.model import MessageTable +from langflow.services.deps import get_memory_base_service, session_scope +from langflow.services.jobs import DuplicateJobError + +router = APIRouter(tags=["Memories"], prefix="/memories", include_in_schema=False) + + +# ------------------------------------------------------------------ # +# Request / Response schemas # +# ------------------------------------------------------------------ # + + +class MessageReadResponse(BaseModel): + """Slim message projection for Memory Base session message listings. + + Only messages that have been ingested into the requested Memory Base are returned. + ``job_id`` and ``ingested_at`` are sourced from MessageIngestionRecord. + """ + + model_config = {"from_attributes": True} + + id: uuid.UUID + timestamp: datetime | None = None + sender: str + sender_name: str + session_id: str + text: str + content_blocks: list[dict[str, Any]] = [] + job_id: uuid.UUID | None = None + ingested_at: datetime | None = None + + +class FlushRequest(BaseModel): + session_id: str + + +class MismatchResponse(BaseModel): + mismatch_detected: bool + + +class RegenerateResponse(BaseModel): + job_ids: list[str] + + +# ------------------------------------------------------------------ # +# CRUD # +# ------------------------------------------------------------------ # + + +@router.post("", status_code=HTTPStatus.CREATED) +@router.post("/", status_code=HTTPStatus.CREATED) +async def create_memory_base( + current_user: CurrentActiveUser, + payload: Annotated[MemoryBaseCreate, Body(embed=False)] = ..., +) -> MemoryBaseRead: + """Create a new Memory Base. + + - kb_name is auto-generated as `{sanitized_name}_{8hex}`. + - KB directory and embedding_metadata.json are created on disk immediately. + - Returns 409 if a Memory Base with the same name already exists for this user. + - Returns 422 if preprocessing=true but preproc_model is missing. + """ + try: + mb = await get_memory_base_service().create(payload, user_id=current_user.id) + except PermissionError as exc: + # Flow not found or belongs to another user — return 404 to avoid info-leak + raise HTTPException(status_code=404, detail=str(exc)) from exc + except ValueError as exc: + raise HTTPException(status_code=409, detail=str(exc)) from exc + return MemoryBaseRead.model_validate(mb) + + +@router.get("", status_code=HTTPStatus.OK) +@router.get("/", status_code=HTTPStatus.OK) +async def list_memory_bases( + current_user: CurrentActiveUser, + params: Annotated[Params, Depends()], + flow_id: uuid.UUID | None = None, +) -> Page[MemoryBaseRead]: + """List all Memory Bases owned by the current user (paginated) for a flow_id. + + Query params (from fastapi-pagination): + page - 1-based page number (default 1) + size - page size (default 50) + """ + async with session_scope() as db: + stmt = get_memory_base_service().list_for_user_stmt(user_id=current_user.id, flow_id=flow_id) + return await apaginate( + db, stmt, params=params, transformer=lambda items: [MemoryBaseRead.model_validate(m) for m in items] + ) + + +@router.get("/{memory_base_id}", status_code=HTTPStatus.OK) +async def get_memory_base( + memory_base_id: uuid.UUID, + current_user: CurrentActiveUser, +) -> MemoryBaseRead: + """Get details for a specific Memory Base.""" + mb = await get_memory_base_service().get(memory_base_id, user_id=current_user.id) + if mb is None: + raise HTTPException(status_code=404, detail="Memory base not found") + return MemoryBaseRead.model_validate(mb) + + +@router.get("/{memory_base_id}/sessions", status_code=HTTPStatus.OK) +async def list_sessions( + memory_base_id: uuid.UUID, + current_user: CurrentActiveUser, + params: Annotated[Params, Depends()], +) -> Page[MemoryBaseSessionRead]: + """List persisted sessions for this Memory Base (paginated). + + Only sessions that have been synced at least once (i.e. have a + MemoryBaseSession row) are returned. Results are ordered by + last_sync_at descending. + + Each item includes ``pending_count``: the number of completed flow runs + remaining before the next auto-capture ingestion is triggered. + """ + from langflow.services.memory_base.ingestion import count_pending_messages + + async with session_scope() as db: + try: + mb = await get_memory_base_service().get_memory_base_or_404(db, memory_base_id, current_user.id) + except ValueError as exc: + raise HTTPException(status_code=404, detail=str(exc)) from exc + + stmt = get_memory_base_service().sessions_stmt(memory_base_id, current_user.id) + raw_page = await apaginate(db, stmt, params=params) + + items: list[MemoryBaseSessionRead] = [] + for s in raw_page.items: + pending_count = await count_pending_messages(db, mb, s) + read = MemoryBaseSessionRead.model_validate(s) + read.pending_count = pending_count + items.append(read) + + return raw_page.model_copy(update={"items": items}) + + +@router.get("/{memory_base_id}/sessions/{session_id}/messages", status_code=HTTPStatus.OK) +async def list_session_messages( + memory_base_id: uuid.UUID, + session_id: str, + current_user: CurrentActiveUser, + params: Annotated[Params, Depends()], +) -> Page[MessageReadResponse]: + """List messages ingested into this Memory Base session (paginated). + + Only messages that have been successfully ingested into the requested Memory Base + are returned. Messages are ordered by timestamp ascending. + Each item includes ``job_id`` and ``ingested_at`` from the MessageIngestionRecord. + + Returns 404 if the Memory Base does not belong to the current user. + """ + from sqlalchemy import and_ + + from langflow.services.database.models.memory_base.model import MessageIngestionRecord + + async with session_scope() as db: + mb_stmt = select(MemoryBase).where(MemoryBase.id == memory_base_id).where(MemoryBase.user_id == current_user.id) + result = await db.exec(mb_stmt) + if result.first() is None: + raise HTTPException(status_code=404, detail="Memory base not found") + + # INNER JOIN — only messages that were actually ingested into this MB/session pair. + # No extra WHERE filters needed: + # - mir.session_id == session_id in the JOIN guarantees msg.session_id == session_id + # (session_id is denormalized from the message at ingestion time — immutable). + # - flow_id is implicitly correct: ingestion only ever touches messages from mb.flow_id, + # and MB ownership is already verified above. + msg_stmt = ( + select(MessageTable, MessageIngestionRecord) + .join( + MessageIngestionRecord, + and_( + MessageIngestionRecord.message_id == MessageTable.id, + MessageIngestionRecord.memory_base_id == memory_base_id, + MessageIngestionRecord.session_id == session_id, + ), + ) + .order_by(col(MessageTable.timestamp).asc()) + ) + return await apaginate( + db, + msg_stmt, + params=params, + transformer=lambda rows: [ + MessageReadResponse( + id=msg.id, + timestamp=msg.timestamp, + sender=msg.sender, + sender_name=msg.sender_name, + session_id=msg.session_id, + text=msg.text, + content_blocks=msg.content_blocks or [], + job_id=mir.job_id, + ingested_at=mir.ingested_at, + ) + for msg, mir in rows + ], + ) + + +@router.patch("/{memory_base_id}", status_code=HTTPStatus.OK) +async def update_memory_base( + memory_base_id: uuid.UUID, + current_user: CurrentActiveUser, + patch: Annotated[MemoryBaseUpdate, Body(embed=False)] = ..., +) -> MemoryBaseRead: + """Update mutable parameters (threshold, auto_capture, preprocessing, etc.). + + Threshold changes only take effect at the next auto-capture trigger. + Any already-running ingestion task continues with its original arguments. + """ + mb = await get_memory_base_service().update(memory_base_id, user_id=current_user.id, patch=patch) + if mb is None: + raise HTTPException(status_code=404, detail="Memory base not found") + return MemoryBaseRead.model_validate(mb) + + +@router.delete("/{memory_base_id}", status_code=HTTPStatus.NO_CONTENT) +async def delete_memory_base( + memory_base_id: uuid.UUID, + current_user: CurrentActiveUser, +) -> None: + """Delete a Memory Base. + + Active ingestion tasks are forcefully cancelled before the DB record is + removed. The associated KB directory is deleted from disk afterwards + (best-effort — a disk failure will not affect the 204 response). + """ + deleted = await get_memory_base_service().delete(memory_base_id, user_id=current_user.id) + if not deleted: + raise HTTPException(status_code=404, detail="Memory base not found") + + +# ------------------------------------------------------------------ # +# Ingestion trigger # +# ------------------------------------------------------------------ # + + +@router.post("/{memory_base_id}/flush", status_code=HTTPStatus.ACCEPTED) +async def flush_memory_base( + memory_base_id: uuid.UUID, + current_user: CurrentActiveUser, + body: Annotated[FlushRequest, Body(embed=False)] = ..., +) -> dict: + """Manually trigger an ingestion / sync job regardless of the threshold. + + Returns 409 Conflict if an ingestion task is already in progress for the + given (memory_base_id, session_id) pair to prevent concurrent indexing. + """ + try: + job_id = await get_memory_base_service().trigger_ingestion( + memory_base_id=memory_base_id, + user_id=current_user.id, + session_id=body.session_id, + ) + except ValueError as exc: + raise HTTPException(status_code=404, detail=str(exc)) from exc + except DuplicateJobError as exc: + raise HTTPException(status_code=409, detail=str(exc)) from exc + except RuntimeError as exc: + raise HTTPException(status_code=409, detail=str(exc)) from exc + + return {"job_id": job_id} + + +# ------------------------------------------------------------------ # +# Mismatch detection & regeneration # +# ------------------------------------------------------------------ # + + +@router.get("/{memory_base_id}/mismatch", status_code=HTTPStatus.OK) +async def check_mismatch( + memory_base_id: uuid.UUID, + current_user: CurrentActiveUser, +) -> MismatchResponse: + """Detect if the vector store is empty while metadata records processed messages. + + The UI should surface a "Mismatch Detected" warning and offer a Regenerate button. + """ + try: + detected = await get_memory_base_service().check_mismatch(memory_base_id, user_id=current_user.id) + except ValueError as exc: + raise HTTPException(status_code=404, detail=str(exc)) from exc + return MismatchResponse(mismatch_detected=detected) + + +@router.post("/{memory_base_id}/regenerate", status_code=HTTPStatus.ACCEPTED) +async def regenerate_memory_base( + memory_base_id: uuid.UUID, + current_user: CurrentActiveUser, +) -> RegenerateResponse: + """Regenerate the Knowledge Base by resetting all session cursors and re-ingesting. + + Use this to recover from external Chroma directory deletions or vector DB corruption. + All MemoryBaseSession.cursor_id values are set to None before re-running ingestion. + """ + try: + job_ids = await get_memory_base_service().regenerate(memory_base_id, user_id=current_user.id) + except ValueError as exc: + raise HTTPException(status_code=404, detail=str(exc)) from exc + return RegenerateResponse(job_ids=job_ids) diff --git a/src/backend/base/langflow/api/v2/workflow.py b/src/backend/base/langflow/api/v2/workflow.py index 8bd7749ed1..3d513a2be6 100644 --- a/src/backend/base/langflow/api/v2/workflow.py +++ b/src/backend/base/langflow/api/v2/workflow.py @@ -29,6 +29,7 @@ from uuid import UUID, uuid4 from fastapi import APIRouter, BackgroundTasks, Depends, HTTPException, Query, Request, status from fastapi.responses import StreamingResponse from lfx.graph.graph.base import Graph +from lfx.log.logger import logger from lfx.schema.workflow import ( WORKFLOW_EXECUTION_RESPONSES, WORKFLOW_STATUS_RESPONSES, @@ -65,7 +66,7 @@ from langflow.services.auth.utils import api_key_security from langflow.services.database.models.flow.model import FlowRead from langflow.services.database.models.jobs.model import JobType from langflow.services.database.models.user.model import UserRead -from langflow.services.deps import get_job_service, get_task_service +from langflow.services.deps import get_job_service, get_memory_base_service, get_task_service # Configuration constants EXECUTION_TIMEOUT = 300 # 5 minutes default timeout for sync execution @@ -391,6 +392,18 @@ async def execute_sync_workflow( stream=False, ) + # Fire memory-base auto-capture hook — non-blocking background effect. + try: + _run_id_uuid = UUID(graph.run_id) if graph.run_id else None # type-cast only; same run_id set on graph + await get_task_service().fire_and_forget_task( + get_memory_base_service().on_flow_output, + flow_id=flow.id, + session_id=execution_session_id, + job_id=_run_id_uuid, + ) + except (RuntimeError, ValueError, OSError): + await logger.awarning("Memory base hook scheduling failed for flow %s", flow.id, exc_info=True) + # Build RunResponse run_response = RunResponse(outputs=task_result, session_id=execution_session_id) # Convert to WorkflowExecutionResponse @@ -470,10 +483,39 @@ async def execute_workflow_background( user_id=api_key_user.id, ) + # Closure captures flow identity for the memory-base hook. + # run_id is the same as job_id — graph.set_run_id(job_id) was called above. + _hook_flow_id = flow.id + _hook_run_id = job_id + + async def _run_and_notify(**kwargs): + """Thin wrapper: execute graph then fire memory-base hook as a background effect. + + The hook is dispatched non-blocking after graph completion. Any failure in + the hook is swallowed so it never affects the job status of the graph run. + """ + result = await run_graph_internal(**kwargs) + _, _effective_session_id = result + try: + # Direct await — we are already inside a background task; awaiting here + # is non-blocking from the client's perspective and avoids the race + # condition that arises when dispatching a second fire_and_forget from + # within an already-running fire_and_forget task. + await get_memory_base_service().on_flow_output( + flow_id=_hook_flow_id, + session_id=_effective_session_id, + job_id=_hook_run_id, + ) + except Exception: # noqa: BLE001 + await logger.awarning( + "Memory base hook failed for flow %s, but workflow succeeded.", _hook_flow_id, exc_info=True + ) + return result + await task_service.fire_and_forget_task( job_service.execute_with_status, job_id=job_id, - run_coro_func=run_graph_internal, + run_coro_func=_run_and_notify, graph=graph, flow_id=flow_id_str, session_id=session_id, @@ -673,7 +715,7 @@ async def stop_workflow( task_service = get_task_service() try: - # 1. Fetch Job + # 1. Fetch Job and verify ownership job = await job_service.get_job_by_job_id(job_id, user_id=api_key_user.id) except Exception as exc: raise HTTPException( diff --git a/src/backend/base/langflow/initial_setup/starter_projects/Instagram Copywriter.json b/src/backend/base/langflow/initial_setup/starter_projects/Instagram Copywriter.json index b1887db7fa..d53ea45b95 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/Instagram Copywriter.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/Instagram Copywriter.json @@ -2064,14 +2064,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -2112,6 +2113,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -2229,7 +2250,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -2532,7 +2553,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/Invoice Summarizer.json b/src/backend/base/langflow/initial_setup/starter_projects/Invoice Summarizer.json index 267b5b3905..a60ff3a096 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/Invoice Summarizer.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/Invoice Summarizer.json @@ -1172,14 +1172,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -1220,6 +1221,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -1337,7 +1358,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -1640,7 +1661,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/Market Research.json b/src/backend/base/langflow/initial_setup/starter_projects/Market Research.json index 9f994436a5..e47ce65eca 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/Market Research.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/Market Research.json @@ -1184,14 +1184,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -1232,6 +1233,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -1349,7 +1370,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -1652,7 +1673,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/News Aggregator.json b/src/backend/base/langflow/initial_setup/starter_projects/News Aggregator.json index 451071cccd..fe141b3d81 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/News Aggregator.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/News Aggregator.json @@ -1166,14 +1166,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -1214,6 +1215,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -1331,7 +1352,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -1634,7 +1655,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/Nvidia Remix.json b/src/backend/base/langflow/initial_setup/starter_projects/Nvidia Remix.json index 62beb48acf..56e47f629b 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/Nvidia Remix.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/Nvidia Remix.json @@ -793,14 +793,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2026-03-20T22:35:04.094Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -841,6 +842,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -960,7 +981,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -1265,7 +1286,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/Pokédex Agent.json b/src/backend/base/langflow/initial_setup/starter_projects/Pokédex Agent.json index 1926a20af3..ff1c6d3795 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/Pokédex Agent.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/Pokédex Agent.json @@ -1231,14 +1231,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -1279,6 +1280,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -1396,7 +1417,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -1699,7 +1720,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/Price Deal Finder.json b/src/backend/base/langflow/initial_setup/starter_projects/Price Deal Finder.json index a66552640d..2300722dac 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/Price Deal Finder.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/Price Deal Finder.json @@ -1600,14 +1600,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -1648,6 +1649,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -1765,7 +1786,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -2068,7 +2089,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/Research Agent.json b/src/backend/base/langflow/initial_setup/starter_projects/Research Agent.json index 2f844458fe..bc6ec24990 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/Research Agent.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/Research Agent.json @@ -2803,14 +2803,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -2851,6 +2852,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -2968,7 +2989,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -3271,7 +3292,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/SaaS Pricing.json b/src/backend/base/langflow/initial_setup/starter_projects/SaaS Pricing.json index 6719f62cae..0f0a743fec 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/SaaS Pricing.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/SaaS Pricing.json @@ -885,14 +885,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -933,6 +934,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -1050,7 +1071,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -1353,7 +1374,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/Search agent.json b/src/backend/base/langflow/initial_setup/starter_projects/Search agent.json index bad7080ed0..634f14d007 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/Search agent.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/Search agent.json @@ -932,14 +932,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -980,6 +981,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -1097,7 +1118,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -1400,7 +1421,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/Sequential Tasks Agents.json b/src/backend/base/langflow/initial_setup/starter_projects/Sequential Tasks Agents.json index 31c4dd41d7..f8250b5b9a 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/Sequential Tasks Agents.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/Sequential Tasks Agents.json @@ -350,14 +350,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -398,6 +399,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -515,7 +536,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -818,7 +839,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], @@ -938,14 +959,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -986,6 +1008,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -1103,7 +1145,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -1406,7 +1448,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], @@ -2383,14 +2425,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -2431,6 +2474,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -2548,7 +2611,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -2851,7 +2914,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/Simple Agent.json b/src/backend/base/langflow/initial_setup/starter_projects/Simple Agent.json index 8d4f41685c..6a64299bbb 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/Simple Agent.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/Simple Agent.json @@ -931,14 +931,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2026-02-12T20:48:13.965Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -979,6 +980,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -1097,7 +1118,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -1402,7 +1423,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/Social Media Agent.json b/src/backend/base/langflow/initial_setup/starter_projects/Social Media Agent.json index 600ed33b36..50742c3033 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/Social Media Agent.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/Social Media Agent.json @@ -1281,14 +1281,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -1329,6 +1330,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -1446,7 +1467,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -1749,7 +1770,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/Travel Planning Agents.json b/src/backend/base/langflow/initial_setup/starter_projects/Travel Planning Agents.json index 5083843f2c..197a09fbcb 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/Travel Planning Agents.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/Travel Planning Agents.json @@ -1697,14 +1697,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -1745,6 +1746,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -1862,7 +1883,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -2165,7 +2186,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], @@ -2280,14 +2301,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -2328,6 +2350,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -2445,7 +2487,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -2748,7 +2790,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], @@ -2863,14 +2905,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-11T21:41:48.407Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -2911,6 +2954,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -3028,7 +3091,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -3331,7 +3394,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/initial_setup/starter_projects/Youtube Analysis.json b/src/backend/base/langflow/initial_setup/starter_projects/Youtube Analysis.json index 170ec2a4fb..64798ffa1a 100644 --- a/src/backend/base/langflow/initial_setup/starter_projects/Youtube Analysis.json +++ b/src/backend/base/langflow/initial_setup/starter_projects/Youtube Analysis.json @@ -265,7 +265,7 @@ "legacy": false, "lf_version": "1.4.3", "metadata": { - "code_hash": "20398e0d18df", + "code_hash": "8c5296516f6c", "dependencies": { "dependencies": [ { @@ -340,7 +340,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from contextlib import contextmanager\n\nimport pandas as pd\nfrom googleapiclient.discovery import build\nfrom googleapiclient.errors import HttpError\n\nfrom lfx.custom.custom_component.component import Component\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, IntInput, MessageTextInput, SecretStrInput\nfrom lfx.schema.dataframe import DataFrame\nfrom lfx.template.field.base import Output\n\n\nclass YouTubeCommentsComponent(Component):\n \"\"\"A component that retrieves comments from YouTube videos.\"\"\"\n\n display_name: str = \"YouTube Comments\"\n description: str = \"Retrieves and analyzes comments from YouTube videos.\"\n icon: str = \"YouTube\"\n\n # Constants\n COMMENTS_DISABLED_STATUS = 403\n NOT_FOUND_STATUS = 404\n API_MAX_RESULTS = 100\n\n inputs = [\n MessageTextInput(\n name=\"video_url\",\n display_name=\"Video URL\",\n info=\"The URL of the YouTube video to get comments from.\",\n tool_mode=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"YouTube API Key\",\n info=\"Your YouTube Data API key.\",\n required=True,\n ),\n IntInput(\n name=\"max_results\",\n display_name=\"Max Results\",\n value=20,\n info=\"The maximum number of comments to return.\",\n ),\n DropdownInput(\n name=\"sort_by\",\n display_name=\"Sort By\",\n options=[\"time\", \"relevance\"],\n value=\"relevance\",\n info=\"Sort comments by time or relevance.\",\n ),\n BoolInput(\n name=\"include_replies\",\n display_name=\"Include Replies\",\n value=False,\n info=\"Whether to include replies to comments.\",\n advanced=True,\n ),\n BoolInput(\n name=\"include_metrics\",\n display_name=\"Include Metrics\",\n value=True,\n info=\"Include metrics like like count and reply count.\",\n advanced=True,\n ),\n ]\n\n outputs = [\n Output(name=\"comments\", display_name=\"Comments\", method=\"get_video_comments\"),\n ]\n\n def _extract_video_id(self, video_url: str) -> str:\n \"\"\"Extracts the video ID from a YouTube URL.\"\"\"\n import re\n\n patterns = [\n r\"(?:youtube\\.com\\/watch\\?v=|youtu.be\\/|youtube.com\\/embed\\/)([^&\\n?#]+)\",\n r\"youtube.com\\/shorts\\/([^&\\n?#]+)\",\n ]\n\n for pattern in patterns:\n match = re.search(pattern, video_url)\n if match:\n return match.group(1)\n\n return video_url.strip()\n\n def _process_reply(self, reply: dict, parent_id: str, *, include_metrics: bool = True) -> dict:\n \"\"\"Process a single reply comment.\"\"\"\n reply_snippet = reply[\"snippet\"]\n reply_data = {\n \"comment_id\": reply[\"id\"],\n \"parent_comment_id\": parent_id,\n \"author\": reply_snippet[\"authorDisplayName\"],\n \"text\": reply_snippet[\"textDisplay\"],\n \"published_at\": reply_snippet[\"publishedAt\"],\n \"is_reply\": True,\n }\n if include_metrics:\n reply_data[\"like_count\"] = reply_snippet[\"likeCount\"]\n reply_data[\"reply_count\"] = 0 # Replies can't have replies\n\n return reply_data\n\n def _process_comment(\n self, item: dict, *, include_metrics: bool = True, include_replies: bool = False\n ) -> list[dict]:\n \"\"\"Process a single comment thread.\"\"\"\n comment = item[\"snippet\"][\"topLevelComment\"][\"snippet\"]\n comment_id = item[\"snippet\"][\"topLevelComment\"][\"id\"]\n\n # Basic comment data\n processed_comments = [\n {\n \"comment_id\": comment_id,\n \"parent_comment_id\": \"\", # Empty for top-level comments\n \"author\": comment[\"authorDisplayName\"],\n \"author_channel_url\": comment.get(\"authorChannelUrl\", \"\"),\n \"text\": comment[\"textDisplay\"],\n \"published_at\": comment[\"publishedAt\"],\n \"updated_at\": comment[\"updatedAt\"],\n \"is_reply\": False,\n }\n ]\n\n # Add metrics if requested\n if include_metrics:\n processed_comments[0].update(\n {\n \"like_count\": comment[\"likeCount\"],\n \"reply_count\": item[\"snippet\"][\"totalReplyCount\"],\n }\n )\n\n # Add replies if requested\n if include_replies and item[\"snippet\"][\"totalReplyCount\"] > 0 and \"replies\" in item:\n for reply in item[\"replies\"][\"comments\"]:\n reply_data = self._process_reply(reply, parent_id=comment_id, include_metrics=include_metrics)\n processed_comments.append(reply_data)\n\n return processed_comments\n\n @contextmanager\n def youtube_client(self):\n \"\"\"Context manager for YouTube API client.\"\"\"\n client = build(\"youtube\", \"v3\", developerKey=self.api_key)\n try:\n yield client\n finally:\n client.close()\n\n def get_video_comments(self) -> DataFrame:\n \"\"\"Retrieves comments from a YouTube video and returns as DataFrame.\"\"\"\n try:\n # Extract video ID from URL\n video_id = self._extract_video_id(self.video_url)\n\n # Use context manager for YouTube API client\n with self.youtube_client() as youtube:\n comments_data = []\n results_count = 0\n request = youtube.commentThreads().list(\n part=\"snippet,replies\",\n videoId=video_id,\n maxResults=min(self.API_MAX_RESULTS, self.max_results),\n order=self.sort_by,\n textFormat=\"plainText\",\n )\n\n while request and results_count < self.max_results:\n response = request.execute()\n\n for item in response.get(\"items\", []):\n if results_count >= self.max_results:\n break\n\n comments = self._process_comment(\n item, include_metrics=self.include_metrics, include_replies=self.include_replies\n )\n comments_data.extend(comments)\n results_count += 1\n\n # Get the next page if available and needed\n if \"nextPageToken\" in response and results_count < self.max_results:\n request = youtube.commentThreads().list(\n part=\"snippet,replies\",\n videoId=video_id,\n maxResults=min(self.API_MAX_RESULTS, self.max_results - results_count),\n order=self.sort_by,\n textFormat=\"plainText\",\n pageToken=response[\"nextPageToken\"],\n )\n else:\n request = None\n\n # Convert to DataFrame\n comments_df = pd.DataFrame(comments_data)\n\n # Add video metadata\n comments_df[\"video_id\"] = video_id\n comments_df[\"video_url\"] = self.video_url\n\n # Sort columns for better organization\n column_order = [\n \"video_id\",\n \"video_url\",\n \"comment_id\",\n \"parent_comment_id\",\n \"is_reply\",\n \"author\",\n \"author_channel_url\",\n \"text\",\n \"published_at\",\n \"updated_at\",\n ]\n\n if self.include_metrics:\n column_order.extend([\"like_count\", \"reply_count\"])\n\n comments_df = comments_df[column_order]\n\n return DataFrame(comments_df)\n\n except HttpError as e:\n error_message = f\"YouTube API error: {e!s}\"\n if e.resp.status == self.COMMENTS_DISABLED_STATUS:\n error_message = \"Comments are disabled for this video or API quota exceeded.\"\n elif e.resp.status == self.NOT_FOUND_STATUS:\n error_message = \"Video not found.\"\n\n return DataFrame(pd.DataFrame({\"error\": [error_message]}))\n" + "value": "from contextlib import contextmanager\n\nimport pandas as pd\nfrom googleapiclient.discovery import build\nfrom googleapiclient.errors import HttpError\n\nfrom lfx.custom.custom_component.component import Component\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, IntInput, MessageTextInput, SecretStrInput\nfrom lfx.schema.dataframe import DataFrame\nfrom lfx.template.field.base import Output\n\n\nclass YouTubeCommentsComponent(Component):\n \"\"\"A component that retrieves comments from YouTube videos.\"\"\"\n\n display_name: str = \"YouTube Comments\"\n description: str = \"Retrieves and analyzes comments from YouTube videos.\"\n icon: str = \"YouTube\"\n\n # Constants\n COMMENTS_DISABLED_STATUS = 403\n NOT_FOUND_STATUS = 404\n API_MAX_RESULTS = 100\n\n inputs = [\n MessageTextInput(\n name=\"video_url\",\n display_name=\"Video URL\",\n info=\"The URL of the YouTube video to get comments from.\",\n tool_mode=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"YouTube API Key\",\n info=\"Your YouTube Data API key.\",\n required=True,\n ),\n IntInput(\n name=\"max_results\",\n display_name=\"Max Results\",\n value=20,\n info=\"The maximum number of comments to return.\",\n ),\n DropdownInput(\n name=\"sort_by\",\n display_name=\"Sort By\",\n options=[\"time\", \"relevance\"],\n value=\"relevance\",\n info=\"Sort comments by time or relevance.\",\n ),\n BoolInput(\n name=\"include_replies\",\n display_name=\"Include Replies\",\n value=False,\n info=\"Whether to include replies to comments.\",\n advanced=True,\n ),\n BoolInput(\n name=\"include_metrics\",\n display_name=\"Include Metrics\",\n value=True,\n info=\"Include metrics like like count and reply count.\",\n advanced=True,\n ),\n ]\n\n outputs = [\n Output(name=\"comments\", display_name=\"Comments\", method=\"get_video_comments\"),\n ]\n\n def _extract_video_id(self, video_url: str) -> str:\n \"\"\"Extracts the video ID from a YouTube URL.\"\"\"\n import re\n\n patterns = [\n r\"(?:youtube\\.com\\/watch\\?v=|youtu.be\\/|youtube.com\\/embed\\/)([^&\\n?#]+)\",\n r\"youtube.com\\/shorts\\/([^&\\n?#]+)\",\n ]\n\n for pattern in patterns:\n match = re.search(pattern, video_url)\n if match:\n return match.group(1)\n\n return video_url.strip()\n\n def _process_reply(self, reply: dict, parent_id: str, *, include_metrics: bool = True) -> dict:\n \"\"\"Process a single reply comment.\"\"\"\n reply_snippet = reply[\"snippet\"]\n reply_data = {\n \"comment_id\": reply[\"id\"],\n \"parent_comment_id\": parent_id,\n \"author\": reply_snippet[\"authorDisplayName\"],\n \"text\": reply_snippet[\"textDisplay\"],\n \"published_at\": reply_snippet[\"publishedAt\"],\n \"is_reply\": True,\n }\n if include_metrics:\n reply_data[\"like_count\"] = reply_snippet[\"likeCount\"]\n reply_data[\"reply_count\"] = 0 # Replies can't have replies\n\n return reply_data\n\n def _process_comment(\n self, item: dict, *, include_metrics: bool = True, include_replies: bool = False\n ) -> list[dict]:\n \"\"\"Process a single comment thread.\"\"\"\n comment = item[\"snippet\"][\"topLevelComment\"][\"snippet\"]\n comment_id = item[\"snippet\"][\"topLevelComment\"][\"id\"]\n\n # Basic comment data\n processed_comments = [\n {\n \"comment_id\": comment_id,\n \"parent_comment_id\": \"\", # Empty for top-level comments\n \"author\": comment[\"authorDisplayName\"],\n \"author_channel_url\": comment.get(\"authorChannelUrl\", \"\"),\n \"text\": comment[\"textDisplay\"],\n \"published_at\": comment[\"publishedAt\"],\n \"updated_at\": comment[\"updatedAt\"],\n \"is_reply\": False,\n }\n ]\n\n # Add metrics if requested\n if include_metrics:\n processed_comments[0].update(\n {\n \"like_count\": comment[\"likeCount\"],\n \"reply_count\": item[\"snippet\"][\"totalReplyCount\"],\n }\n )\n\n # Add replies if requested\n if include_replies and item[\"snippet\"][\"totalReplyCount\"] > 0 and \"replies\" in item:\n for reply in item[\"replies\"][\"comments\"]:\n reply_data = self._process_reply(reply, parent_id=comment_id, include_metrics=include_metrics)\n processed_comments.append(reply_data)\n\n return processed_comments\n\n @contextmanager\n def youtube_client(self):\n \"\"\"Context manager for YouTube API client.\"\"\"\n client = build(\"youtube\", \"v3\", developerKey=self.api_key)\n try:\n yield client\n finally:\n client.close()\n\n def get_video_comments(self) -> DataFrame:\n \"\"\"Retrieves comments from a YouTube video and returns as DataFrame.\"\"\"\n try:\n # Extract video ID from URL\n video_id = self._extract_video_id(self.video_url)\n\n # Use context manager for YouTube API client\n with self.youtube_client() as youtube:\n comments_data = []\n results_count = 0\n request = youtube.commentThreads().list(\n part=\"snippet,replies\",\n videoId=video_id,\n maxResults=min(self.API_MAX_RESULTS, self.max_results),\n order=self.sort_by,\n textFormat=\"plainText\",\n )\n\n while request and results_count < self.max_results:\n response = request.execute()\n\n for item in response.get(\"items\", []):\n if results_count >= self.max_results:\n break\n\n comments = self._process_comment(\n item, include_metrics=self.include_metrics, include_replies=self.include_replies\n )\n comments_data.extend(comments)\n results_count += 1\n\n # Get the next page if available and needed\n if \"nextPageToken\" in response and results_count < self.max_results:\n request = youtube.commentThreads().list(\n part=\"snippet,replies\",\n videoId=video_id,\n maxResults=min(self.API_MAX_RESULTS, self.max_results - results_count),\n order=self.sort_by,\n textFormat=\"plainText\",\n pageToken=response[\"nextPageToken\"],\n )\n else:\n request = None\n\n # Define column order\n column_order = [\n \"video_id\",\n \"video_url\",\n \"comment_id\",\n \"parent_comment_id\",\n \"is_reply\",\n \"author\",\n \"author_channel_url\",\n \"text\",\n \"published_at\",\n \"updated_at\",\n ]\n\n if self.include_metrics:\n column_order.extend([\"like_count\", \"reply_count\"])\n\n # Handle empty comments case\n if not comments_data:\n # Create empty DataFrame with proper columns\n comments_df = pd.DataFrame(columns=column_order)\n else:\n # Convert to DataFrame\n comments_df = pd.DataFrame(comments_data)\n\n # Add video metadata\n comments_df[\"video_id\"] = video_id\n comments_df[\"video_url\"] = self.video_url\n\n # Reorder columns\n comments_df = comments_df[column_order]\n\n return DataFrame(comments_df)\n\n except HttpError as e:\n error_message = f\"YouTube API error: {e!s}\"\n if e.resp.status == self.COMMENTS_DISABLED_STATUS:\n error_message = \"Comments are disabled for this video or API quota exceeded.\"\n elif e.resp.status == self.NOT_FOUND_STATUS:\n error_message = \"Video not found.\"\n\n return DataFrame(pd.DataFrame({\"error\": [error_message]}))\n" }, "include_metrics": { "_input_type": "BoolInput", @@ -493,14 +493,15 @@ "verbose", "max_iterations", "agent_description", - "add_current_date_tool" + "add_current_date_tool", + "add_calculator_tool" ], "frozen": false, "icon": "bot", "last_updated": "2025-12-22T21:08:01.050Z", "legacy": false, "metadata": { - "code_hash": "154c71cf7441", + "code_hash": "239359d94298", "dependencies": { "dependencies": [ { @@ -541,6 +542,26 @@ "pinned": false, "template": { "_type": "Component", + "add_calculator_tool": { + "_input_type": "BoolInput", + "advanced": true, + "display_name": "Calculator", + "dynamic": false, + "info": "If true, adds a zero-config arithmetic calculator tool to the agent (safe: only +, -, *, /, ** operators via AST).", + "list": false, + "list_add_label": "Add More", + "name": "add_calculator_tool", + "override_skip": false, + "placeholder": "", + "required": false, + "show": true, + "title_case": false, + "tool_mode": false, + "trace_as_metadata": true, + "track_in_telemetry": true, + "type": "bool", + "value": true + }, "add_current_date_tool": { "_input_type": "BoolInput", "advanced": true, @@ -658,7 +679,7 @@ "show": true, "title_case": false, "type": "code", - "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=\"System Prompt: Initial instructions and context provided to guide the agent's behavior.\",\n value=\"You are a helpful assistant that can use tools to answer questions and perform tasks.\",\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self.system_prompt,\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt)\n agent_instructions = getattr(self, \"system_prompt\", \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" + "value": "from __future__ import annotations\n\nimport json\nimport re\nfrom datetime import datetime, timezone\nfrom typing import TYPE_CHECKING\n\nfrom pydantic import ValidationError\n\nfrom lfx.components.models_and_agents.memory import MemoryComponent\n\nif TYPE_CHECKING:\n from langchain_core.tools import Tool\n\nfrom lfx.base.agents.agent import LCToolsAgentComponent\nfrom lfx.base.agents.events import ExceptionWithMessageError\nfrom lfx.base.models.unified_models import (\n get_language_model_options,\n get_llm,\n handle_model_input_update,\n)\nfrom lfx.base.models.watsonx_constants import IBM_WATSONX_URLS\nfrom lfx.components.agentics.helpers.model_config import validate_model_selection\nfrom lfx.components.helpers import CalculatorComponent, CurrentDateComponent\nfrom lfx.components.langchain_utilities.tool_calling import ToolCallingAgentComponent\nfrom lfx.custom.custom_component.component import get_component_toolkit\nfrom lfx.field_typing.range_spec import RangeSpec\nfrom lfx.helpers.base_model import build_model_from_schema\nfrom lfx.inputs.inputs import BoolInput, DropdownInput, ModelInput, StrInput\nfrom lfx.io import IntInput, MessageTextInput, MultilineInput, Output, SecretStrInput, TableInput\nfrom lfx.log.logger import logger\nfrom lfx.schema.data import Data\nfrom lfx.schema.dotdict import dotdict\nfrom lfx.schema.message import Message\nfrom lfx.schema.table import EditMode\n\n\ndef set_advanced_true(component_input):\n component_input.advanced = True\n return component_input\n\n\nclass AgentComponent(ToolCallingAgentComponent):\n display_name: str = \"Agent\"\n description: str = \"Define the agent's instructions, then enter a task to complete using tools.\"\n documentation: str = \"https://docs.langflow.org/agents\"\n icon = \"bot\"\n beta = False\n name = \"Agent\"\n\n memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]\n\n inputs = [\n ModelInput(\n name=\"model\",\n display_name=\"Language Model\",\n info=\"Select your model provider\",\n real_time_refresh=True,\n required=True,\n ),\n SecretStrInput(\n name=\"api_key\",\n display_name=\"API Key\",\n info=\"Overrides global provider settings. Leave blank to use your pre-configured API Key.\",\n real_time_refresh=True,\n advanced=True,\n ),\n DropdownInput(\n name=\"base_url_ibm_watsonx\",\n display_name=\"watsonx API Endpoint\",\n info=\"The base URL of the API (IBM watsonx.ai only)\",\n options=IBM_WATSONX_URLS,\n value=IBM_WATSONX_URLS[0],\n combobox=True,\n show=False,\n real_time_refresh=True,\n ),\n StrInput(\n name=\"project_id\",\n display_name=\"watsonx Project ID\",\n info=\"The project ID associated with the foundation model (IBM watsonx.ai only)\",\n show=False,\n required=False,\n ),\n MultilineInput(\n name=\"system_prompt\",\n display_name=\"Agent Instructions\",\n info=(\n \"System Prompt: Initial instructions and context provided to guide the agent's behavior. \"\n \"Supports dynamic placeholders: {current_date}, {model_name}.\"\n ),\n value=(\n \"You are a helpful assistant that can use tools to answer questions and perform tasks. \"\n \"Today is {current_date}. You are powered by {model_name}.\"\n ),\n advanced=False,\n ),\n MessageTextInput(\n name=\"context_id\",\n display_name=\"Context ID\",\n info=\"The context ID of the chat. Adds an extra layer to the local memory.\",\n value=\"\",\n advanced=True,\n ),\n IntInput(\n name=\"n_messages\",\n display_name=\"Number of Chat History Messages\",\n value=100,\n info=\"Number of chat history messages to retrieve.\",\n advanced=True,\n show=True,\n ),\n IntInput(\n name=\"max_tokens\",\n display_name=\"Max Tokens\",\n info=\"Maximum number of tokens to generate. Field name varies by provider.\",\n advanced=True,\n range_spec=RangeSpec(min=1, max=128000, step=1, step_type=\"int\"),\n ),\n MultilineInput(\n name=\"format_instructions\",\n display_name=\"Output Format Instructions\",\n info=\"Generic Template for structured output formatting. Valid only with Structured response.\",\n value=(\n \"You are an AI that extracts structured JSON objects from unstructured text. \"\n \"Use a predefined schema with expected types (str, int, float, bool, dict). \"\n \"Extract ALL relevant instances that match the schema - if multiple patterns exist, capture them all. \"\n \"Fill missing or ambiguous values with defaults: null for missing values. \"\n \"Remove exact duplicates but keep variations that have different field values. \"\n \"Always return valid JSON in the expected format, never throw errors. \"\n \"If multiple objects can be extracted, return them all in the structured format.\"\n ),\n advanced=True,\n ),\n TableInput(\n name=\"output_schema\",\n display_name=\"Output Schema\",\n info=(\n \"Schema Validation: Define the structure and data types for structured output. \"\n \"No validation if no output schema.\"\n ),\n advanced=True,\n required=False,\n value=[],\n table_schema=[\n {\n \"name\": \"name\",\n \"display_name\": \"Name\",\n \"type\": \"str\",\n \"description\": \"Specify the name of the output field.\",\n \"default\": \"field\",\n \"edit_mode\": EditMode.INLINE,\n },\n {\n \"name\": \"description\",\n \"display_name\": \"Description\",\n \"type\": \"str\",\n \"description\": \"Describe the purpose of the output field.\",\n \"default\": \"description of field\",\n \"edit_mode\": EditMode.POPOVER,\n },\n {\n \"name\": \"type\",\n \"display_name\": \"Type\",\n \"type\": \"str\",\n \"edit_mode\": EditMode.INLINE,\n \"description\": (\"Indicate the data type of the output field (e.g., str, int, float, bool, dict).\"),\n \"options\": [\"str\", \"int\", \"float\", \"bool\", \"dict\"],\n \"default\": \"str\",\n },\n {\n \"name\": \"multiple\",\n \"display_name\": \"As List\",\n \"type\": \"boolean\",\n \"description\": \"Set to True if this output field should be a list of the specified type.\",\n \"default\": \"False\",\n \"edit_mode\": EditMode.INLINE,\n },\n ],\n ),\n *LCToolsAgentComponent.get_base_inputs(),\n # removed memory inputs from agent component\n # *memory_inputs,\n BoolInput(\n name=\"add_current_date_tool\",\n display_name=\"Current Date\",\n advanced=True,\n info=\"If true, will add a tool to the agent that returns the current date.\",\n value=True,\n ),\n BoolInput(\n name=\"add_calculator_tool\",\n display_name=\"Calculator\",\n advanced=True,\n info=(\n \"If true, adds a zero-config arithmetic calculator tool to the agent \"\n \"(safe: only +, -, *, /, ** operators via AST).\"\n ),\n value=True,\n ),\n ]\n outputs = [\n Output(name=\"response\", display_name=\"Response\", method=\"message_response\"),\n ]\n\n def _resolve_selected_model(self):\n \"\"\"Resolve the selected model, including legacy agent_llm/model_name inputs.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return self.model\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n return self.model\n\n legacy_provider = getattr(self, \"agent_llm\", None)\n legacy_model_name = getattr(self, \"model_name\", None)\n if not legacy_provider or not legacy_model_name:\n return self.model\n\n options = get_language_model_options(user_id=self.user_id)\n for option in options:\n if option.get(\"provider\") == legacy_provider and option.get(\"name\") == legacy_model_name:\n return [option]\n\n return [\n {\n \"name\": legacy_model_name,\n \"provider\": legacy_provider,\n \"metadata\": {},\n }\n ]\n\n def _get_max_tokens_value(self):\n \"\"\"Return the user-supplied max_tokens or None when unset/zero.\"\"\"\n val = getattr(self, \"max_tokens\", None)\n if val in {\"\", 0}:\n return None\n return val\n\n def _get_llm(self):\n \"\"\"Override parent to include max_tokens from the Agent's input field.\"\"\"\n return get_llm(\n model=self.model,\n user_id=self.user_id,\n api_key=getattr(self, \"api_key\", None),\n max_tokens=self._get_max_tokens_value(),\n watsonx_url=getattr(self, \"base_url_ibm_watsonx\", None),\n watsonx_project_id=getattr(self, \"project_id\", None),\n )\n\n async def get_agent_requirements(self):\n \"\"\"Get the agent requirements for the agent.\"\"\"\n from langchain_core.tools import StructuredTool\n\n selected_model = self._resolve_selected_model()\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n is_connected_model = isinstance(selected_model, BaseLanguageModel)\n except ImportError:\n is_connected_model = False\n\n if not is_connected_model:\n validate_model_selection(selected_model)\n\n # Ensure _get_llm() uses the resolved model (e.g. from legacy agent_llm/model_name)\n self.model = selected_model\n llm_model = self._get_llm()\n if llm_model is None:\n msg = \"No language model selected. Please choose a model to proceed.\"\n raise ValueError(msg)\n\n # Get memory data\n self.chat_history = await self.get_memory_data()\n await logger.adebug(f\"Retrieved {len(self.chat_history)} chat history messages\")\n if isinstance(self.chat_history, Message):\n self.chat_history = [self.chat_history]\n\n # Add current date tool if enabled\n if self.add_current_date_tool:\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n current_date_tool = (await CurrentDateComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(current_date_tool, StructuredTool):\n msg = \"CurrentDateComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(current_date_tool)\n\n # Add calculator tool if enabled (zero-config arithmetic)\n if getattr(self, \"add_calculator_tool\", False):\n if not isinstance(self.tools, list): # type: ignore[has-type]\n self.tools = []\n calculator_tool = (await CalculatorComponent(**self.get_base_args()).to_toolkit()).pop(0)\n\n if not isinstance(calculator_tool, StructuredTool):\n msg = \"CalculatorComponent must be converted to a StructuredTool\"\n raise TypeError(msg)\n self.tools.append(calculator_tool)\n\n # Set shared callbacks for tracing the tools used by the agent\n self.set_tools_callbacks(self.tools, self._get_shared_callbacks())\n\n return llm_model, self.chat_history, self.tools\n\n def _get_resolved_model_name(self) -> str:\n \"\"\"Best-effort human-readable model name for {model_name} injection.\"\"\"\n try:\n from langchain_core.language_models import BaseLanguageModel\n\n if isinstance(self.model, BaseLanguageModel):\n return type(self.model).__name__\n except ImportError:\n pass\n\n if isinstance(self.model, list) and self.model:\n first = self.model[0]\n if isinstance(first, dict):\n name = first.get(\"name\")\n if isinstance(name, str) and name:\n return name\n\n legacy_model_name = getattr(self, \"model_name\", None)\n if isinstance(legacy_model_name, str) and legacy_model_name:\n return legacy_model_name\n return \"\"\n\n def _inject_dynamic_prompt_values(self, prompt: str | None) -> str | None:\n \"\"\"Replace {current_date} / {model_name} placeholders in the system prompt.\n\n Uses str.replace (not str.format) so user prompts containing literal braces\n such as JSON examples ({\"key\": 1}) never break the agent.\n \"\"\"\n if not prompt:\n return prompt\n replacements = {\n \"{current_date}\": datetime.now(tz=timezone.utc).strftime(\"%Y-%m-%d %H:%M:%S UTC\"),\n \"{model_name}\": self._get_resolved_model_name(),\n }\n for placeholder, value in replacements.items():\n prompt = prompt.replace(placeholder, value)\n return prompt\n\n async def message_response(self) -> Message:\n try:\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n # Set up and run agent\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=self._inject_dynamic_prompt_values(self.system_prompt),\n )\n agent = self.create_agent_runnable()\n result = await self.run_agent(agent)\n\n # Store result for potential JSON output\n self._agent_result = result\n\n except (ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"{type(e).__name__}: {e!s}\")\n raise\n except ExceptionWithMessageError as e:\n await logger.aerror(f\"ExceptionWithMessageError occurred: {e}\")\n raise\n # Avoid catching blind Exception; let truly unexpected exceptions propagate\n except Exception as e:\n await logger.aerror(f\"Unexpected error: {e!s}\")\n raise\n else:\n return result\n\n def _preprocess_schema(self, schema):\n \"\"\"Preprocess schema to ensure correct data types for build_model_from_schema.\"\"\"\n processed_schema = []\n for field in schema:\n processed_field = {\n \"name\": str(field.get(\"name\", \"field\")),\n \"type\": str(field.get(\"type\", \"str\")),\n \"description\": str(field.get(\"description\", \"\")),\n \"multiple\": field.get(\"multiple\", False),\n }\n # Ensure multiple is handled correctly\n if isinstance(processed_field[\"multiple\"], str):\n processed_field[\"multiple\"] = processed_field[\"multiple\"].lower() in [\n \"true\",\n \"1\",\n \"t\",\n \"y\",\n \"yes\",\n ]\n processed_schema.append(processed_field)\n return processed_schema\n\n async def build_structured_output_base(self, content: str):\n \"\"\"Build structured output with optional BaseModel validation.\"\"\"\n json_pattern = r\"\\{.*\\}\"\n schema_error_msg = \"Try setting an output schema\"\n\n # Try to parse content as JSON first\n json_data = None\n try:\n json_data = json.loads(content)\n except json.JSONDecodeError:\n json_match = re.search(json_pattern, content, re.DOTALL)\n if json_match:\n try:\n json_data = json.loads(json_match.group())\n except json.JSONDecodeError:\n return {\"content\": content, \"error\": schema_error_msg}\n else:\n return {\"content\": content, \"error\": schema_error_msg}\n\n # If no output schema provided, return parsed JSON without validation\n if not hasattr(self, \"output_schema\") or not self.output_schema or len(self.output_schema) == 0:\n return json_data\n\n # Use BaseModel validation with schema\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n\n # Validate against the schema\n if isinstance(json_data, list):\n # Multiple objects\n validated_objects = []\n for item in json_data:\n try:\n validated_obj = output_model.model_validate(item)\n validated_objects.append(validated_obj.model_dump())\n except ValidationError as e:\n await logger.aerror(f\"Validation error for item: {e}\")\n # Include invalid items with error info\n validated_objects.append({\"data\": item, \"validation_error\": str(e)})\n return validated_objects\n\n # Single object\n try:\n validated_obj = output_model.model_validate(json_data)\n return [validated_obj.model_dump()] # Return as list for consistency\n except ValidationError as e:\n await logger.aerror(f\"Validation error: {e}\")\n return [{\"data\": json_data, \"validation_error\": str(e)}]\n\n except (TypeError, ValueError) as e:\n await logger.aerror(f\"Error building structured output: {e}\")\n # Fallback to parsed JSON without validation\n return json_data\n\n async def json_response(self) -> Data:\n \"\"\"Convert agent response to structured JSON Data output with schema validation.\"\"\"\n # Always use structured chat agent for JSON response mode for better JSON formatting\n try:\n system_components = []\n\n # 1. Agent Instructions (system_prompt).\n # Inject dynamic placeholders HERE so user-authored format_instructions\n # and schema descriptions appended later keep their literal {...} tokens.\n agent_instructions = self._inject_dynamic_prompt_values(getattr(self, \"system_prompt\", \"\") or \"\") or \"\"\n if agent_instructions:\n system_components.append(f\"{agent_instructions}\")\n\n # 2. Format Instructions\n format_instructions = getattr(self, \"format_instructions\", \"\") or \"\"\n if format_instructions:\n system_components.append(f\"Format instructions: {format_instructions}\")\n\n # 3. Schema Information from BaseModel\n if hasattr(self, \"output_schema\") and self.output_schema and len(self.output_schema) > 0:\n try:\n processed_schema = self._preprocess_schema(self.output_schema)\n output_model = build_model_from_schema(processed_schema)\n schema_dict = output_model.model_json_schema()\n schema_info = (\n \"You are given some text that may include format instructions, \"\n \"explanations, or other content alongside a JSON schema.\\n\\n\"\n \"Your task:\\n\"\n \"- Extract only the JSON schema.\\n\"\n \"- Return it as valid JSON.\\n\"\n \"- Do not include format instructions, explanations, or extra text.\\n\\n\"\n \"Input:\\n\"\n f\"{json.dumps(schema_dict, indent=2)}\\n\\n\"\n \"Output (only JSON schema):\"\n )\n system_components.append(schema_info)\n except (ValidationError, ValueError, TypeError, KeyError) as e:\n await logger.aerror(f\"Could not build schema for prompt: {e}\", exc_info=True)\n\n # Combine all components. Injection already applied on agent_instructions\n # above; do NOT re-run it here so literal {...} tokens in format\n # instructions / schema descriptions stay intact.\n combined_instructions = \"\\n\\n\".join(system_components) if system_components else \"\"\n llm_model, self.chat_history, self.tools = await self.get_agent_requirements()\n self.set(\n llm=llm_model,\n tools=self.tools or [],\n chat_history=self.chat_history,\n input_value=self.input_value,\n system_prompt=combined_instructions,\n )\n\n # Create and run structured chat agent\n try:\n structured_agent = self.create_agent_runnable()\n except (NotImplementedError, ValueError, TypeError) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n raise\n try:\n result = await self.run_agent(structured_agent)\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n RuntimeError,\n ) as e:\n await logger.aerror(f\"Error with structured agent result: {e}\")\n raise\n # Extract content from structured agent result\n if hasattr(result, \"content\"):\n content = result.content\n elif hasattr(result, \"text\"):\n content = result.text\n else:\n content = str(result)\n\n except (\n ExceptionWithMessageError,\n ValueError,\n TypeError,\n NotImplementedError,\n AttributeError,\n ) as e:\n await logger.aerror(f\"Error with structured chat agent: {e}\")\n # Fallback to regular agent\n content_str = \"No content returned from agent\"\n return Data(data={\"content\": content_str, \"error\": str(e)})\n\n # Process with structured output validation\n try:\n structured_output = await self.build_structured_output_base(content)\n\n # Handle different output formats\n if isinstance(structured_output, list) and structured_output:\n if len(structured_output) == 1:\n return Data(data=structured_output[0])\n return Data(data={\"results\": structured_output})\n if isinstance(structured_output, dict):\n return Data(data=structured_output)\n return Data(data={\"content\": content})\n\n except (ValueError, TypeError) as e:\n await logger.aerror(f\"Error in structured output processing: {e}\")\n return Data(data={\"content\": content, \"error\": str(e)})\n\n async def get_memory_data(self):\n # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.\n messages = (\n await MemoryComponent(**self.get_base_args())\n .set(\n session_id=self.graph.session_id,\n context_id=self.context_id,\n order=\"Ascending\",\n n_messages=self.n_messages,\n )\n .retrieve_messages()\n )\n return [\n message for message in messages if getattr(message, \"id\", None) != getattr(self.input_value, \"id\", None)\n ]\n\n def update_input_types(self, build_config: dotdict) -> dotdict:\n \"\"\"Update input types for all fields in build_config.\"\"\"\n for key, value in build_config.items():\n if isinstance(value, dict):\n if value.get(\"input_types\") is None:\n build_config[key][\"input_types\"] = []\n elif hasattr(value, \"input_types\") and value.input_types is None:\n value.input_types = []\n return build_config\n\n async def update_build_config(\n self,\n build_config: dotdict,\n field_value: list[dict],\n field_name: str | None = None,\n ) -> dotdict:\n # Update model options with caching (for all field changes)\n # Agents require tool calling, so filter for only tool-calling capable models\n build_config = handle_model_input_update(\n component=self,\n build_config=dict(build_config),\n field_value=field_value,\n field_name=field_name,\n cache_key_prefix=\"language_model_options_tool_calling\",\n get_options_func=lambda user_id=None: get_language_model_options(user_id=user_id, tool_calling=True),\n )\n build_config = dotdict(build_config)\n\n if field_name == \"model\":\n build_config = self.update_input_types(build_config)\n\n # Validate required keys\n default_keys = [\n \"code\",\n \"_type\",\n \"model\",\n \"tools\",\n \"input_value\",\n \"add_current_date_tool\",\n \"add_calculator_tool\",\n \"system_prompt\",\n \"agent_description\",\n \"max_iterations\",\n \"handle_parsing_errors\",\n \"verbose\",\n ]\n missing_keys = [key for key in default_keys if key not in build_config]\n if missing_keys:\n msg = f\"Missing required keys in build_config: {missing_keys}\"\n raise ValueError(msg)\n return dotdict({k: v.to_dict() if hasattr(v, \"to_dict\") else v for k, v in build_config.items()})\n\n async def _get_tools(self) -> list[Tool]:\n component_toolkit = get_component_toolkit()\n tools_names = self._build_tools_names()\n agent_description = self.get_tool_description()\n # TODO: Agent Description Depreciated Feature to be removed\n description = f\"{agent_description}{tools_names}\"\n\n tools = component_toolkit(component=self).get_tools(\n tool_name=\"Call_Agent\",\n tool_description=description,\n # here we do not use the shared callbacks as we are exposing the agent as a tool\n callbacks=self.get_langchain_callbacks(),\n )\n if hasattr(self, \"tools_metadata\"):\n tools = component_toolkit(component=self, metadata=self.tools_metadata).update_tools_metadata(tools=tools)\n\n return tools\n" }, "context_id": { "_input_type": "MessageTextInput", @@ -961,7 +982,7 @@ "copy_field": false, "display_name": "Agent Instructions", "dynamic": false, - "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior.", + "info": "System Prompt: Initial instructions and context provided to guide the agent's behavior. Supports dynamic placeholders: {current_date}, {model_name}.", "input_types": [ "Message" ], diff --git a/src/backend/base/langflow/services/database/models/__init__.py b/src/backend/base/langflow/services/database/models/__init__.py index 2ecc8d0153..b00e515af3 100644 --- a/src/backend/base/langflow/services/database/models/__init__.py +++ b/src/backend/base/langflow/services/database/models/__init__.py @@ -8,6 +8,7 @@ from .flow_version import FlowVersion from .flow_version_deployment_attachment import FlowVersionDeploymentAttachment from .folder import Folder from .jobs import Job +from .memory_base import MemoryBase, MemoryBaseSession, MemoryBaseWorkflowRun, MessageIngestionRecord from .message import MessageTable from .traces.model import SpanTable, TraceTable from .transactions import TransactionTable @@ -24,6 +25,10 @@ __all__ = [ "FlowVersionDeploymentAttachment", "Folder", "Job", + "MemoryBase", + "MemoryBaseSession", + "MemoryBaseWorkflowRun", + "MessageIngestionRecord", "MessageTable", "SSOConfig", "SSOUserProfile", diff --git a/src/backend/base/langflow/services/database/models/jobs/model.py b/src/backend/base/langflow/services/database/models/jobs/model.py index c18b4415f1..a58213d651 100644 --- a/src/backend/base/langflow/services/database/models/jobs/model.py +++ b/src/backend/base/langflow/services/database/models/jobs/model.py @@ -63,6 +63,9 @@ class JobBase(SQLModel): asset_type: str | None = Field( index=False, nullable=True ) # Polymorphic: records if job is related to an entity like a KB, workflow, etc. + dedupe_key: str | None = Field( + index=True, nullable=True + ) # Optional idempotency key to prevent duplicate jobs for the same asset and operation. class Job(JobBase, table=True): # type: ignore[call-arg] diff --git a/src/backend/base/langflow/services/database/models/memory_base/__init__.py b/src/backend/base/langflow/services/database/models/memory_base/__init__.py new file mode 100644 index 0000000000..842eb8826f --- /dev/null +++ b/src/backend/base/langflow/services/database/models/memory_base/__init__.py @@ -0,0 +1,21 @@ +from langflow.services.database.models.memory_base.model import ( + MemoryBase, + MemoryBaseCreate, + MemoryBaseRead, + MemoryBaseSession, + MemoryBaseSessionRead, + MemoryBaseUpdate, + MemoryBaseWorkflowRun, + MessageIngestionRecord, +) + +__all__ = [ + "MemoryBase", + "MemoryBaseCreate", + "MemoryBaseRead", + "MemoryBaseSession", + "MemoryBaseSessionRead", + "MemoryBaseUpdate", + "MemoryBaseWorkflowRun", + "MessageIngestionRecord", +] diff --git a/src/backend/base/langflow/services/database/models/memory_base/model.py b/src/backend/base/langflow/services/database/models/memory_base/model.py new file mode 100644 index 0000000000..26a3ce08b1 --- /dev/null +++ b/src/backend/base/langflow/services/database/models/memory_base/model.py @@ -0,0 +1,199 @@ +from datetime import datetime, timezone +from uuid import UUID, uuid4 + +import sqlalchemy as sa +from pydantic import model_validator +from sqlalchemy import Column, DateTime, ForeignKey, Index, UniqueConstraint +from sqlmodel import Field, Relationship, SQLModel + + +class MemoryBaseBase(SQLModel): + name: str = Field(index=False) + flow_id: UUID = Field(index=True) + user_id: UUID = Field(index=True) + threshold: int = Field(default=50) + auto_capture: bool = Field(default=True) + # Preprocessing config — accepted in payload but logic deferred to future scope + embedding_model: str = Field(default="") + preprocessing: bool = Field(default=False) + preproc_model: str | None = Field(default=None) + preproc_instructions: str | None = Field(default=None) + + +class MemoryBase(MemoryBaseBase, table=True): # type: ignore[call-arg] + __tablename__ = "memory_base" + __table_args__ = (UniqueConstraint("user_id", "name", name="uq_memory_base_user_name"),) + + id: UUID = Field(default_factory=uuid4, primary_key=True) + # kb_name is auto-generated at creation time — not user-supplied + kb_name: str = Field(default="") + created_at: datetime = Field( + default_factory=lambda: datetime.now(timezone.utc), + sa_column=Column(DateTime(timezone=True), nullable=False), + ) + + sessions: list["MemoryBaseSession"] = Relationship( + back_populates="memory_base", + sa_relationship_kwargs={"cascade": "all, delete-orphan"}, + ) + + +class MemoryBaseCreate(MemoryBaseBase): + user_id: UUID | None = None # Derived from auth token in the endpoint; not required in request body + + @model_validator(mode="after") + def preproc_model_required_when_preprocessing(self) -> "MemoryBaseCreate": + if self.preprocessing and not self.preproc_model: + msg = "preproc_model is required when preprocessing is enabled" + raise ValueError(msg) + return self + + +class MemoryBaseUpdate(SQLModel): + name: str | None = None + threshold: int | None = None + auto_capture: bool | None = None + preprocessing: bool | None = None + preproc_model: str | None = None + preproc_instructions: str | None = None + + +class MemoryBaseRead(MemoryBaseBase): + id: UUID + kb_name: str + created_at: datetime + + +class MemoryBaseSessionBase(SQLModel): + """Fields shared between the table class and response schemas.""" + + session_id: str = Field(index=True) + cursor_id: UUID | None = Field(default=None) + total_processed: int = Field(default=0) + last_sync_at: datetime | None = Field( + default=None, + sa_column=Column(DateTime(timezone=True), nullable=True), + ) + + +class MemoryBaseSession(MemoryBaseSessionBase, table=True): # type: ignore[call-arg] + __tablename__ = "memory_base_session" + + __table_args__ = ( + UniqueConstraint("memory_base_id", "session_id", name="uq_memory_base_session"), + Index("ix_memory_base_session_lookup", "memory_base_id", "session_id"), + ) + + id: UUID = Field(default_factory=uuid4, primary_key=True) + + # FK defined via sa_column so Alembic sees the same shape as the migration: + # inline ForeignKey on the column with ondelete="CASCADE". + # This matches the pattern used by the File model (ForeignKey on sa_column). + memory_base_id: UUID = Field( + sa_column=Column( + sa.Uuid(), + ForeignKey("memory_base.id", ondelete="CASCADE"), + nullable=False, + index=True, + ) + ) + + memory_base: MemoryBase = Relationship(back_populates="sessions") + + +class MemoryBaseSessionRead(MemoryBaseSessionBase): + id: UUID + memory_base_id: UUID # Explicit — not in base to keep base free of DB-layer FK + pending_count: int = Field(default=0) + + +class MemoryBaseWorkflowRun(SQLModel, table=True): # type: ignore[call-arg] + """Tracks WORKFLOW job runs per (memory_base, session) for threshold-based ingestion. + + One row per WORKFLOW job, per session, per memory base. + - ``workflow_job_id``: the WORKFLOW job that produced this run (SET NULL on job deletion). + - ``ingestion_job_id``: set only after the ingestion job that processed this run completes + successfully. NULL means the run is still pending (not yet counted toward an ingestion). + + Count pending = COUNT(*) WHERE ingestion_job_id IS NULL for a given (memory_base_id, session_id). + """ + + __tablename__ = "memory_base_workflow_run" + __table_args__ = ( + UniqueConstraint("memory_base_id", "session_id", "workflow_job_id", name="uq_mbwr_mb_session_wf_job"), + Index("ix_mbwr_mb_session", "memory_base_id", "session_id"), + Index("ix_mbwr_ingestion_job_id", "ingestion_job_id"), + ) + + id: UUID = Field(default_factory=uuid4, primary_key=True) + memory_base_id: UUID = Field( + sa_column=Column( + sa.Uuid(), + ForeignKey("memory_base.id", ondelete="CASCADE"), + nullable=False, + ) + ) + session_id: str = Field(sa_column=Column(sa.String(), nullable=False)) + workflow_job_id: UUID | None = Field( + default=None, + sa_column=Column( + sa.Uuid(), + ForeignKey("job.job_id", ondelete="SET NULL"), + nullable=True, + ), + ) + ingestion_job_id: UUID | None = Field( + default=None, + sa_column=Column( + sa.Uuid(), + ForeignKey("job.job_id", ondelete="SET NULL"), + nullable=True, + ), + ) + recorded_at: datetime = Field(sa_column=Column(DateTime(timezone=True), nullable=False)) + + +class MessageIngestionRecord(SQLModel, table=True): # type: ignore[call-arg] + """M-N join table recording which messages were ingested into which Memory Base by which job. + + One record per (message, session, memory_base) — enforced by the unique constraint. + Records are written only after a confirmed successful Chroma write (write-on-success). + On regenerate, all records for the memory_base are deleted atomically alongside the + cursor reset so that re-ingestion starts clean. + """ + + __tablename__ = "message_ingestion_record" + __table_args__ = ( + UniqueConstraint("message_id", "session_id", "memory_base_id", name="uq_mir_message_session_mb"), + Index("ix_mir_message_id", "message_id"), + Index("ix_mir_job_id", "job_id"), + Index("ix_mir_memory_base_session", "memory_base_id", "session_id"), + ) + + id: UUID = Field(default_factory=uuid4, primary_key=True) + + message_id: UUID = Field( + sa_column=Column( + sa.Uuid(), + ForeignKey("message.id", ondelete="CASCADE"), + nullable=False, + ) + ) + memory_base_id: UUID = Field( + sa_column=Column( + sa.Uuid(), + ForeignKey("memory_base.id", ondelete="CASCADE"), + nullable=False, + ) + ) + job_id: UUID | None = Field( + default=None, + sa_column=Column( + sa.Uuid(), + ForeignKey("job.job_id", ondelete="SET NULL"), + nullable=True, + ), + ) + # Denormalized from MessageTable.session_id — immutable, avoids JOIN on the hot query path + session_id: str = Field(sa_column=Column(sa.String(), nullable=False)) + ingested_at: datetime = Field(sa_column=Column(DateTime(timezone=True), nullable=False)) diff --git a/src/backend/base/langflow/services/database/models/message/model.py b/src/backend/base/langflow/services/database/models/message/model.py index 52bfd68267..a10291c27c 100644 --- a/src/backend/base/langflow/services/database/models/message/model.py +++ b/src/backend/base/langflow/services/database/models/message/model.py @@ -32,6 +32,7 @@ class MessageBase(SQLModel): properties: Properties = Field(default_factory=Properties) category: str = Field(default="message") content_blocks: list[ContentBlock] = Field(default_factory=list) + session_metadata: dict | None = Field(default=None) @field_serializer("timestamp") def serialize_timestamp(self, value): @@ -59,7 +60,7 @@ class MessageBase(SQLModel): return value @classmethod - def from_message(cls, message: "Message", flow_id: str | UUID | None = None): + def from_message(cls, message: "Message", flow_id: str | UUID | None = None, run_id: str | UUID | None = None): if message.text is None or not message.sender or not message.sender_name: msg = "The message does not have the required fields (text, sender, sender_name)." raise ValueError(msg) @@ -114,6 +115,13 @@ class MessageBase(SQLModel): msg = f"Flow ID {flow_id} is not a valid UUID" raise ValueError(msg) from exc + if isinstance(run_id, str): + try: + run_id = UUID(run_id) + except ValueError as exc: + msg = f"Run ID {run_id} is not a valid UUID" + raise ValueError(msg) from exc + return cls( sender=message.sender, sender_name=message.sender_name, @@ -123,6 +131,7 @@ class MessageBase(SQLModel): files=message.files or [], timestamp=timestamp, flow_id=flow_id, + run_id=run_id, properties=properties, category=message.category, content_blocks=content_blocks, @@ -149,6 +158,8 @@ class MessageTable(MessageBase, table=True): # type: ignore[call-arg] id: UUID = Field(default_factory=uuid4, primary_key=True) flow_id: UUID | None = Field(default=None) + run_id: UUID | None = Field(default=None, index=True) + is_output: bool = Field(default=False) files: list[str] = Field(sa_column=Column(JSON)) properties: dict | Properties = Field( # type: ignore[assignment] @@ -207,11 +218,9 @@ class MessageTable(MessageBase, table=True): # type: ignore[call-arg] @field_serializer("properties", "content_blocks", "session_metadata") @classmethod - def serialize_properties_or_content_blocks(cls, value) -> dict | list[dict] | None: + def serialize_properties_or_content_blocks(cls, value) -> dict | list[dict]: # Redundant sanitization here acts as a defensive measure for rows # already in the database that might contain NaN/Infinity values. - if value is None: - return None if isinstance(value, list): value = [cls.serialize_properties_or_content_blocks(item) for item in value] elif hasattr(value, "model_dump"): @@ -224,8 +233,9 @@ class MessageTable(MessageBase, table=True): # type: ignore[call-arg] class MessageRead(MessageBase): id: UUID - flow_id: UUID | None = Field() + flow_id: UUID | None = None session_metadata: dict | None = None + run_id: UUID | None = None class MessageCreate(MessageBase): @@ -243,3 +253,5 @@ class MessageUpdate(SQLModel): error: bool | None = None properties: Properties | None = None session_metadata: dict | None = None + category: str | None = None + content_blocks: list[ContentBlock] | None = None diff --git a/src/backend/base/langflow/services/deps.py b/src/backend/base/langflow/services/deps.py index 14f9bcbdc9..ae33611197 100644 --- a/src/backend/base/langflow/services/deps.py +++ b/src/backend/base/langflow/services/deps.py @@ -269,3 +269,10 @@ def get_flow_events_service(): from langflow.services.flow_events.factory import FlowEventsServiceFactory return get_service(ServiceType.FLOW_EVENTS_SERVICE, FlowEventsServiceFactory()) + + +def get_memory_base_service(): + """Retrieves the MemoryBaseService instance from the service manager.""" + from langflow.services.memory_base.factory import MemoryBaseServiceFactory + + return get_service(ServiceType.MEMORY_BASE_SERVICE, MemoryBaseServiceFactory()) diff --git a/src/backend/base/langflow/services/jobs/__init__.py b/src/backend/base/langflow/services/jobs/__init__.py index 8fd4b155b2..53e37c4096 100644 --- a/src/backend/base/langflow/services/jobs/__init__.py +++ b/src/backend/base/langflow/services/jobs/__init__.py @@ -1,5 +1,6 @@ """Job service package.""" +from langflow.services.jobs.exceptions import DuplicateJobError from langflow.services.jobs.service import JobService -__all__ = ["JobService"] +__all__ = ["DuplicateJobError", "JobService"] diff --git a/src/backend/base/langflow/services/jobs/exceptions.py b/src/backend/base/langflow/services/jobs/exceptions.py new file mode 100644 index 0000000000..385a596a1f --- /dev/null +++ b/src/backend/base/langflow/services/jobs/exceptions.py @@ -0,0 +1,16 @@ +"""Domain exceptions for the jobs service.""" + +from __future__ import annotations + + +class JobError(RuntimeError): + """Base exception for job-domain errors.""" + + +class DuplicateJobError(JobError): + """Raised by create_job() when a non-retryable job with the same dedupe_key already exists. + + (QUEUED, IN_PROGRESS, or COMPLETED). + FAILED and CANCELLED are retryable and do not trigger this error. + Extends RuntimeError so existing except RuntimeError callers keep working. + """ diff --git a/src/backend/base/langflow/services/jobs/service.py b/src/backend/base/langflow/services/jobs/service.py index 914e683e83..b258aec8c9 100644 --- a/src/backend/base/langflow/services/jobs/service.py +++ b/src/backend/base/langflow/services/jobs/service.py @@ -10,15 +10,16 @@ if TYPE_CHECKING: from datetime import datetime, timezone from uuid import UUID +from sqlmodel import col, func, select + from langflow.services.base import Service from langflow.services.database.models.jobs.crud import ( - get_job_by_job_id, - get_jobs_by_flow_id, get_latest_jobs_by_asset_ids, update_job_status, ) from langflow.services.database.models.jobs.model import Job, JobStatus, JobType from langflow.services.deps import session_scope +from langflow.services.jobs.exceptions import DuplicateJobError class JobService(Service): @@ -30,11 +31,14 @@ class JobService(Service): """Initialize the job service.""" self.set_ready() - async def get_jobs_by_flow_id(self, flow_id: UUID | str, page: int = 1, page_size: int = 10) -> list[Job]: - """Get jobs for a specific flow with pagination. + async def get_jobs_by_flow_id( + self, flow_id: UUID | str, user_id: UUID, page: int = 1, page_size: int = 10 + ) -> list[Job]: + """Get jobs for a specific flow with pagination, filtered by user. Args: flow_id: The flow ID to filter jobs by + user_id: The user ID to enforce ownership page: Page number (1-indexed) page_size: Number of jobs per page @@ -45,7 +49,16 @@ class JobService(Service): flow_id = UUID(flow_id) async with session_scope() as session: - return await get_jobs_by_flow_id(session, flow_id, page=page, size=page_size) + stmt = ( + select(Job) + .where(Job.flow_id == flow_id) + .where((Job.user_id == user_id) | (Job.user_id.is_(None))) + .order_by(col(Job.created_at).desc()) + .offset((page - 1) * page_size) + .limit(page_size) + ) + result = await session.exec(stmt) + return list(result.all()) async def get_job_by_job_id(self, job_id: UUID | str, user_id: UUID | None = None) -> Job | None: """Get job for a specific job ID. @@ -62,7 +75,11 @@ class JobService(Service): job_id = UUID(job_id) async with session_scope() as session: - return await get_job_by_job_id(session, job_id, user_id=user_id) + stmt = select(Job).where(Job.job_id == job_id) + if user_id: + stmt = stmt.where((Job.user_id == user_id) | (Job.user_id.is_(None))) + result = await session.exec(stmt) + return result.first() async def create_job( self, @@ -72,16 +89,19 @@ class JobService(Service): asset_id: UUID | None = None, asset_type: str | None = None, user_id: UUID | None = None, + dedupe_key: str | None = None, ) -> Job: """Create a new job record with QUEUED status. Args: job_id: The job ID flow_id: The flow ID + user_id: The user ID job_type: The job type asset_id: The asset ID asset_type: The asset type user_id: The user ID who owns this job + dedupe_key: Optional idempotency key to prevent duplicate jobs for the same batch Returns: Created Job object @@ -93,6 +113,18 @@ class JobService(Service): flow_id = UUID(flow_id) async with session_scope() as session: + if dedupe_key is not None: + stmt = ( + select(func.count()) + .select_from(Job) + .where(Job.dedupe_key == dedupe_key) + .where(col(Job.status).in_([JobStatus.QUEUED, JobStatus.IN_PROGRESS, JobStatus.COMPLETED])) + ) + result = await session.exec(stmt) + if result.one() > 0: + msg = f"A non-retryable job with dedupe_key={dedupe_key!r} already exists" + raise DuplicateJobError(msg) + job = Job( job_id=job_id, flow_id=flow_id, @@ -101,6 +133,7 @@ class JobService(Service): asset_id=asset_id, asset_type=asset_type, user_id=user_id, + dedupe_key=dedupe_key, ) session.add(job) await session.flush() @@ -210,3 +243,18 @@ class JobService(Service): await logger.ainfo(f"Job {job_id} completed successfully") await self.update_job_status(job_id, JobStatus.COMPLETED, finished_timestamp=True) return result + + async def _validate_ownership(self, job_id: UUID, user_id: UUID) -> Job: + """Verify that a job exists and belongs to the specified user. + + Raises: + ValueError: If the job is not found or is NOT owned by the user. + """ + job = await self.get_job_by_job_id(job_id) + if job is None: + msg = f"Job {job_id} not found" + raise ValueError(msg) + if job.user_id is not None and job.user_id != user_id: + msg = f"Access denied for job {job_id}" + raise ValueError(msg) + return job diff --git a/src/backend/base/langflow/services/memory_base/__init__.py b/src/backend/base/langflow/services/memory_base/__init__.py new file mode 100644 index 0000000000..c7b1f2c756 --- /dev/null +++ b/src/backend/base/langflow/services/memory_base/__init__.py @@ -0,0 +1,3 @@ +from langflow.services.memory_base.service import MemoryBaseService + +__all__ = ["MemoryBaseService"] diff --git a/src/backend/base/langflow/services/memory_base/document_builders.py b/src/backend/base/langflow/services/memory_base/document_builders.py new file mode 100644 index 0000000000..bef249e417 --- /dev/null +++ b/src/backend/base/langflow/services/memory_base/document_builders.py @@ -0,0 +1,139 @@ +"""Document building and KB metadata sync helpers for Memory Base ingestion. + +Extracted from task.py to separate "document shaping" from "ingestion orchestration". +""" + +from __future__ import annotations + +import json +from typing import TYPE_CHECKING + +from langchain_core.documents import Document +from langchain_text_splitters import RecursiveCharacterTextSplitter +from lfx.log.logger import logger + +from langflow.api.utils.kb_helpers import KBAnalysisHelper, KBStorageHelper + +if TYPE_CHECKING: + from pathlib import Path + + from langchain_chroma import Chroma + + from langflow.services.database.models.message.model import MessageTable + +# Chunk size for splitting long messages before embedding +MESSAGE_CHUNK_SIZE = 1000 +MESSAGE_CHUNK_OVERLAP = 100 + + +def extract_content_block_text(content_blocks: list) -> str: + """Extract embeddable text from content blocks of type text, code, and json. + + Blocks of any other type (tool_use, error, media, etc.) are skipped. + Each extracted piece is separated by a blank line so chunk boundaries + remain readable in the vector store. + """ + parts: list[str] = [] + for block in content_blocks: + # content_blocks are stored as JSON; each block is a dict at runtime. + contents: list = block.get("contents", []) if isinstance(block, dict) else [] + for entry in contents: + if not isinstance(entry, dict): + continue + entry_type = entry.get("type") + if entry_type == "text": + fragment = (entry.get("text") or "").strip() + elif entry_type == "code": + lang = entry.get("language") or "" + code = (entry.get("code") or "").strip() + fragment = f"```{lang}\n{code}\n```" if code else "" + elif entry_type == "json": + data = entry.get("data") + fragment = json.dumps(data, ensure_ascii=False) if data is not None else "" + else: + continue + if fragment: + parts.append(fragment) + return "\n\n".join(parts) + + +def build_documents_from_messages( + messages: list[MessageTable], + *, + session_id: str, + flow_id: str, + job_id: str = "", +) -> list[Document]: + """Convert MessageTable rows into LangChain Documents. + + Each message's embeddable text is the concatenation of msg.text and any + content-block fragments whose type is text, code, or json. Other block + types (tool_use, error, media, ...) are ignored. Long combined texts are + split by RecursiveCharacterTextSplitter before embedding. + """ + splitter = RecursiveCharacterTextSplitter( + chunk_size=MESSAGE_CHUNK_SIZE, + chunk_overlap=MESSAGE_CHUNK_OVERLAP, + ) + docs: list[Document] = [] + for msg in messages: + parts: list[str] = [] + if msg.text and msg.text.strip(): + parts.append(msg.text.strip()) + cb_text = extract_content_block_text(msg.content_blocks or []) + if cb_text: + parts.append(cb_text) + + text = "\n\n".join(parts) + if not text: + continue + chunks = splitter.split_text(text) + for i, chunk in enumerate(chunks): + docs.append( + Document( + page_content=chunk, + metadata={ + "message_id": str(msg.id), + "session_id": session_id, + "flow_id": flow_id, + "sender": msg.sender, + "sender_name": msg.sender_name, + "timestamp": msg.timestamp.isoformat() if msg.timestamp else "", + "run_id": str(msg.run_id) if msg.run_id else "", + "chunk_index": i, + "total_chunks": len(chunks), + "source": f"memory_base/{session_id}", + "job_id": job_id, + }, + ) + ) + return docs + + +def sync_kb_metadata(*, kb_path: Path, chroma: Chroma) -> None: + """Update embedding_metadata.json after a successful Memory Base ingestion. + + Mirrors the post-write metadata sync in ``KBIngestionHelper.perform_ingestion``: + - Refreshes chunk / word / character counts from the live Chroma collection. + - Updates on-disk size. + - Stamps ``is_memory_base: true`` (required for Knowledge Retrieval filtering). + - Sets ``source_types: ["memory"]`` to distinguish from file-based KBs. + + Called while the Chroma client is still open so that ``update_text_metrics`` + can query the collection directly without opening a second client. + """ + try: + metadata = KBAnalysisHelper.get_metadata(kb_path, fast=True) + KBAnalysisHelper.update_text_metrics(kb_path, metadata, chroma=chroma) + metadata["size"] = KBStorageHelper.get_directory_size(kb_path) + metadata["is_memory_base"] = True + # Preserve any existing source_types but always include "memory" + existing = set(metadata.get("source_types") or []) + existing.add("memory") + metadata["source_types"] = sorted(existing) + (kb_path / "embedding_metadata.json").write_text(json.dumps(metadata, indent=2)) + except (OSError, json.JSONDecodeError, ValueError): + # Metadata sync is best-effort; a failure here must not block the cursor advance. + # Note: this runs inside asyncio.to_thread so we use sync logging here. + # The lfx logger's sync .warning() method goes through the same structured pipeline. + logger.warning("KB metadata sync failed for kb_path=%s", kb_path, exc_info=True) diff --git a/src/backend/base/langflow/services/memory_base/embedding_helpers.py b/src/backend/base/langflow/services/memory_base/embedding_helpers.py new file mode 100644 index 0000000000..5e51394c83 --- /dev/null +++ b/src/backend/base/langflow/services/memory_base/embedding_helpers.py @@ -0,0 +1,26 @@ +"""Embedding provider inference for MemoryBase. + +Extracted from MemoryBaseService to keep single-responsibility per file. +""" + +from __future__ import annotations + +# Provider inference map — mirrors provider_patterns in KBAnalysisHelper._detect_embedding_provider +# so we can derive the provider from a model name string without filesystem access. +_MODEL_TO_PROVIDER: list[tuple[list[str], str]] = [ + (["text-embedding", "ada-", "gpt-"], "OpenAI"), + (["embed-english", "embed-multilingual"], "Cohere"), + (["sentence-transformers", "bert-", "huggingface"], "HuggingFace"), + (["palm", "gecko", "google"], "Google"), + (["ollama"], "Ollama"), + (["azure"], "Azure OpenAI"), +] + + +def infer_embedding_provider(embedding_model: str) -> str: + """Derive embedding provider name from a model string.""" + lower = embedding_model.lower() + for patterns, provider in _MODEL_TO_PROVIDER: + if any(p in lower for p in patterns): + return provider + return "OpenAI" # Safe default — matches _resolve_embedding fallback diff --git a/src/backend/base/langflow/services/memory_base/factory.py b/src/backend/base/langflow/services/memory_base/factory.py new file mode 100644 index 0000000000..d7492258dd --- /dev/null +++ b/src/backend/base/langflow/services/memory_base/factory.py @@ -0,0 +1,14 @@ +"""Factory for creating MemoryBaseService instances.""" + +from langflow.services.factory import ServiceFactory +from langflow.services.memory_base.service import MemoryBaseService + + +class MemoryBaseServiceFactory(ServiceFactory): + """Factory for creating MemoryBaseService instances.""" + + def __init__(self): + super().__init__(MemoryBaseService) + + def create(self): + return MemoryBaseService() diff --git a/src/backend/base/langflow/services/memory_base/ingestion.py b/src/backend/base/langflow/services/memory_base/ingestion.py new file mode 100644 index 0000000000..d6576c55c9 --- /dev/null +++ b/src/backend/base/langflow/services/memory_base/ingestion.py @@ -0,0 +1,421 @@ +"""Ingestion orchestration for MemoryBase — auto-capture, regeneration, and mismatch detection. + +Extracted from MemoryBaseService to keep single-responsibility per file. +The MemoryBaseService delegates to these functions for all ingestion-related work. +""" + +from __future__ import annotations + +import asyncio +import uuid +from typing import TYPE_CHECKING + +from lfx.log.logger import logger +from sqlmodel import col, func, select + +from langflow.api.utils.kb_helpers import KBAnalysisHelper, KBStorageHelper +from langflow.services.database.models.jobs.model import Job, JobStatus, JobType +from langflow.services.database.models.memory_base.model import ( + MemoryBase, + MemoryBaseSession, + MemoryBaseWorkflowRun, +) +from langflow.services.deps import get_job_service, get_task_service, session_scope +from langflow.services.jobs import DuplicateJobError +from langflow.services.memory_base.kb_path_helpers import ( + hash_session_id, + resolve_embedding, + resolve_kb_username, + resolve_kb_username_by_user_id, + validate_kb_path, +) +from langflow.services.memory_base.task import IngestionRequest, ingest_memory_task + +if TYPE_CHECKING: + from sqlmodel.ext.asyncio.session import AsyncSession + + +async def trigger_ingestion( + memory_base_id: uuid.UUID, + user_id: uuid.UUID, + session_id: str, + *, + get_mb_or_raise, + get_or_create_session, +) -> str: + """Manually trigger (or auto-trigger) an ingestion sync. + + Returns: + job_id string for the newly created job. + + Raises: + ValueError: If MemoryBase not found. + RuntimeError: If a job is already active (caller should return 409). + """ + async with session_scope() as db: + mb = await get_mb_or_raise(db, memory_base_id, user_id) + + # Ensure a session record exists + mbs = await get_or_create_session(db, memory_base_id, session_id) + + # Snapshot the cursor NOW (immutable arg for the task) + cursor_id_snapshot = mbs.cursor_id + + # Build dedupe_key from the latest uncovered WORKFLOW run for idempotency. + latest_job_id = await _get_latest_pending_workflow_job_id(db, mb, mbs) + dedupe_key: str | None = None + if latest_job_id is not None: + dedupe_key = f"ingestion:{memory_base_id}:{session_id}:{latest_job_id}" + + kb_username = await resolve_kb_username(db, mb.user_id) + embedding_provider, embedding_model = resolve_embedding(mb.kb_name, kb_username) + + # Create tracking job + job_service = get_job_service() + job_id = uuid.uuid4() + await job_service.create_job( + job_id=job_id, + flow_id=mb.flow_id, + user_id=mb.user_id, + job_type=JobType.INGESTION, + asset_id=memory_base_id, + asset_type="memory_base", + dedupe_key=dedupe_key, + ) + + task_service = get_task_service() + await task_service.fire_and_forget_task( + job_service.execute_with_status, + job_id=job_id, + run_coro_func=ingest_memory_task, + request=IngestionRequest( + memory_base_id=memory_base_id, + session_id=session_id, + flow_id=mb.flow_id, + kb_name=mb.kb_name, + kb_username=kb_username, + user_id=mb.user_id, + embedding_provider=embedding_provider, + embedding_model=embedding_model, + cursor_id=cursor_id_snapshot, + task_job_id=job_id, + job_service=job_service, + ), + ) + + return str(job_id) + + +async def on_flow_output( + flow_id: uuid.UUID, + session_id: str, + job_id: uuid.UUID | None, + *, + get_or_create_session, +) -> None: + """Called after a flow run completes. + + For every MemoryBase watching this flow with auto_capture=True: + 1. Record the workflow run in the tracking table (inside _maybe_trigger). + 2. Count uncovered WORKFLOW runs for this session. + 3. If count >= threshold, fire ingestion task. + """ + async with session_scope() as db: + stmt = ( + select(MemoryBase).where(MemoryBase.flow_id == flow_id).where(MemoryBase.auto_capture == True) # noqa: E712 + ) + result = await db.exec(stmt) + memory_bases = list(result.all()) + + hashed_sid = hash_session_id(session_id) + for mb in memory_bases: + try: + await logger.adebug( + "Auto-capture check | memory_base=%s threshold=%s session=%s", + mb.id, + mb.threshold, + hashed_sid, + ) + await _maybe_trigger( + mb=mb, session_id=session_id, job_id=job_id, get_or_create_session=get_or_create_session + ) + except (RuntimeError, ValueError, OSError): + await logger.aerror("Auto-capture failed for memory_base=%s session=%s", mb.id, hashed_sid, exc_info=True) + + +async def _maybe_trigger( + *, + mb: MemoryBase, + session_id: str, + job_id: uuid.UUID | None, + get_or_create_session, +) -> None: + async with session_scope() as db: + mbs = await get_or_create_session(db, mb.id, session_id) + + # Record this workflow run before evaluating the threshold. + await _insert_workflow_run(db, mb.id, session_id, job_id) + + pending = await count_pending_messages(db, mb, mbs) + + if pending < mb.threshold: + return + + cursor_id_snapshot = mbs.cursor_id + + # Build dedupe_key from the latest pending WORKFLOW run for idempotency. + latest_wf_job_id = await _get_latest_pending_workflow_job_id(db, mb, mbs) + dedupe_key: str | None = None + if latest_wf_job_id is not None: + dedupe_key = f"ingestion:{mb.id}:{session_id}:{latest_wf_job_id}" + + kb_username = await resolve_kb_username(db, mb.user_id) + + embedding_provider, embedding_model = resolve_embedding(mb.kb_name, kb_username) + + job_service = get_job_service() + job_id = uuid.uuid4() + try: + await job_service.create_job( + job_id=job_id, + flow_id=mb.flow_id, + user_id=mb.user_id, + job_type=JobType.INGESTION, + asset_id=mb.id, + asset_type="memory_base", + dedupe_key=dedupe_key, + ) + except DuplicateJobError: + await logger.adebug("Auto-capture: duplicate job for dedupe_key=%s - skipping.", dedupe_key) + return + + task_service = get_task_service() + await task_service.fire_and_forget_task( + job_service.execute_with_status, + job_id=job_id, + run_coro_func=ingest_memory_task, + request=IngestionRequest( + memory_base_id=mb.id, + session_id=session_id, + flow_id=mb.flow_id, + kb_name=mb.kb_name, + kb_username=kb_username, + user_id=mb.user_id, + embedding_provider=embedding_provider, + embedding_model=embedding_model, + cursor_id=cursor_id_snapshot, + task_job_id=job_id, + job_service=job_service, + ), + ) + + +async def check_mismatch( + memory_base_id: uuid.UUID, + user_id: uuid.UUID, + *, + get_mb_or_raise, +) -> bool: + """Return True if metadata claims processed rows but vector store is empty.""" + async with session_scope() as db: + mb = await get_mb_or_raise(db, memory_base_id, user_id) + stmt = select(func.sum(MemoryBaseSession.total_processed)).where( + MemoryBaseSession.memory_base_id == memory_base_id + ) + result = await db.exec(stmt) + total_processed: int = result.first() or 0 + + if total_processed == 0: + return False + + kb_username = await resolve_kb_username_by_user_id(user_id) + kb_root = KBStorageHelper.get_root_path() + if not kb_root: + return False + kb_path = kb_root / kb_username / mb.kb_name + validate_kb_path(kb_root, kb_path) + if not await asyncio.to_thread(kb_path.exists): + return True + + metadata = KBAnalysisHelper.get_metadata(kb_path, fast=True) + return int(metadata.get("chunks", 0)) == 0 + + +async def regenerate( + memory_base_id: uuid.UUID, + user_id: uuid.UUID, + *, + get_mb_or_raise, + trigger_ingestion_fn, +) -> list[str]: + """Reset all session cursors to None and re-trigger ingestion per session. + + Used to recover from FS / Vector DB mismatch (Chroma dir deleted externally). + Returns list of newly created job IDs. + Also deletes all MessageIngestionRecord rows for this memory base atomically + with the cursor reset so that re-ingestion starts clean without hitting the + unique constraint. + """ + from sqlalchemy import delete as sa_delete + + from langflow.services.database.models.memory_base.model import MessageIngestionRecord + + async with session_scope() as db: + await get_mb_or_raise(db, memory_base_id, user_id) + + stmt = select(MemoryBaseSession).where(MemoryBaseSession.memory_base_id == memory_base_id) + result = await db.exec(stmt) + sessions = list(result.all()) + + for s in sessions: + s.cursor_id = None + db.add(s) + + # Delete existing ingestion records so re-ingestion inserts fresh rows + await db.exec( # type: ignore[call-overload] + sa_delete(MessageIngestionRecord).where(MessageIngestionRecord.memory_base_id == memory_base_id) + ) + await db.commit() + + job_ids: list[str] = [] + for s in sessions: + try: + jid = await trigger_ingestion_fn(memory_base_id, user_id, s.session_id) + job_ids.append(jid) + except DuplicateJobError: + await logger.awarning( + "Regenerate: duplicate batch already ingested for session %s - skipped.", + hash_session_id(s.session_id), + ) + except RuntimeError: + await logger.awarning( + "Regenerate: active job exists for session %s - reset cursor but skipped trigger.", + hash_session_id(s.session_id), + ) + return job_ids + + +async def cancel_active_jobs(*, memory_base_id: uuid.UUID, db: AsyncSession) -> None: + """Cancel all IN_PROGRESS or QUEUED jobs for this memory base.""" + stmt = ( + select(Job) + .where(Job.asset_id == memory_base_id) + .where(Job.asset_type == "memory_base") + .where(col(Job.status).in_([JobStatus.IN_PROGRESS, JobStatus.QUEUED])) + ) + result = await db.exec(stmt) + active_jobs = list(result.all()) + + task_service = get_task_service() + job_service = get_job_service() + for job in active_jobs: + try: + await task_service.revoke_task(job.job_id) + await job_service.update_job_status(job.job_id, JobStatus.CANCELLED) + await logger.ainfo("Cancelled job %s for memory_base %s", job.job_id, memory_base_id) + except (RuntimeError, ValueError, OSError): + await logger.awarning( + "Could not cancel job %s for memory_base %s", job.job_id, memory_base_id, exc_info=True + ) + + +# ------------------------------------------------------------------ # +# Shared query helpers (public — used by service.py and memories.py) # +# ------------------------------------------------------------------ # + + +async def count_pending_messages(db: AsyncSession, mb: MemoryBase, mbs: MemoryBaseSession) -> int: + """Count WORKFLOW runs for this (memory_base, session) not yet covered by a completed ingestion. + + A row in memory_base_workflow_run with ingestion_job_id IS NULL means the run + has not been processed by any ingestion job. Count pending = number of such rows. + This is session-scoped and time-independent; job failures leave rows NULL so they + are correctly re-counted on the next threshold check. + """ + stmt = ( + select(func.count()) + .select_from(MemoryBaseWorkflowRun) + .where(MemoryBaseWorkflowRun.memory_base_id == mb.id) + .where(MemoryBaseWorkflowRun.session_id == mbs.session_id) + .where(MemoryBaseWorkflowRun.ingestion_job_id == None) # noqa: E711 + ) + try: + result = await db.exec(stmt) + row = result.first() + if row is None: + return 0 + return int(row) + except (TypeError, ValueError, OSError) as e: + await logger.aerror("Error counting pending workflow runs: %s", e) + return 0 + + +async def _insert_workflow_run( + db: AsyncSession, + memory_base_id: uuid.UUID, + session_id: str, + job_id: uuid.UUID | None, +) -> None: + """Record a WORKFLOW job run for (memory_base_id, session_id). + + Verifies that job_id refers to a WORKFLOW type job before inserting. + Uses dialect-specific INSERT ... ON CONFLICT DO NOTHING for idempotency — + safe to call multiple times with the same arguments. + Skips silently if job_id is None or the job is not of WORKFLOW type. + """ + from datetime import datetime, timezone + + hashed_sid = hash_session_id(session_id) + if job_id is None: + await logger.awarning( + "on_flow_output called with no job_id for memory_base=%s session=%s — run not recorded.", + memory_base_id, + hashed_sid, + ) + return + + job_result = await db.exec(select(Job).where(Job.job_id == job_id).where(Job.type == JobType.WORKFLOW)) + if job_result.first() is None: + await logger.awarning( + "job_id=%s is not a WORKFLOW job — skipping workflow run record for memory_base=%s session=%s.", + job_id, + memory_base_id, + hashed_sid, + ) + return + + row = { + "id": uuid.uuid4(), + "memory_base_id": memory_base_id, + "session_id": session_id, + "workflow_job_id": job_id, + "ingestion_job_id": None, + "recorded_at": datetime.now(timezone.utc), + } + conn = await db.connection() + if conn.dialect.name == "postgresql": + from sqlalchemy.dialects.postgresql import insert as pg_insert + + stmt = pg_insert(MemoryBaseWorkflowRun).values([row]).on_conflict_do_nothing() + else: + from sqlalchemy.dialects.sqlite import insert as sqlite_insert + + stmt = sqlite_insert(MemoryBaseWorkflowRun).values([row]).on_conflict_do_nothing() + await db.exec(stmt) # type: ignore[call-overload] + await db.commit() + + +async def _get_latest_pending_workflow_job_id( + db: AsyncSession, mb: MemoryBase, mbs: MemoryBaseSession +) -> uuid.UUID | None: + """Return the workflow_job_id of the most recent uncovered workflow run for this session.""" + stmt = ( + select(MemoryBaseWorkflowRun.workflow_job_id) + .where(MemoryBaseWorkflowRun.memory_base_id == mb.id) + .where(MemoryBaseWorkflowRun.session_id == mbs.session_id) + .where(MemoryBaseWorkflowRun.ingestion_job_id == None) # noqa: E711 + .order_by(col(MemoryBaseWorkflowRun.recorded_at).desc()) + .limit(1) + ) + result = await db.exec(stmt) + return result.first() diff --git a/src/backend/base/langflow/services/memory_base/kb_path_helpers.py b/src/backend/base/langflow/services/memory_base/kb_path_helpers.py new file mode 100644 index 0000000000..3887a22425 --- /dev/null +++ b/src/backend/base/langflow/services/memory_base/kb_path_helpers.py @@ -0,0 +1,149 @@ +"""KB path resolution and username helpers for MemoryBase. + +Extracted from MemoryBaseService to keep single-responsibility per file. +""" + +from __future__ import annotations + +import asyncio +import hashlib +import json +import re +import uuid +from datetime import datetime, timezone +from typing import TYPE_CHECKING + +from lfx.log.logger import logger +from sqlmodel import select + +from langflow.api.utils.kb_helpers import KBAnalysisHelper, KBStorageHelper +from langflow.services.deps import session_scope + +if TYPE_CHECKING: + from pathlib import Path + + from sqlmodel.ext.asyncio.session import AsyncSession + + +def validate_kb_path(kb_root: Path, kb_path: Path) -> None: + """Assert that kb_path is contained within kb_root (path traversal guard). + + Prevents crafted usernames with '..' segments from escaping the KB root directory. + Follows the same pattern as services/storage/local.py:save_file. + """ + kb_root_resolved = kb_root.resolve() + kb_path_resolved = kb_path.resolve() + if not kb_path_resolved.is_relative_to(kb_root_resolved): + msg = "KB path escapes root directory" + raise ValueError(msg) + + +def hash_session_id(session_id: str) -> str: + """Return a truncated SHA-256 hash for safe logging of session IDs.""" + return hashlib.sha256(session_id.encode()).hexdigest()[:12] + + +def sanitize_kb_name(name: str) -> str: + """Lowercase, replace spaces/hyphens with underscores, strip non-alphanum.""" + sanitized = name.strip().lower() + sanitized = re.sub(r"[\s\-]+", "_", sanitized) + sanitized = re.sub(r"[^\w]", "", sanitized) + return sanitized or "memory" + + +async def resolve_kb_username(db: AsyncSession, user_id: uuid.UUID) -> str: + """Look up the username for a user_id within an existing DB session.""" + from langflow.services.database.models.user.model import User + + stmt = select(User.username).where(User.id == user_id) + result = await db.exec(stmt) + username = result.first() + if not username: + msg = f"User {user_id} not found" + raise ValueError(msg) + return username + + +async def resolve_kb_username_by_user_id(user_id: uuid.UUID) -> str: + """Look up the username for a user_id using a fresh DB session.""" + async with session_scope() as db: + return await resolve_kb_username(db, user_id) + + +def resolve_embedding(kb_name: str, kb_username: str) -> tuple[str, str]: + """Read embedding provider/model from KB metadata.json, with sane defaults.""" + kb_root = KBStorageHelper.get_root_path() + if not kb_root: + return "OpenAI", "text-embedding-3-small" + kb_path: Path = kb_root / kb_username / kb_name + metadata = KBAnalysisHelper.get_metadata(kb_path, fast=True) + provider = metadata.get("embedding_provider") or "OpenAI" + model = metadata.get("embedding_model") or "text-embedding-3-small" + return provider, model + + +async def initialize_kb( + *, + kb_name: str, + kb_username: str, + embedding_provider: str, + embedding_model: str, +) -> None: + """Create KB directory, initialize Chroma, and write embedding_metadata.json. + + Mirrors the logic in knowledge_bases.py:create_knowledge_base so Memory Base + KBs are immediately visible with the correct metadata (including is_memory_base: true). + """ + import chromadb + + kb_root = KBStorageHelper.get_root_path() + if not kb_root: + await logger.awarning("KB root path not configured — Memory Base KB will not be initialized on disk.") + return + + kb_path: Path = kb_root / kb_username / kb_name + validate_kb_path(kb_root, kb_path) + await asyncio.to_thread(kb_path.mkdir, parents=True, exist_ok=True) + + # Initialize Chroma collection so the directory is non-empty and readable + try: + client = KBStorageHelper.get_fresh_chroma_client(kb_path) + client.create_collection(name=kb_name) + except (OSError, ValueError, chromadb.errors.ChromaError) as exc: + await logger.awarning("Initial Chroma setup for %s failed: %s", kb_name, exc) + finally: + client = None # type: ignore[assignment] + KBStorageHelper.release_chroma_resources(kb_path) + + embedding_metadata = { + "id": str(uuid.uuid4()), + "embedding_provider": embedding_provider, + "embedding_model": embedding_model, + "is_memory_base": True, + "created_at": datetime.now(timezone.utc).isoformat(), + "chunks": 0, + "words": 0, + "characters": 0, + "avg_chunk_size": 0.0, + "size": 0, + "source_types": ["memory"], + } + await asyncio.to_thread( + (kb_path / "embedding_metadata.json").write_text, + json.dumps(embedding_metadata, indent=2), + ) + + +async def delete_kb(*, kb_name: str, kb_username: str) -> None: + """Remove the KB directory from disk. Logs on failure, does not raise.""" + if not kb_name: + return + kb_root = KBStorageHelper.get_root_path() + if not kb_root: + return + kb_path = kb_root / kb_username / kb_name + validate_kb_path(kb_root, kb_path) + try: + await asyncio.to_thread(KBStorageHelper.delete_storage, kb_path, kb_name) + except (OSError, ValueError): + await logger.awarning("Could not delete KB '%s' from disk after Memory Base deletion.", kb_name, exc_info=True) diff --git a/src/backend/base/langflow/services/memory_base/service.py b/src/backend/base/langflow/services/memory_base/service.py new file mode 100644 index 0000000000..b2c280df0e --- /dev/null +++ b/src/backend/base/langflow/services/memory_base/service.py @@ -0,0 +1,288 @@ +"""MemoryBase service — CRUD and session state management. + +Ingestion orchestration, KB path helpers, and embedding inference are in +separate modules (ingestion.py, kb_path_helpers.py, embedding_helpers.py) +to keep this file focused on data access and business-rule enforcement. + +Edge cases handled: +- Name uniqueness per user: 409 if a Memory Base with the same name already exists. +- Deletion during sync: cancels active tasks before DB deletion. +- KB deletion on delete: removes the associated KB directory from disk. +- Concurrent task prevention: returns 409 if a job is already IN_PROGRESS. +- Threshold updates: deferred; does not re-evaluate pending count immediately. +""" + +from __future__ import annotations + +import uuid +from typing import TYPE_CHECKING + +from sqlmodel import col, select + +from langflow.services.base import Service +from langflow.services.database.models.memory_base.model import ( + MemoryBase, + MemoryBaseCreate, + MemoryBaseSession, + MemoryBaseUpdate, +) +from langflow.services.deps import session_scope +from langflow.services.memory_base.embedding_helpers import infer_embedding_provider +from langflow.services.memory_base.ingestion import ( + cancel_active_jobs, +) +from langflow.services.memory_base.ingestion import ( + check_mismatch as _check_mismatch, +) +from langflow.services.memory_base.ingestion import ( + on_flow_output as _on_flow_output, +) +from langflow.services.memory_base.ingestion import ( + regenerate as _regenerate, +) +from langflow.services.memory_base.ingestion import ( + trigger_ingestion as _trigger_ingestion, +) +from langflow.services.memory_base.kb_path_helpers import ( + delete_kb, + initialize_kb, + resolve_kb_username, + sanitize_kb_name, +) + +if TYPE_CHECKING: + from sqlmodel.ext.asyncio.session import AsyncSession + + +class MemoryBaseService(Service): + """Service layer for MemoryBase CRUD and session state management.""" + + name = "memory_base_service" + + # ------------------------------------------------------------------ # + # CRUD # + # ------------------------------------------------------------------ # + + async def create(self, payload: MemoryBaseCreate, user_id: uuid.UUID) -> MemoryBase: + # 1. Verify that the referenced flow belongs to this user. + async with session_scope() as db: + from langflow.services.database.models.flow.model import Flow + + flow_result = await db.exec(select(Flow).where(Flow.id == payload.flow_id).where(Flow.user_id == user_id)) + if flow_result.first() is None: + msg = f"Flow {payload.flow_id} not found" + raise PermissionError(msg) + + # 2. Resolve username — needed for the KB path. + async with session_scope() as db: + kb_username = await resolve_kb_username(db, user_id) + + # 3. Auto-generate kb_name: sanitized_name_<8hex> + kb_name = f"{sanitize_kb_name(payload.name)}_{uuid.uuid4().hex[:8]}" + + # 4. Create KB directory and embedding_metadata.json on disk. + embedding_provider = infer_embedding_provider(payload.embedding_model) + await initialize_kb( + kb_name=kb_name, + kb_username=kb_username, + embedding_provider=embedding_provider, + embedding_model=payload.embedding_model, + ) + + # 5. Uniqueness check + insert. + from sqlalchemy.exc import IntegrityError + + async with session_scope() as db: + existing = await db.exec( + select(MemoryBase).where(MemoryBase.user_id == user_id).where(MemoryBase.name == payload.name) + ) + if existing.first() is not None: + msg = f"A Memory Base named '{payload.name}' already exists for this user" + raise ValueError(msg) + + mb = MemoryBase( + **payload.model_dump(exclude={"user_id"}), + user_id=user_id, + kb_name=kb_name, + ) + db.add(mb) + try: + await db.commit() + except IntegrityError: + msg = f"A Memory Base named '{payload.name}' already exists for this user" + raise ValueError(msg) from None + await db.refresh(mb) + + return mb + + async def list_for_user(self, user_id: uuid.UUID) -> list[MemoryBase]: + async with session_scope() as db: + stmt = select(MemoryBase).where(MemoryBase.user_id == user_id) + result = await db.exec(stmt) + return list(result.all()) + + def list_for_user_stmt(self, user_id: uuid.UUID, flow_id: uuid.UUID | None = None): # type: ignore[return] + """Return the SQLModel select statement for pagination at the API layer.""" + stmt = select(MemoryBase).where(MemoryBase.user_id == user_id) + if flow_id is not None: + stmt = stmt.where(MemoryBase.flow_id == flow_id) + return stmt + + async def get(self, memory_base_id: uuid.UUID, user_id: uuid.UUID) -> MemoryBase | None: + async with session_scope() as db: + stmt = select(MemoryBase).where(MemoryBase.id == memory_base_id).where(MemoryBase.user_id == user_id) + result = await db.exec(stmt) + return result.first() + + async def update( + self, + memory_base_id: uuid.UUID, + user_id: uuid.UUID, + patch: MemoryBaseUpdate, + ) -> MemoryBase | None: + """Update mutable fields. + + Threshold changes take effect on the NEXT auto-capture trigger; any + already-running ingestion task ignores the change (immutable args). + """ + async with session_scope() as db: + stmt = select(MemoryBase).where(MemoryBase.id == memory_base_id).where(MemoryBase.user_id == user_id) + result = await db.exec(stmt) + mb = result.first() + if mb is None: + return None + for field, value in patch.model_dump(exclude_unset=True).items(): + setattr(mb, field, value) + db.add(mb) + await db.commit() + await db.refresh(mb) + return mb + + async def delete(self, memory_base_id: uuid.UUID, user_id: uuid.UUID) -> bool: + """Delete a MemoryBase and its associated KB directory.""" + async with session_scope() as db: + stmt = select(MemoryBase).where(MemoryBase.id == memory_base_id).where(MemoryBase.user_id == user_id) + result = await db.exec(stmt) + mb = result.first() + if mb is None: + return False + + kb_name = mb.kb_name + kb_username = await resolve_kb_username(db, user_id) + + # Cancel active ingestion jobs before removing the DB record + await cancel_active_jobs(memory_base_id=memory_base_id, db=db) + + await db.delete(mb) + await db.commit() + + # Delete the corresponding KB from disk (best-effort — DB already committed) + await delete_kb(kb_name=kb_name, kb_username=kb_username) + + return True + + # ------------------------------------------------------------------ # + # Sessions # + # ------------------------------------------------------------------ # + + async def verify_ownership(self, memory_base_id: uuid.UUID, user_id: uuid.UUID) -> None: + """Raise ValueError if the Memory Base does not belong to user_id.""" + async with session_scope() as db: + await self.get_memory_base_or_404(db, memory_base_id, user_id) + + def sessions_stmt(self, memory_base_id: uuid.UUID, user_id: uuid.UUID): # type: ignore[return] + """Return the select statement for persisted sessions, for use with apaginate. + + Inline-joins MemoryBase to verify ownership in the SQL itself, so a + caller that forgets a pre-check cannot leak other users' sessions. + """ + return ( + select(MemoryBaseSession) + .join(MemoryBase, MemoryBase.id == MemoryBaseSession.memory_base_id) + .where(MemoryBaseSession.memory_base_id == memory_base_id) + .where(MemoryBase.user_id == user_id) + .order_by(col(MemoryBaseSession.last_sync_at).desc()) + ) + + # ------------------------------------------------------------------ # + # Ingestion delegation # + # ------------------------------------------------------------------ # + + async def trigger_ingestion( + self, + memory_base_id: uuid.UUID, + user_id: uuid.UUID, + session_id: str, + ) -> str: + return await _trigger_ingestion( + memory_base_id, + user_id, + session_id, + get_mb_or_raise=self.get_memory_base_or_404, + get_or_create_session=self._get_or_create_session, + ) + + async def on_flow_output( + self, + flow_id: uuid.UUID, + session_id: str, + job_id: uuid.UUID | None, + ) -> None: + await _on_flow_output( + flow_id, + session_id, + job_id, + get_or_create_session=self._get_or_create_session, + ) + + async def check_mismatch(self, memory_base_id: uuid.UUID, user_id: uuid.UUID) -> bool: + return await _check_mismatch( + memory_base_id, + user_id, + get_mb_or_raise=self.get_memory_base_or_404, + ) + + async def regenerate(self, memory_base_id: uuid.UUID, user_id: uuid.UUID) -> list[str]: + return await _regenerate( + memory_base_id, + user_id, + get_mb_or_raise=self.get_memory_base_or_404, + trigger_ingestion_fn=self.trigger_ingestion, + ) + + # ------------------------------------------------------------------ # + # Public query helpers # + # ------------------------------------------------------------------ # + + async def get_memory_base_or_404( + self, db: AsyncSession, memory_base_id: uuid.UUID, user_id: uuid.UUID + ) -> MemoryBase: + """Fetch a MemoryBase or raise ValueError (mapped to 404 at the API layer).""" + stmt = select(MemoryBase).where(MemoryBase.id == memory_base_id).where(MemoryBase.user_id == user_id) + result = await db.exec(stmt) + mb = result.first() + if mb is None: + msg = f"MemoryBase {memory_base_id} not found" + raise ValueError(msg) + return mb + + # ------------------------------------------------------------------ # + # Internal helpers # + # ------------------------------------------------------------------ # + + async def _get_or_create_session( + self, db: AsyncSession, memory_base_id: uuid.UUID, session_id: str + ) -> MemoryBaseSession: + stmt = ( + select(MemoryBaseSession) + .where(MemoryBaseSession.memory_base_id == memory_base_id) + .where(MemoryBaseSession.session_id == session_id) + ) + result = await db.exec(stmt) + mbs = result.first() + if mbs is None: + mbs = MemoryBaseSession(memory_base_id=memory_base_id, session_id=session_id) + db.add(mbs) + await db.commit() + await db.refresh(mbs) + return mbs diff --git a/src/backend/base/langflow/services/memory_base/task.py b/src/backend/base/langflow/services/memory_base/task.py new file mode 100644 index 0000000000..a41d764951 --- /dev/null +++ b/src/backend/base/langflow/services/memory_base/task.py @@ -0,0 +1,444 @@ +"""Background task for Memory Base ingestion. + +Design principles enforced here: +- Cursor atomicity: cursor_id is NEVER updated before ingestion confirms success. +- Retry safety: If a job fails, cursor_id remains at the last known good position. +- Serialization: A per-(memory_base_id, session_id) distributed lock prevents concurrent + jobs from racing to write the same messages into Chroma. Uses PostgreSQL advisory locks + for cross-worker safety, with an in-process asyncio.Lock fallback for SQLite (dev/test). + The lock is acquired before any DB or Chroma access and released in a finally block. +- Live cursor: After acquiring the lock, the current cursor_id is re-read from the DB + (not the dispatch-time snapshot) so the pending message fetch always starts from the + true latest position, even if a prior job advanced the cursor while this job waited. +- Path safety: kb_path is validated against kb_root before any filesystem operation. + +The actual Chroma write logic is shared with KB file ingestion via +``KBIngestionHelper.write_documents_to_chroma`` — no duplicate batching/retry code here. + +Document building and KB metadata sync live in ``document_builders.py``. +""" + +from __future__ import annotations + +import asyncio +import hashlib +import types +import weakref +from dataclasses import dataclass +from datetime import datetime, timezone +from typing import TYPE_CHECKING + +from langchain_chroma import Chroma +from lfx.log.logger import logger +from sqlalchemy import text +from sqlmodel import Session, col, select + +from langflow.api.utils.kb_helpers import KBIngestionHelper, KBStorageHelper +from langflow.services.database.models.memory_base.model import MemoryBaseSession, MemoryBaseWorkflowRun +from langflow.services.database.models.message.model import MessageTable +from langflow.services.deps import get_settings_service, session_scope +from langflow.services.memory_base.document_builders import build_documents_from_messages, sync_kb_metadata +from langflow.services.memory_base.kb_path_helpers import hash_session_id, validate_kb_path + +if TYPE_CHECKING: + import uuid + from pathlib import Path + + from langflow.services.jobs.service import JobService + + +@dataclass(frozen=True, slots=True) +class IngestionRequest: + """Typed parameter bundle for ``ingest_memory_task``. + + All fields needed to run an ingestion job are grouped here so callers + construct one object instead of threading 11+ loose kwargs. + """ + + memory_base_id: uuid.UUID + session_id: str + flow_id: uuid.UUID + kb_name: str + kb_username: str + user_id: uuid.UUID + embedding_provider: str + embedding_model: str + cursor_id: uuid.UUID | None + task_job_id: uuid.UUID + job_service: JobService + + +# The ingestion lock timeout is read from settings (max_ingestion_timeout_secs). +# If the timeout expires before the lock is acquired, an asyncio.TimeoutError is raised. + +# --------------------------------------------------------------------------- +# Distributed locking: PostgreSQL advisory locks with in-process fallback +# --------------------------------------------------------------------------- +# In multi-worker deployments, an asyncio.Lock is process-local and cannot +# serialize across workers. We use PostgreSQL session-level advisory locks +# keyed on a hash of (memory_base_id, session_id). For SQLite (dev/test) we +# fall back to the in-process asyncio.Lock which is sufficient for a single worker. + +_session_ingestion_locks: weakref.WeakValueDictionary[tuple, asyncio.Lock] = weakref.WeakValueDictionary() + + +def _get_or_create_session_lock(key: tuple) -> asyncio.Lock: + """Return the asyncio.Lock for the given key (SQLite fallback only).""" + lock = _session_ingestion_locks.get(key) + if lock is None: + lock = asyncio.Lock() + _session_ingestion_locks[key] = lock + return lock + + +def _compute_advisory_key(memory_base_id: uuid.UUID, session_id: str) -> int: + """Compute a stable int64 advisory lock key from (memory_base_id, session_id).""" + raw = f"{memory_base_id}:{session_id}".encode() + return int(hashlib.sha256(raw).hexdigest()[:16], 16) % (2**63 - 1) + + +async def _is_postgres() -> bool: + """Return True if the database backend is PostgreSQL.""" + from langflow.services.deps import get_db_service + + db_service = get_db_service() + return db_service.engine.dialect.name == "postgresql" + + +async def _pg_advisory_lock(db: Session, key: int) -> None: + """Acquire a PostgreSQL session-level advisory lock with retry and timeout. + + The lock is held on the specific connection of the shared 'db' session. + """ + timeout = get_settings_service().settings.max_ingestion_timeout_secs + deadline = asyncio.get_event_loop().time() + timeout + backoff = 0.1 + max_backoff = 5.0 + + while True: + conn = await db.connection() + result = await conn.execute(text(f"SELECT pg_try_advisory_lock({key})")) + acquired = result.scalar() + + if acquired: + return + + remaining = deadline - asyncio.get_event_loop().time() + if remaining <= 0: + raise asyncio.TimeoutError + + await asyncio.sleep(min(backoff, remaining)) + backoff = min(backoff * 2, max_backoff) + + +async def _pg_advisory_unlock(db: Session, key: int) -> None: + """Release a PostgreSQL session-level advisory lock on the shared session.""" + conn = await db.connection() + await conn.execute(text(f"SELECT pg_advisory_unlock({key})")) + + +async def _acquire_session_lock(db: Session, memory_base_id: uuid.UUID, session_id: str) -> int | asyncio.Lock: + """Acquire the distributed ingestion lock. Returns the key (PG) or Lock (SQLite).""" + timeout = get_settings_service().settings.max_ingestion_timeout_secs + if await _is_postgres(): + key = _compute_advisory_key(memory_base_id, session_id) + await _pg_advisory_lock(db, key) + return key + lock = _get_or_create_session_lock((memory_base_id, session_id)) + await asyncio.wait_for(lock.acquire(), timeout=timeout) + return lock + + +async def _release_session_lock(db: Session, lock_handle: int | asyncio.Lock) -> None: + """Release the distributed ingestion lock.""" + if isinstance(lock_handle, int): + await _pg_advisory_unlock(db, lock_handle) + else: + lock_handle.release() + + +async def _read_live_cursor(db: Session, memory_base_id: uuid.UUID, session_id: str) -> uuid.UUID | None: + """Read current cursor_id from shared 'db' session inside the serialization lock.""" + stmt = ( + select(MemoryBaseSession.cursor_id) + .where(MemoryBaseSession.memory_base_id == memory_base_id) + .where(MemoryBaseSession.session_id == session_id) + ) + result = await db.exec(stmt) + return result.first() + + +async def ingest_memory_task(*, request: IngestionRequest) -> dict: + """Ingest pending output messages from a session into the target Knowledge Base. + + Accepts a single ``IngestionRequest`` dataclass that bundles all required parameters. + + Serialization: acquires a per-(memory_base_id, session_id) distributed lock before + any DB or Chroma access. Uses PostgreSQL advisory locks for cross-worker + serialization (multi-worker safe) with an in-process asyncio.Lock fallback for + SQLite. Concurrent jobs for the same session wait up to max_ingestion_timeout_secs; + if the lock cannot be acquired in time, asyncio.TimeoutError is re-raised so + execute_with_status records JobStatus.TIMED_OUT. + + Live cursor: after acquiring the lock, the current cursor_id is re-read from the DB. + ``cursor_id`` on the request is the dispatch-time snapshot kept only for logging. + """ + # Unpack for readability within the function body + memory_base_id = request.memory_base_id + session_id = request.session_id + flow_id = request.flow_id + kb_name = request.kb_name + kb_username = request.kb_username + user_id = request.user_id + embedding_provider = request.embedding_provider + embedding_model = request.embedding_model + cursor_id = request.cursor_id + task_job_id = request.task_job_id + job_service = request.job_service + + hashed_sid = hash_session_id(session_id) + await logger.adebug( + "Ingestion job started | memory_base=%s session=%s dispatch_cursor=%s job=%s", + memory_base_id, + hashed_sid, + cursor_id, + task_job_id, + ) + kb_root = KBStorageHelper.get_root_path() + if not kb_root: + msg = "Knowledge base root path is not configured" + raise RuntimeError(msg) + + kb_path: Path = kb_root / kb_username / kb_name + + # ---- Path traversal guard ---- + validate_kb_path(kb_root, kb_path) + + # ---- 0. Acquire per-session serialization lock ---- + async with session_scope() as db: + try: + lock_handle = await _acquire_session_lock(db, memory_base_id, session_id) + except asyncio.TimeoutError: + await logger.awarning( + "Ingestion lock wait timeout | memory_base=%s session=%s job=%s.", + memory_base_id, + hashed_sid, + task_job_id, + ) + raise + + try: + # ---- 0b. Re-read live cursor inside the lock ---- + live_cursor_id = await _read_live_cursor(db, memory_base_id, session_id) + await logger.adebug( + "Ingestion lock acquired | memory_base=%s session=%s live_cursor=%s job=%s", + memory_base_id, + hashed_sid, + live_cursor_id, + task_job_id, + ) + + # ---- 1. Fetch pending output messages for this session ---- + messages = await _fetch_pending_messages( + db, + flow_id=flow_id, + session_id=session_id, + cursor_id=live_cursor_id, + ) + if not messages: + await logger.ainfo( + "MemoryBase %s / session %s: no pending messages, skipping.", memory_base_id, hashed_sid + ) + return {"message": "No pending messages", "ingested": 0} + + # ---- 2. Build documents from messages ---- + job_id_str = str(task_job_id) + documents = build_documents_from_messages( + messages, session_id=session_id, flow_id=str(flow_id), job_id=job_id_str + ) + + if not documents: + return {"message": "No non-empty messages to ingest", "ingested": 0} + + # ---- 3. Check cancellation before touching the vector store ---- + if await KBIngestionHelper.is_job_cancelled(job_service, task_job_id): + return {"message": "Job cancelled before ingestion", "ingested": 0} + + # ---- 4. Open Chroma, write, then sync KB metadata ---- + user_stub = types.SimpleNamespace(id=user_id) + embeddings = await KBIngestionHelper.build_embeddings(embedding_provider, embedding_model, user_stub) + + client = KBStorageHelper.get_fresh_chroma_client(kb_path) + written = 0 + try: + chroma = Chroma(client=client, embedding_function=embeddings, collection_name=kb_name) + + written = await KBIngestionHelper.write_documents_to_chroma( + documents=documents, + chroma=chroma, + task_job_id=task_job_id, + job_service=job_service, + ) + + if written == len(documents): + await asyncio.to_thread(sync_kb_metadata, kb_path=kb_path, chroma=chroma) + except Exception: + await logger.aerror( + "Ingestion write failed | memory_base=%s session=%s job=%s. Rolling back partial writes...", + memory_base_id, + hashed_sid, + task_job_id, + ) + await KBIngestionHelper.cleanup_chroma_chunks_by_job(task_job_id, kb_path, kb_name) + raise + finally: + KBStorageHelper.release_chroma_resources(kb_path) + + if written < len(documents): + await logger.awarning("Ingestion job %s was cancelled. Cleaning up partial data...", task_job_id) + await KBIngestionHelper.cleanup_chroma_chunks_by_job(task_job_id, kb_path, kb_name) + return {"message": "Job cancelled during ingestion", "ingested": 0} + + # ---- 5. Bulk-stamp ingestion metadata ---- + await _mark_messages_ingested(db, messages=messages, job_id=task_job_id, memory_base_id=memory_base_id) + + # ---- 6. Update cursor atomically ONLY after confirmed success ---- + last_message_id = messages[-1].id + ingested_count = len(messages) + await _advance_cursor( + db, + memory_base_id=memory_base_id, + session_id=session_id, + new_cursor_id=last_message_id, + ingested_count=ingested_count, + task_job_id=task_job_id, + ) + + await logger.ainfo( + "Ingestion job finished | memory_base=%s session=%s job=%s ingested=%d", + memory_base_id, + hashed_sid, + task_job_id, + ingested_count, + ) + return {"message": "Success", "ingested": ingested_count} + + finally: + await _release_session_lock(db, lock_handle) + + +async def _fetch_pending_messages( + db: Session, + *, + flow_id: uuid.UUID, + session_id: str, + cursor_id: uuid.UUID | None, +) -> list[MessageTable]: + """Fetch all messages for this session that come after cursor_id using shared session.""" + from sqlalchemy import and_, or_ + + stmt = ( + select(MessageTable) + .where(MessageTable.flow_id == flow_id) + .where(MessageTable.session_id == session_id) + .order_by(col(MessageTable.timestamp).asc(), col(MessageTable.id).asc()) + ) + if cursor_id is not None: + cursor_stmt = select(MessageTable.timestamp, MessageTable.id).where(MessageTable.id == cursor_id) + result = await db.exec(cursor_stmt) + cursor_row = result.first() + if cursor_row: + cursor_ts, c_id = cursor_row + stmt = stmt.where( + or_( + col(MessageTable.timestamp) > cursor_ts, + and_( + col(MessageTable.timestamp) == cursor_ts, + col(MessageTable.id) > c_id, + ), + ) + ) + + result = await db.exec(stmt) + return list(result.all()) + + +async def _mark_messages_ingested( + db: Session, + *, + messages: list[MessageTable], + job_id: uuid.UUID, + memory_base_id: uuid.UUID, +) -> None: + """Batch-insert ingestion records for all successfully ingested messages using shared session.""" + from uuid import uuid4 as _uuid4 + + from langflow.services.database.models.memory_base.model import MessageIngestionRecord + + ingested_at = datetime.now(timezone.utc) + rows = [ + { + "id": _uuid4(), + "message_id": msg.id, + "memory_base_id": memory_base_id, + "job_id": job_id, + "session_id": msg.session_id, + "ingested_at": ingested_at, + } + for msg in messages + ] + conn = await db.connection() + if conn.dialect.name == "postgresql": + from sqlalchemy.dialects.postgresql import insert as pg_insert + + stmt = pg_insert(MessageIngestionRecord).values(rows).on_conflict_do_nothing() + else: + from sqlalchemy.dialects.sqlite import insert as sqlite_insert + + stmt = sqlite_insert(MessageIngestionRecord).values(rows).on_conflict_do_nothing() + await db.exec(stmt) # type: ignore[call-overload] + await db.commit() + + +async def _advance_cursor( + db: Session, + *, + memory_base_id: uuid.UUID, + session_id: str, + new_cursor_id: uuid.UUID, + ingested_count: int, + task_job_id: uuid.UUID, +) -> None: + """Atomically advance the cursor using the shared 'db' session.""" + from sqlalchemy import update as sa_update + + stmt = ( + select(MemoryBaseSession) + .where(MemoryBaseSession.memory_base_id == memory_base_id) + .where(MemoryBaseSession.session_id == session_id) + ) + result = await db.exec(stmt) + mbs = result.first() + if mbs is None: + await logger.awarning( + "MemoryBaseSession for (%s, %s) vanished before cursor update.", + memory_base_id, + hash_session_id(session_id), + ) + return + + mbs.cursor_id = new_cursor_id + mbs.total_processed += ingested_count + mbs.last_sync_at = datetime.now(timezone.utc) + db.add(mbs) + + # Stamp all pending workflow run rows for this session. + await db.exec( # type: ignore[call-overload] + sa_update(MemoryBaseWorkflowRun) + .where(MemoryBaseWorkflowRun.memory_base_id == memory_base_id) + .where(MemoryBaseWorkflowRun.session_id == session_id) + .where(MemoryBaseWorkflowRun.ingestion_job_id == None) # noqa: E711 + .values(ingestion_job_id=task_job_id) + ) + + await db.commit() diff --git a/src/backend/base/langflow/services/schema.py b/src/backend/base/langflow/services/schema.py index 44b075b08c..6dbe4b8abf 100644 --- a/src/backend/base/langflow/services/schema.py +++ b/src/backend/base/langflow/services/schema.py @@ -22,3 +22,4 @@ class ServiceType(str, Enum): MCP_COMPOSER_SERVICE = "mcp_composer_service" JOB_SERVICE = "jobs_service" FLOW_EVENTS_SERVICE = "flow_events_service" + MEMORY_BASE_SERVICE = "memory_base_service" diff --git a/src/backend/base/langflow/services/utils.py b/src/backend/base/langflow/services/utils.py index f40bb89572..4a901496f2 100644 --- a/src/backend/base/langflow/services/utils.py +++ b/src/backend/base/langflow/services/utils.py @@ -94,6 +94,12 @@ async def setup_superuser(settings_service: SettingsService, session: AsyncSessi ) if user is not None: await logger.adebug("Superuser created successfully.") + # When the default superuser is recreated (e.g. after a DB reset in + # AUTO_LOGIN mode) the per-user MCP servers config file saved under the + # previous UUID becomes orphaned on disk. Best-effort recover it so + # users don't lose their MCP server configuration across restarts. + if is_default and settings_service.auth_settings.AUTO_LOGIN: + await migrate_orphaned_mcp_servers_config(session, settings_service, user) except Exception as exc: logger.exception(exc) msg = "Could not create superuser. Please create a superuser manually." @@ -103,6 +109,149 @@ async def setup_superuser(settings_service: SettingsService, session: AsyncSessi settings_service.auth_settings.reset_credentials() +async def migrate_orphaned_mcp_servers_config( + session: AsyncSession, + settings_service: SettingsService, + current_user, +) -> bool: + """Best-effort recovery of MCP servers config files orphaned by a DB reset. + + The MCP servers config is persisted on disk at + ``{config_dir}/{user_id}/_mcp_servers_{user_id}.json`` and tracked in the DB + via a ``File`` row. When Langflow starts with a fresh database but the same + config directory (common in containerized deployments without a persisted + DB volume), the default superuser is recreated with a new UUID and the + previously saved MCP config files become unreachable. + + Recovery rules (intentionally conservative to avoid importing another user's + config — MCP server entries can contain ``env`` and ``headers`` auth material): + + * If the new user already has an MCP config file on disk without a matching + ``File`` row, re-register the row (self-heal a partial previous migration). + * If exactly one orphaned ``_mcp_servers_{uuid}.json`` is found in the config + directory, migrate it. + * If multiple orphans are found, skip and log — we can't safely identify the + previous default superuser's file without extra metadata, so leave manual + recovery to the operator. + + Returns True when a file was migrated or a missing DB row was restored. + """ + from pathlib import Path + from uuid import UUID + + import aiofiles + import anyio + + from langflow.services.database.models.file.model import File as UserFile + + try: + config_dir_value = settings_service.settings.config_dir + if not config_dir_value: + return False + + config_dir = Path(config_dir_value) + if not config_dir.exists() or not config_dir.is_dir(): + return False + + # The current user's DB record is fresh; nothing to migrate if they + # somehow already have an MCP config row (defensive guard). + name_without_ext = f"_mcp_servers_{current_user.id}" + existing_stmt = ( + select(UserFile).where(UserFile.user_id == current_user.id).where(UserFile.name == name_without_ext) + ) + if (await session.exec(existing_stmt)).first() is not None: + return False + + current_user_dir = str(current_user.id) + new_dir = config_dir / current_user_dir + new_filename = f"_mcp_servers_{current_user.id}.json" + new_file_path = new_dir / new_filename + db_path = f"{current_user.id}/{new_filename}" + + # Case 1: a previous migration attempt copied the file but failed before + # committing the DB row. Re-register the existing file instead of + # returning early and leaving the user with an invisible config. + if new_file_path.exists(): + try: + size = new_file_path.stat().st_size + except OSError as exc: + await logger.awarning( + "Cannot stat existing MCP config %s while self-healing DB row: %s", + new_file_path, + exc, + ) + return False + session.add(UserFile(user_id=current_user.id, name=name_without_ext, path=db_path, size=size)) + await session.commit() + await logger.ainfo( + "Restored missing MCP servers config DB row for user %s from existing file %s", + current_user.id, + new_file_path, + ) + return True + + def _find_orphans() -> list[tuple[float, Path]]: + orphans: list[tuple[float, Path]] = [] + for entry in config_dir.iterdir(): + if not entry.is_dir() or entry.name == current_user_dir: + continue + try: + UUID(entry.name) + except ValueError: + continue + mcp_path = entry / f"_mcp_servers_{entry.name}.json" + if mcp_path.is_file(): + try: + mtime = mcp_path.stat().st_mtime + except OSError: + continue + orphans.append((mtime, mcp_path)) + return orphans + + orphans = await anyio.to_thread.run_sync(_find_orphans) + if not orphans: + return False + + # Ambiguous: more than one candidate could belong to different users. + # Refuse to migrate rather than risk importing unrelated auth material. + if len(orphans) > 1: + orphan_paths = ", ".join(str(p) for _, p in sorted(orphans, key=lambda i: i[0], reverse=True)) + await logger.awarning( + "Found %d orphaned MCP servers config files in %s; skipping automatic " + "migration to avoid restoring the wrong one. Move the intended file to " + "%s to recover. Candidates: %s", + len(orphans), + config_dir, + new_file_path, + orphan_paths, + ) + return False + + _, orphan_path = orphans[0] + + async with aiofiles.open(str(orphan_path), "rb") as src: + data = await src.read() + + await anyio.to_thread.run_sync(lambda: new_dir.mkdir(parents=True, exist_ok=True)) + async with aiofiles.open(str(new_file_path), "wb") as dst: + await dst.write(data) + + session.add(UserFile(user_id=current_user.id, name=name_without_ext, path=db_path, size=len(data))) + await session.commit() + + await logger.ainfo( + "Migrated orphaned MCP servers config from %s to user %s", + orphan_path, + current_user.id, + ) + except Exception as exc: # noqa: BLE001 + # Never let migration failure block startup. + await logger.awarning("Failed to migrate orphaned MCP servers config: %s", exc) + return False + else: + return True + + async def teardown_superuser(settings_service, session: AsyncSession) -> None: """Teardown the superuser.""" # If AUTO_LOGIN is True, we will remove the default superuser diff --git a/src/backend/tests/unit/components/models_and_agents/test_agent_component.py b/src/backend/tests/unit/components/models_and_agents/test_agent_component.py index 3cb4648ea4..b55c0ac5e9 100644 --- a/src/backend/tests/unit/components/models_and_agents/test_agent_component.py +++ b/src/backend/tests/unit/components/models_and_agents/test_agent_component.py @@ -39,6 +39,7 @@ class TestAgentComponent(ComponentTestBaseWithoutClient): return { "_type": "Agent", "add_current_date_tool": True, + "add_calculator_tool": True, "agent_description": "A helpful agent", "model": MockLanguageModel(), "handle_parsing_errors": True, @@ -450,6 +451,199 @@ class TestAgentComponent(ComponentTestBaseWithoutClient): # Note: The provider-specific field name mapping happens inside get_llm, # so we just verify max_tokens is passed correctly + async def test_should_append_calculator_tool_when_add_calculator_toggle_is_true( + self, component_class, default_kwargs + ): + """Calculator tool is appended when the toggle is enabled. + + Given add_calculator_tool=True, When get_agent_requirements runs, + Then self.tools contains a StructuredTool derived from CalculatorComponent. + """ + from unittest.mock import AsyncMock + + from langchain_core.tools import StructuredTool + + default_kwargs["add_calculator_tool"] = True + default_kwargs["add_current_date_tool"] = False # isolate: only calculator + component = await self.component_setup(component_class, default_kwargs) + component.model = [{"name": "gpt-4o", "provider": "OpenAI", "metadata": {}}] + component.get_memory_data = AsyncMock(return_value=[]) + component._get_shared_callbacks = list + component.set_tools_callbacks = lambda *_: None + + with patch("lfx.components.models_and_agents.agent.get_llm") as mock_get_llm: + mock_get_llm.return_value = MockLanguageModel() + _, _, tools = await component.get_agent_requirements() + + assert len(tools) == 1 + assert isinstance(tools[0], StructuredTool) + assert "evaluate" in tools[0].name.lower(), f"Expected a Calculator-derived tool; got name={tools[0].name!r}" + + async def test_should_not_append_calculator_tool_when_add_calculator_toggle_is_false( + self, component_class, default_kwargs + ): + """Calculator tool is skipped when the toggle is disabled. + + Given add_calculator_tool=False, When get_agent_requirements runs, + Then no Calculator tool is appended to self.tools. + """ + from unittest.mock import AsyncMock + + default_kwargs["add_calculator_tool"] = False + default_kwargs["add_current_date_tool"] = False + component = await self.component_setup(component_class, default_kwargs) + component.model = [{"name": "gpt-4o", "provider": "OpenAI", "metadata": {}}] + component.get_memory_data = AsyncMock(return_value=[]) + component._get_shared_callbacks = list + component.set_tools_callbacks = lambda *_: None + + with patch("lfx.components.models_and_agents.agent.get_llm") as mock_get_llm: + mock_get_llm.return_value = MockLanguageModel() + _, _, tools = await component.get_agent_requirements() + + assert tools == [] + + def test_should_replace_current_date_and_model_name_when_both_placeholders_present(self, component_class): + """Unit test: helper replaces both placeholders with concrete values.""" + component = component_class() + component.model = [{"name": "gpt-4o", "provider": "OpenAI", "metadata": {}}] + + prompt = "Today is {current_date}. You are powered by {model_name}." + result = component._inject_dynamic_prompt_values(prompt) + + assert "{current_date}" not in result + assert "{model_name}" not in result + assert "gpt-4o" in result + + def test_should_leave_literal_braces_untouched_when_prompt_has_no_known_placeholders(self, component_class): + """Adversarial: prompts with literal JSON like {"key": 1} must not raise and must stay intact.""" + component = component_class() + component.model = [{"name": "gpt-4o", "provider": "OpenAI", "metadata": {}}] + + prompt = 'Respond with JSON: {"key": 1, "nested": {"a": [1, 2]}}.' + result = component._inject_dynamic_prompt_values(prompt) + + assert result == prompt + + def test_should_return_empty_when_prompt_is_empty(self, component_class): + """Edge case: empty/None prompt is returned as-is without raising.""" + component = component_class() + assert component._inject_dynamic_prompt_values("") == "" + assert component._inject_dynamic_prompt_values(None) is None + + async def test_should_inject_dynamic_values_into_system_prompt_when_message_response_runs( + self, component_class, default_kwargs + ): + """Integration: message_response must call self.set with the resolved system_prompt.""" + from unittest.mock import AsyncMock, MagicMock + + default_kwargs["system_prompt"] = "Powered by {model_name}." + default_kwargs["add_calculator_tool"] = False + default_kwargs["add_current_date_tool"] = False + component = await self.component_setup(component_class, default_kwargs) + component.model = [{"name": "gpt-4o", "provider": "OpenAI", "metadata": {}}] + component.get_memory_data = AsyncMock(return_value=[]) + component._get_shared_callbacks = list + component.set_tools_callbacks = lambda *_: None + + captured: dict = {} + + def fake_set(**kwargs): + captured.update(kwargs) + return component + + component.set = fake_set + component.create_agent_runnable = MagicMock(return_value=MagicMock()) + component.run_agent = AsyncMock(return_value=MagicMock()) + + with patch("lfx.components.models_and_agents.agent.get_llm") as mock_get_llm: + mock_get_llm.return_value = MockLanguageModel() + await component.message_response() + + assert captured.get("system_prompt") == "Powered by gpt-4o." + + async def test_should_not_mutate_format_instructions_when_json_response_runs(self, component_class, default_kwargs): + """Regression: injection must only touch agent_instructions, not format_instructions. + + Ensures literal {current_date}/{model_name} tokens in user-authored + format_instructions survive intact while the main system_prompt is + still replaced by the helper. + """ + from unittest.mock import AsyncMock, MagicMock + + default_kwargs["system_prompt"] = "Powered by {model_name}." + default_kwargs["format_instructions"] = "Return JSON with fields {current_date} and {model_name} preserved." + default_kwargs["add_calculator_tool"] = False + default_kwargs["add_current_date_tool"] = False + component = await self.component_setup(component_class, default_kwargs) + component.model = [{"name": "gpt-4o", "provider": "OpenAI", "metadata": {}}] + component.get_memory_data = AsyncMock(return_value=[]) + component._get_shared_callbacks = list + component.set_tools_callbacks = lambda *_: None + + captured: dict = {} + + def fake_set(**kwargs): + captured.update(kwargs) + return component + + component.set = fake_set + component.create_agent_runnable = MagicMock(return_value=MagicMock()) + component.run_agent = AsyncMock(return_value=MagicMock(content="{}")) + + with patch("lfx.components.models_and_agents.agent.get_llm") as mock_get_llm: + mock_get_llm.return_value = MockLanguageModel() + await component.json_response() + + prompt = captured.get("system_prompt") or "" + assert "Powered by gpt-4o." in prompt, "agent_instructions should have placeholders replaced" + assert "{current_date}" in prompt, "format_instructions literal braces must survive" + assert "{model_name} preserved" in prompt, "format_instructions literal braces must survive" + + async def test_should_accept_add_calculator_tool_in_default_keys(self, component_class, default_kwargs): + """update_build_config's default_keys validation must include add_calculator_tool.""" + from lfx.schema.dotdict import dotdict + + with patch("lfx.components.models_and_agents.agent.get_language_model_options") as mock_opts: + mock_opts.return_value = [ + { + "name": "gpt-4o", + "provider": "OpenAI", + "icon": "OpenAI", + "metadata": { + "model_class": "ChatOpenAI", + "model_name_param": "model", + "api_key_param": "api_key", + }, + } + ] + component = await self.component_setup(component_class, default_kwargs) + frontend_node = component.to_frontend_node() + build_config = frontend_node["data"]["node"]["template"] + + # add_calculator_tool must be present in the build_config already; if not, + # update_build_config will error listing it as missing. + assert "add_calculator_tool" in build_config + + updated_config = await component.update_build_config( + dotdict(build_config), mock_opts.return_value, field_name="model" + ) + assert "add_calculator_tool" in updated_config + + def test_should_have_placeholders_in_default_system_prompt(self, component_class): + """Default system_prompt ships with placeholders for the dynamic injection. + + Ensures {current_date} and {model_name} are visible on a fresh agent so + that the dynamic injection has an observable effect out-of-the-box. + """ + prompt_input = next( + (inp for inp in component_class.inputs if getattr(inp, "name", None) == "system_prompt"), + None, + ) + assert prompt_input is not None + assert "{current_date}" in prompt_input.value + assert "{model_name}" in prompt_input.value + class TestAgentComponentWithClient(ComponentTestBaseWithClient): @pytest.fixture diff --git a/src/backend/tests/unit/components/vectorstores/test_chroma_vector_store_component.py b/src/backend/tests/unit/components/vectorstores/test_chroma_vector_store_component.py index 2ebbb0662c..8ee0a20441 100644 --- a/src/backend/tests/unit/components/vectorstores/test_chroma_vector_store_component.py +++ b/src/backend/tests/unit/components/vectorstores/test_chroma_vector_store_component.py @@ -1,5 +1,6 @@ from pathlib import Path from typing import Any +from unittest.mock import MagicMock, patch import pytest from lfx.components.chroma import ChromaVectorStoreComponent @@ -8,6 +9,37 @@ from lfx.schema.data import Data from tests.base import ComponentTestBaseWithoutClient, VersionComponentMapping +def test_remote_chroma_server_uses_http_client() -> None: + mock_client = MagicMock() + mock_chroma = MagicMock() + mock_chroma.get.return_value = {"ids": [], "documents": [], "metadatas": []} + + with ( + patch("chromadb.HttpClient", return_value=mock_client) as mock_http_client, + patch("langchain_chroma.Chroma", return_value=mock_chroma) as mock_chroma_class, + ): + component = ChromaVectorStoreComponent().set( + collection_name="remote_collection", + persist_directory=None, + embedding=None, + chroma_server_host="chroma.example.com", + chroma_server_http_port=8100, + chroma_server_ssl_enabled=True, + ingest_data=[], + limit=None, + ) + + assert component.build_vector_store() is mock_chroma + + mock_http_client.assert_called_once_with(host="chroma.example.com", port=8100, ssl=True) + mock_chroma_class.assert_called_once_with( + persist_directory=None, + client=mock_client, + embedding_function=None, + collection_name="remote_collection", + ) + + @pytest.mark.api_key_required class TestChromaVectorStoreComponent(ComponentTestBaseWithoutClient): @pytest.fixture diff --git a/src/backend/tests/unit/test_knowledge_bases_api.py b/src/backend/tests/unit/test_knowledge_bases_api.py index 548eb55351..35f6699b3c 100644 --- a/src/backend/tests/unit/test_knowledge_bases_api.py +++ b/src/backend/tests/unit/test_knowledge_bases_api.py @@ -714,7 +714,7 @@ class TestPerformIngestionTask: @patch("langflow.api.utils.kb_helpers.KBStorageHelper.get_fresh_chroma_client") @patch("langflow.api.utils.kb_helpers.Chroma") - @patch("langflow.api.utils.kb_helpers.KBIngestionHelper._build_embeddings", new_callable=AsyncMock) + @patch("langflow.api.utils.kb_helpers.KBIngestionHelper.build_embeddings", new_callable=AsyncMock) @patch("langflow.api.utils.kb_helpers.KBAnalysisHelper.get_metadata") @patch("langflow.api.utils.kb_helpers.KBStorageHelper.get_directory_size") @patch("langflow.api.utils.kb_helpers.KBAnalysisHelper.update_text_metrics") @@ -764,7 +764,7 @@ class TestPerformIngestionTask: @patch("langflow.api.utils.kb_helpers.KBStorageHelper.get_fresh_chroma_client") @patch("langflow.api.utils.kb_helpers.Chroma") - @patch("langflow.api.utils.kb_helpers.KBIngestionHelper._build_embeddings", new_callable=AsyncMock) + @patch("langflow.api.utils.kb_helpers.KBIngestionHelper.build_embeddings", new_callable=AsyncMock) @patch("langflow.api.utils.kb_helpers.KBIngestionHelper.cleanup_chroma_chunks_by_job", new_callable=AsyncMock) async def test_perform_ingestion_rollback( self, mock_cleanup, mock_build, mock_chroma, mock_fresh_client, mock_kb_path diff --git a/src/backend/tests/unit/test_mcp_servers_orphan_migration.py b/src/backend/tests/unit/test_mcp_servers_orphan_migration.py new file mode 100644 index 0000000000..f0a1e7857a --- /dev/null +++ b/src/backend/tests/unit/test_mcp_servers_orphan_migration.py @@ -0,0 +1,247 @@ +"""Tests for migrate_orphaned_mcp_servers_config in langflow.services.utils. + +Verifies that MCP server config files written under a previous default +superuser's UUID are picked up and migrated to the new default superuser +when the database is reset but the config directory is preserved +(typical of containerized deployments without a persisted DB volume). +""" + +from __future__ import annotations + +import json +import os +from typing import TYPE_CHECKING +from uuid import uuid4 + +import pytest + +if TYPE_CHECKING: + from pathlib import Path +from langflow.services.database.models.file.model import File as UserFile +from langflow.services.deps import get_settings_service, session_scope +from langflow.services.utils import migrate_orphaned_mcp_servers_config +from sqlmodel import select + + +@pytest.fixture +async def initialized_services(monkeypatch, tmp_path): + """Initialize DB + services with an isolated config dir.""" + from langflow.services.utils import initialize_services, teardown_services + from lfx.services.manager import get_service_manager + + db_path = tmp_path / "test.db" + config_dir = tmp_path / "config" + config_dir.mkdir(parents=True, exist_ok=True) + + monkeypatch.setenv("LANGFLOW_DATABASE_URL", f"sqlite:///{db_path}") + monkeypatch.setenv("LANGFLOW_CONFIG_DIR", str(config_dir)) + monkeypatch.setenv("LANGFLOW_AUTO_LOGIN", "true") + + get_service_manager().factories.clear() + get_service_manager().services.clear() + + await initialize_services() + + yield config_dir + + await teardown_services() + + +def _write_orphan(config_dir: Path, payload: dict, *, mtime_offset: float = 0.0) -> Path: + """Create an orphaned _mcp_servers_{uuid}.json file in a UUID-named folder.""" + orphan_id = uuid4() + orphan_dir = config_dir / str(orphan_id) + orphan_dir.mkdir(parents=True, exist_ok=True) + orphan_path = orphan_dir / f"_mcp_servers_{orphan_id}.json" + orphan_path.write_text(json.dumps(payload)) + if mtime_offset: + stat = orphan_path.stat() + os.utime(orphan_path, (stat.st_atime + mtime_offset, stat.st_mtime + mtime_offset)) + return orphan_path + + +@pytest.mark.asyncio +@pytest.mark.timeout(30) +async def test_migrate_orphaned_mcp_servers_config_recovers_previous_user_config( + initialized_services, +): + """A single orphaned file is migrated to the current default superuser.""" + config_dir: Path = initialized_services + expected_payload = {"mcpServers": {"my-server": {"command": "uvx", "args": ["mcp-proxy"]}}} + orphan_path = _write_orphan(config_dir, expected_payload) + + settings = get_settings_service() + + async with session_scope() as session: + from langflow.services.database.models.user.model import User + from lfx.services.settings.constants import DEFAULT_SUPERUSER + + user = (await session.exec(select(User).where(User.username == DEFAULT_SUPERUSER))).first() + assert user is not None, "default superuser should exist after initialize_services" + + # Simulate the fresh-DB scenario: the user has no MCP config row yet. + stmt = select(UserFile).where(UserFile.user_id == user.id).where(UserFile.name == f"_mcp_servers_{user.id}") + assert (await session.exec(stmt)).first() is None + + migrated = await migrate_orphaned_mcp_servers_config(session, settings, user) + assert migrated is True + + # DB record should exist and point at the new user-specific path. + stmt = select(UserFile).where(UserFile.user_id == user.id).where(UserFile.name == f"_mcp_servers_{user.id}") + new_record = (await session.exec(stmt)).first() + assert new_record is not None + assert new_record.path == f"{user.id}/_mcp_servers_{user.id}.json" + + # File should live under the new user's folder with the same contents. + target = config_dir / str(user.id) / f"_mcp_servers_{user.id}.json" + assert target.exists() + assert json.loads(target.read_text()) == expected_payload + # Orphan source is left intact (best-effort copy, not destructive move). + assert orphan_path.exists() + + +@pytest.mark.asyncio +@pytest.mark.timeout(30) +async def test_migrate_orphaned_mcp_servers_config_skips_when_multiple_orphans( + initialized_services, +): + """With multiple orphan candidates we refuse to guess and leave state untouched. + + MCP server entries can contain env/headers auth material, so importing an + unrelated user's config would be a security hazard. Operators must resolve + the ambiguity manually. + """ + config_dir: Path = initialized_services + + old_payload = {"mcpServers": {"old": {}}} + new_payload = {"mcpServers": {"new": {}}} + + old_path = _write_orphan(config_dir, old_payload, mtime_offset=-3600) + new_path = _write_orphan(config_dir, new_payload) + + settings = get_settings_service() + + async with session_scope() as session: + from langflow.services.database.models.user.model import User + from lfx.services.settings.constants import DEFAULT_SUPERUSER + + user = (await session.exec(select(User).where(User.username == DEFAULT_SUPERUSER))).first() + migrated = await migrate_orphaned_mcp_servers_config(session, settings, user) + assert migrated is False + + stmt = select(UserFile).where(UserFile.user_id == user.id).where(UserFile.name == f"_mcp_servers_{user.id}") + assert (await session.exec(stmt)).first() is None + + target = config_dir / str(user.id) / f"_mcp_servers_{user.id}.json" + assert not target.exists() + # Both orphans are preserved on disk for manual recovery. + assert json.loads(old_path.read_text()) == old_payload + assert json.loads(new_path.read_text()) == new_payload + + +@pytest.mark.asyncio +@pytest.mark.timeout(30) +async def test_migrate_orphaned_mcp_servers_config_no_orphans_is_noop( + initialized_services, +): + """With no orphaned files the function returns False and does not touch the DB.""" + config_dir: Path = initialized_services + # No UUID-named subdirectories should exist. + from uuid import UUID + + def _is_uuid_dir(p: Path) -> bool: + if not p.is_dir(): + return False + try: + UUID(p.name) + except ValueError: + return False + return True + + assert not [p for p in config_dir.iterdir() if _is_uuid_dir(p)] + + settings = get_settings_service() + + async with session_scope() as session: + from langflow.services.database.models.user.model import User + from lfx.services.settings.constants import DEFAULT_SUPERUSER + + user = (await session.exec(select(User).where(User.username == DEFAULT_SUPERUSER))).first() + migrated = await migrate_orphaned_mcp_servers_config(session, settings, user) + assert migrated is False + + stmt = select(UserFile).where(UserFile.user_id == user.id).where(UserFile.name == f"_mcp_servers_{user.id}") + assert (await session.exec(stmt)).first() is None + + +@pytest.mark.asyncio +@pytest.mark.timeout(30) +async def test_migrate_orphaned_mcp_servers_config_self_heals_missing_db_row( + initialized_services, +): + """Re-register missing DB rows when the on-disk file already exists. + + If the user's file is on disk but the DB row is missing, the migration + should recreate the row instead of leaving the config invisible. This + covers recovery from a previous migration attempt that wrote the file + but crashed before committing the DB row. + """ + config_dir: Path = initialized_services + settings = get_settings_service() + + async with session_scope() as session: + from langflow.services.database.models.user.model import User + from lfx.services.settings.constants import DEFAULT_SUPERUSER + + user = (await session.exec(select(User).where(User.username == DEFAULT_SUPERUSER))).first() + + # Simulate a file-exists / DB-row-missing state. + existing_dir = config_dir / str(user.id) + existing_dir.mkdir(parents=True, exist_ok=True) + existing_payload = {"mcpServers": {"keep-me": {}}} + existing_path = existing_dir / f"_mcp_servers_{user.id}.json" + existing_path.write_text(json.dumps(existing_payload)) + + migrated = await migrate_orphaned_mcp_servers_config(session, settings, user) + assert migrated is True + + stmt = select(UserFile).where(UserFile.user_id == user.id).where(UserFile.name == f"_mcp_servers_{user.id}") + row = (await session.exec(stmt)).first() + assert row is not None + assert row.path == f"{user.id}/_mcp_servers_{user.id}.json" + assert row.size == existing_path.stat().st_size + + # File contents are untouched. + assert json.loads(existing_path.read_text()) == existing_payload + + +@pytest.mark.asyncio +@pytest.mark.timeout(30) +async def test_migrate_orphaned_mcp_servers_config_skips_when_row_already_present( + initialized_services, +): + """When the DB row already exists, do nothing — avoids double-registration.""" + config_dir: Path = initialized_services + _write_orphan(config_dir, {"mcpServers": {"orphan": {}}}) + + settings = get_settings_service() + + async with session_scope() as session: + from langflow.services.database.models.user.model import User + from lfx.services.settings.constants import DEFAULT_SUPERUSER + + user = (await session.exec(select(User).where(User.username == DEFAULT_SUPERUSER))).first() + + # Pre-register an MCP file row to simulate an already-migrated user. + session.add( + UserFile( + user_id=user.id, + name=f"_mcp_servers_{user.id}", + path=f"{user.id}/_mcp_servers_{user.id}.json", + size=0, + ) + ) + await session.commit() + + migrated = await migrate_orphaned_mcp_servers_config(session, settings, user) + assert migrated is False diff --git a/src/backend/tests/unit/test_memory_base_task.py b/src/backend/tests/unit/test_memory_base_task.py new file mode 100644 index 0000000000..db01cc8371 --- /dev/null +++ b/src/backend/tests/unit/test_memory_base_task.py @@ -0,0 +1,1018 @@ +"""Unit tests for langflow.services.memory_base.task. + +Covers the gaps not addressed by TestIngestMemoryTask in test_memory_bases.py: +- ingest_memory_task: missing kb_root, pre-ingestion cancel, zero-document early-out +- extract_content_block_text: all block types, edge cases +- build_documents_from_messages: chunking, content-block text, missing fields +- sync_kb_metadata: source_types merge +- _advance_cursor: vanished session, normal update +- _mark_messages_ingested: correct DB update shape +""" + +from __future__ import annotations + +import asyncio +import json +import uuid +from datetime import datetime, timezone +from unittest.mock import AsyncMock, MagicMock, patch + +import pytest +from langflow.services.database.models.message.model import MessageTable + +# ------------------------------------------------------------------ # +# Shared helpers # +# ------------------------------------------------------------------ # + + +def _make_message( + *, + flow_id: uuid.UUID | None = None, + session_id: str = "sess-1", + text: str = "hello", + run_id: uuid.UUID | None = None, + timestamp: datetime | None = None, + content_blocks: list | None = None, +) -> MessageTable: + return MessageTable( + id=uuid.uuid4(), + sender="AI", + sender_name="Bot", + session_id=session_id, + text=text, + flow_id=flow_id or uuid.uuid4(), + timestamp=timestamp or datetime.now(timezone.utc), + run_id=run_id, + content_blocks=content_blocks or [], + ) + + +def _fake_scope(mock_db): + class _FakeCtx: + async def __aenter__(self): + return mock_db + + async def __aexit__(self, *a): + pass + + scope = MagicMock() + scope.return_value = _FakeCtx() + return scope + + +# ------------------------------------------------------------------ # +# ingest_memory_task — orchestrator edge cases # +# ------------------------------------------------------------------ # + + +class TestIngestMemoryTaskEdgeCases: + @pytest.mark.asyncio + async def test_raises_when_kb_root_not_configured(self): + from langflow.services.memory_base.task import IngestionRequest, ingest_memory_task + + with ( + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=None, + ), + pytest.raises(RuntimeError, match="Knowledge base root path is not configured"), + ): + await ingest_memory_task( + request=IngestionRequest( + memory_base_id=uuid.uuid4(), + session_id="s1", + flow_id=uuid.uuid4(), + kb_name="kb", + kb_username="user", + user_id=uuid.uuid4(), + embedding_provider="OpenAI", + embedding_model="text-embedding-3-small", + cursor_id=None, + task_job_id=uuid.uuid4(), + job_service=MagicMock(), + ), + ) + + @pytest.mark.asyncio + async def test_returns_early_when_job_cancelled_before_write(self, tmp_path): + """is_job_cancelled=True after fetch must return without touching Chroma.""" + from langflow.services.memory_base.task import IngestionRequest, ingest_memory_task + + flow_id = uuid.uuid4() + msg = _make_message(flow_id=flow_id) + chroma_client_called = False + + def fake_get_client(_path): + nonlocal chroma_client_called + chroma_client_called = True + return MagicMock() + + with ( + patch( + "langflow.services.memory_base.task._acquire_session_lock", + AsyncMock(return_value=asyncio.Lock()), + ), + patch("langflow.services.memory_base.task._release_session_lock", AsyncMock()), + patch("langflow.services.memory_base.task._read_live_cursor", AsyncMock(return_value=None)), + patch( + "langflow.services.memory_base.task._fetch_pending_messages", + AsyncMock(return_value=[msg]), + ), + patch( + "langflow.services.memory_base.task.build_documents_from_messages", + return_value=[MagicMock()], + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.is_job_cancelled", + AsyncMock(return_value=True), + ), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_fresh_chroma_client", + side_effect=fake_get_client, + ), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path, + ), + ): + result = await ingest_memory_task( + request=IngestionRequest( + memory_base_id=uuid.uuid4(), + session_id="s1", + flow_id=flow_id, + kb_name="kb", + kb_username="user", + user_id=uuid.uuid4(), + embedding_provider="OpenAI", + embedding_model="text-embedding-3-small", + cursor_id=None, + task_job_id=uuid.uuid4(), + job_service=MagicMock(), + ), + ) + + assert result == {"message": "Job cancelled before ingestion", "ingested": 0} + assert not chroma_client_called + + @pytest.mark.asyncio + async def test_returns_early_when_documents_list_is_empty(self, tmp_path): + """All-whitespace messages produce zero documents — early exit before Chroma.""" + from langflow.services.memory_base.task import IngestionRequest, ingest_memory_task + + flow_id = uuid.uuid4() + msg = _make_message(flow_id=flow_id, text=" ") # whitespace only + + with ( + patch( + "langflow.services.memory_base.task._acquire_session_lock", + AsyncMock(return_value=asyncio.Lock()), + ), + patch("langflow.services.memory_base.task._release_session_lock", AsyncMock()), + patch("langflow.services.memory_base.task._read_live_cursor", AsyncMock(return_value=None)), + patch( + "langflow.services.memory_base.task._fetch_pending_messages", + AsyncMock(return_value=[msg]), + ), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path, + ), + ): + result = await ingest_memory_task( + request=IngestionRequest( + memory_base_id=uuid.uuid4(), + session_id="s1", + flow_id=flow_id, + kb_name="kb", + kb_username="user", + user_id=uuid.uuid4(), + embedding_provider="OpenAI", + embedding_model="text-embedding-3-small", + cursor_id=None, + task_job_id=uuid.uuid4(), + job_service=MagicMock(), + ), + ) + + assert result == {"message": "No non-empty messages to ingest", "ingested": 0} + + @pytest.mark.asyncio + async def test_mark_messages_ingested_called_on_success(self, tmp_path): + """_mark_messages_ingested must be called exactly once on a successful run.""" + from langflow.services.memory_base.task import IngestionRequest, ingest_memory_task + + flow_id = uuid.uuid4() + msg = _make_message(flow_id=flow_id) + + mark_ingested_mock = AsyncMock() + + with ( + patch( + "langflow.services.memory_base.task._acquire_session_lock", + AsyncMock(return_value=asyncio.Lock()), + ), + patch("langflow.services.memory_base.task._release_session_lock", AsyncMock()), + patch("langflow.services.memory_base.task._read_live_cursor", AsyncMock(return_value=None)), + patch( + "langflow.services.memory_base.task._fetch_pending_messages", + AsyncMock(return_value=[msg]), + ), + patch( + "langflow.services.memory_base.task.build_documents_from_messages", + return_value=[MagicMock()], + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.is_job_cancelled", + AsyncMock(return_value=False), + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.build_embeddings", + AsyncMock(return_value=MagicMock()), + ), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_fresh_chroma_client", + return_value=MagicMock(), + ), + patch("langflow.services.memory_base.task.Chroma"), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.write_documents_to_chroma", + AsyncMock(return_value=1), + ), + patch("langflow.services.memory_base.task.sync_kb_metadata"), + patch("langflow.services.memory_base.task._mark_messages_ingested", mark_ingested_mock), + patch("langflow.services.memory_base.task._advance_cursor", AsyncMock()), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path, + ), + patch("langflow.services.memory_base.task.KBStorageHelper.release_chroma_resources"), + ): + await ingest_memory_task( + request=IngestionRequest( + memory_base_id=uuid.uuid4(), + session_id="s1", + flow_id=flow_id, + kb_name="kb", + kb_username="user", + user_id=uuid.uuid4(), + embedding_provider="OpenAI", + embedding_model="text-embedding-3-small", + cursor_id=None, + task_job_id=uuid.uuid4(), + job_service=MagicMock(), + ), + ) + + mark_ingested_mock.assert_awaited_once() + + @pytest.mark.asyncio + async def test_mark_messages_ingested_not_called_when_cancelled(self, tmp_path): + """When write returns fewer docs than sent, messages must NOT be stamped.""" + from langflow.services.memory_base.task import IngestionRequest, ingest_memory_task + + flow_id = uuid.uuid4() + msg = _make_message(flow_id=flow_id) + + mark_ingested_mock = AsyncMock() + + with ( + patch( + "langflow.services.memory_base.task._acquire_session_lock", + AsyncMock(return_value=asyncio.Lock()), + ), + patch("langflow.services.memory_base.task._release_session_lock", AsyncMock()), + patch("langflow.services.memory_base.task._read_live_cursor", AsyncMock(return_value=None)), + patch( + "langflow.services.memory_base.task._fetch_pending_messages", + AsyncMock(return_value=[msg]), + ), + patch( + "langflow.services.memory_base.task.build_documents_from_messages", + return_value=[MagicMock()], + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.is_job_cancelled", + AsyncMock(return_value=False), + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.build_embeddings", + AsyncMock(return_value=MagicMock()), + ), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_fresh_chroma_client", + return_value=MagicMock(), + ), + patch("langflow.services.memory_base.task.Chroma"), + # Partial write simulates mid-run cancellation + patch( + "langflow.services.memory_base.task.KBIngestionHelper.write_documents_to_chroma", + AsyncMock(return_value=0), + ), + patch("langflow.services.memory_base.task.sync_kb_metadata"), + patch("langflow.services.memory_base.task._mark_messages_ingested", mark_ingested_mock), + patch("langflow.services.memory_base.task._advance_cursor", AsyncMock()), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path, + ), + patch("langflow.services.memory_base.task.KBStorageHelper.release_chroma_resources"), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.cleanup_chroma_chunks_by_job", + AsyncMock(), + ) as cleanup_mock, + ): + result = await ingest_memory_task( + request=IngestionRequest( + memory_base_id=uuid.uuid4(), + session_id="s1", + flow_id=flow_id, + kb_name="kb", + kb_username="user", + user_id=uuid.uuid4(), + embedding_provider="OpenAI", + embedding_model="text-embedding-3-small", + cursor_id=None, + task_job_id=uuid.uuid4(), + job_service=MagicMock(), + ), + ) + + mark_ingested_mock.assert_not_awaited() + assert "cancelled" in result["message"].lower() + cleanup_mock.assert_awaited_once() + + @pytest.mark.asyncio + async def test_cleanup_called_on_write_exception(self, tmp_path): + """When write_documents_to_chroma raises, partial chunks must be cleaned up.""" + from langflow.services.memory_base.task import IngestionRequest, ingest_memory_task + + flow_id = uuid.uuid4() + msg = _make_message(flow_id=flow_id) + + with ( + patch( + "langflow.services.memory_base.task._acquire_session_lock", + AsyncMock(return_value=asyncio.Lock()), + ), + patch("langflow.services.memory_base.task._release_session_lock", AsyncMock()), + patch("langflow.services.memory_base.task._read_live_cursor", AsyncMock(return_value=None)), + patch( + "langflow.services.memory_base.task._fetch_pending_messages", + AsyncMock(return_value=[msg]), + ), + patch( + "langflow.services.memory_base.task.build_documents_from_messages", + return_value=[MagicMock()], + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.is_job_cancelled", + AsyncMock(return_value=False), + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.build_embeddings", + AsyncMock(return_value=MagicMock()), + ), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_fresh_chroma_client", + return_value=MagicMock(), + ), + patch("langflow.services.memory_base.task.Chroma"), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.write_documents_to_chroma", + AsyncMock(side_effect=RuntimeError("Chroma write failed")), + ), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path, + ), + patch("langflow.services.memory_base.task.KBStorageHelper.release_chroma_resources"), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.cleanup_chroma_chunks_by_job", + AsyncMock(), + ) as cleanup_mock, + pytest.raises(RuntimeError, match="Chroma write failed"), + ): + await ingest_memory_task( + request=IngestionRequest( + memory_base_id=uuid.uuid4(), + session_id="s1", + flow_id=flow_id, + kb_name="kb", + kb_username="user", + user_id=uuid.uuid4(), + embedding_provider="OpenAI", + embedding_model="text-embedding-3-small", + cursor_id=None, + task_job_id=uuid.uuid4(), + job_service=MagicMock(), + ), + ) + + cleanup_mock.assert_awaited_once() + + +# ------------------------------------------------------------------ # +# extract_content_block_text # +# ------------------------------------------------------------------ # + + +class TestExtractContentBlockText: + def _call(self, blocks): + from langflow.services.memory_base.document_builders import extract_content_block_text + + return extract_content_block_text(blocks) + + def test_empty_list_returns_empty_string(self): + assert self._call([]) == "" + + def test_text_block_extracted(self): + blocks = [{"contents": [{"type": "text", "text": "hello world"}]}] + result = self._call(blocks) + assert result == "hello world" + + def test_text_block_whitespace_only_skipped(self): + blocks = [{"contents": [{"type": "text", "text": " "}]}] + result = self._call(blocks) + assert result == "" + + def test_code_block_with_language(self): + blocks = [{"contents": [{"type": "code", "language": "python", "code": "print('hi')"}]}] + result = self._call(blocks) + assert result == "```python\nprint('hi')\n```" + + def test_code_block_without_language(self): + blocks = [{"contents": [{"type": "code", "language": "", "code": "x = 1"}]}] + result = self._call(blocks) + assert result == "```\nx = 1\n```" + + def test_code_block_empty_code_skipped(self): + blocks = [{"contents": [{"type": "code", "language": "python", "code": ""}]}] + result = self._call(blocks) + assert result == "" + + def test_json_block_serialized(self): + data = {"key": "value", "num": 42} + blocks = [{"contents": [{"type": "json", "data": data}]}] + result = self._call(blocks) + assert result == json.dumps(data, ensure_ascii=False) + + def test_json_block_none_data_skipped(self): + blocks = [{"contents": [{"type": "json", "data": None}]}] + result = self._call(blocks) + assert result == "" + + def test_unknown_block_type_skipped(self): + blocks = [ + { + "contents": [ + {"type": "tool_use", "tool": "search"}, + {"type": "error", "message": "failed"}, + {"type": "media", "url": "http://x.com/img.png"}, + {"type": "text", "text": "kept"}, + ] + } + ] + result = self._call(blocks) + assert result == "kept" + + def test_non_dict_entry_skipped(self): + # entries that aren't dicts should be silently skipped + blocks = [{"contents": ["not a dict", 42, None, {"type": "text", "text": "ok"}]}] + result = self._call(blocks) + assert result == "ok" + + def test_non_dict_block_skipped(self): + # top-level block that isn't a dict + blocks = ["string block", {"contents": [{"type": "text", "text": "valid"}]}] + result = self._call(blocks) + assert result == "valid" + + def test_multiple_blocks_joined_with_double_newline(self): + blocks = [ + {"contents": [{"type": "text", "text": "first"}]}, + {"contents": [{"type": "text", "text": "second"}]}, + ] + result = self._call(blocks) + assert result == "first\n\nsecond" + + def test_multiple_entries_in_same_block_joined(self): + blocks = [ + { + "contents": [ + {"type": "text", "text": "a"}, + {"type": "text", "text": "b"}, + ] + } + ] + result = self._call(blocks) + assert result == "a\n\nb" + + +# ------------------------------------------------------------------ # +# build_documents_from_messages # +# ------------------------------------------------------------------ # + + +class TestBuildDocumentsFromMessages: + def _call(self, messages, *, session_id="s1", flow_id=None, job_id="test-job-id"): + from langflow.services.memory_base.document_builders import build_documents_from_messages + + return build_documents_from_messages( + messages, + session_id=session_id, + flow_id=flow_id or str(uuid.uuid4()), + job_id=job_id, + ) + + def test_content_blocks_contribute_to_doc_text(self): + flow_id = uuid.uuid4() + msg = _make_message( + flow_id=flow_id, + text="", + content_blocks=[{"contents": [{"type": "text", "text": "from block"}]}], + ) + docs = self._call([msg], flow_id=str(flow_id)) + assert len(docs) == 1 + assert "from block" in docs[0].page_content + + def test_text_and_content_blocks_combined(self): + flow_id = uuid.uuid4() + msg = _make_message( + flow_id=flow_id, + text="msg text", + content_blocks=[{"contents": [{"type": "text", "text": "block text"}]}], + ) + docs = self._call([msg], flow_id=str(flow_id)) + assert len(docs) == 1 + assert "msg text" in docs[0].page_content + assert "block text" in docs[0].page_content + + def test_long_message_split_into_multiple_chunks(self): + from langflow.services.memory_base.document_builders import MESSAGE_CHUNK_SIZE + + flow_id = uuid.uuid4() + # Craft text longer than chunk size + long_text = "x " * (MESSAGE_CHUNK_SIZE + 100) + msg = _make_message(flow_id=flow_id, text=long_text) + docs = self._call([msg], flow_id=str(flow_id)) + assert len(docs) > 1 + + def test_chunk_index_and_total_chunks_metadata(self): + from langflow.services.memory_base.document_builders import MESSAGE_CHUNK_SIZE + + flow_id = uuid.uuid4() + long_text = "word " * (MESSAGE_CHUNK_SIZE // 4) + msg = _make_message(flow_id=flow_id, text=long_text) + docs = self._call([msg], flow_id=str(flow_id)) + for i, doc in enumerate(docs): + assert doc.metadata["chunk_index"] == i + assert doc.metadata["total_chunks"] == len(docs) + + def test_missing_run_id_stored_as_empty_string(self): + flow_id = uuid.uuid4() + msg = _make_message(flow_id=flow_id, run_id=None) + docs = self._call([msg], flow_id=str(flow_id)) + assert docs[0].metadata["run_id"] == "" + + def test_run_id_stored_as_string(self): + flow_id = uuid.uuid4() + run_id = uuid.uuid4() + msg = _make_message(flow_id=flow_id, run_id=run_id) + docs = self._call([msg], flow_id=str(flow_id)) + assert docs[0].metadata["run_id"] == str(run_id) + + def test_missing_timestamp_stored_as_empty_string(self): + flow_id = uuid.uuid4() + msg = _make_message(flow_id=flow_id) + # validate_assignment=True on the model prevents setting timestamp=None + # directly, so bypass Pydantic validation via object.__setattr__. + object.__setattr__(msg, "timestamp", None) + docs = self._call([msg], flow_id=str(flow_id)) + assert docs[0].metadata["timestamp"] == "" + + def test_source_metadata_uses_session_id(self): + flow_id = uuid.uuid4() + msg = _make_message(flow_id=flow_id, session_id="my-session") + docs = self._call([msg], session_id="my-session", flow_id=str(flow_id)) + assert docs[0].metadata["source"] == "memory_base/my-session" + + def test_job_id_included_in_metadata(self): + flow_id = uuid.uuid4() + job_id = str(uuid.uuid4()) + msg = _make_message(flow_id=flow_id) + docs = self._call([msg], flow_id=str(flow_id), job_id=job_id) + assert docs[0].metadata["job_id"] == job_id + + def test_job_id_defaults_to_empty_string(self): + flow_id = uuid.uuid4() + msg = _make_message(flow_id=flow_id) + docs = self._call([msg], flow_id=str(flow_id), job_id="") + assert docs[0].metadata["job_id"] == "" + + def test_multiple_messages_produce_separate_docs(self): + flow_id = uuid.uuid4() + msgs = [_make_message(flow_id=flow_id, text=f"msg {i}") for i in range(3)] + docs = self._call(msgs, flow_id=str(flow_id)) + assert len(docs) == 3 + message_ids = [d.metadata["message_id"] for d in docs] + assert len(set(message_ids)) == 3 + + +# ------------------------------------------------------------------ # +# sync_kb_metadata # +# ------------------------------------------------------------------ # + + +class TestSyncKbMetadata: + def test_preserves_existing_source_types(self, tmp_path): + from langflow.services.memory_base.document_builders import sync_kb_metadata as _sync_kb_metadata + + kb_path = tmp_path / "kb" + kb_path.mkdir() + + with ( + patch( + "langflow.services.memory_base.document_builders.KBAnalysisHelper.get_metadata", + return_value={"chunks": 5, "source_types": ["file"]}, + ), + patch("langflow.services.memory_base.document_builders.KBAnalysisHelper.update_text_metrics"), + patch( + "langflow.services.memory_base.document_builders.KBStorageHelper.get_directory_size", return_value=2048 + ), + ): + _sync_kb_metadata(kb_path=kb_path, chroma=MagicMock()) + + written = json.loads((kb_path / "embedding_metadata.json").read_text()) + assert "file" in written["source_types"] + assert "memory" in written["source_types"] + + def test_source_types_sorted(self, tmp_path): + from langflow.services.memory_base.document_builders import sync_kb_metadata as _sync_kb_metadata + + kb_path = tmp_path / "kb" + kb_path.mkdir() + + with ( + patch( + "langflow.services.memory_base.document_builders.KBAnalysisHelper.get_metadata", + return_value={"chunks": 0, "source_types": ["zzz", "aaa"]}, + ), + patch("langflow.services.memory_base.document_builders.KBAnalysisHelper.update_text_metrics"), + patch("langflow.services.memory_base.document_builders.KBStorageHelper.get_directory_size", return_value=0), + ): + _sync_kb_metadata(kb_path=kb_path, chroma=MagicMock()) + + written = json.loads((kb_path / "embedding_metadata.json").read_text()) + assert written["source_types"] == sorted(written["source_types"]) + + def test_json_decode_error_swallowed(self, tmp_path): + from langflow.services.memory_base.document_builders import sync_kb_metadata as _sync_kb_metadata + + kb_path = tmp_path / "kb" + kb_path.mkdir() + + with patch( + "langflow.services.memory_base.document_builders.KBAnalysisHelper.get_metadata", + side_effect=json.JSONDecodeError("bad", "", 0), + ): + # Must not raise + _sync_kb_metadata(kb_path=kb_path, chroma=MagicMock()) + + def test_value_error_swallowed(self, tmp_path): + from langflow.services.memory_base.document_builders import sync_kb_metadata as _sync_kb_metadata + + kb_path = tmp_path / "kb" + kb_path.mkdir() + + with ( + patch( + "langflow.services.memory_base.document_builders.KBAnalysisHelper.get_metadata", + return_value={}, + ), + patch( + "langflow.services.memory_base.document_builders.KBAnalysisHelper.update_text_metrics", + side_effect=ValueError("bad metric"), + ), + ): + _sync_kb_metadata(kb_path=kb_path, chroma=MagicMock()) + + +# ------------------------------------------------------------------ # +# _advance_cursor # +# ------------------------------------------------------------------ # + + +class TestAdvanceCursor: + @pytest.mark.asyncio + async def test_normal_update(self): + from langflow.services.database.models.memory_base.model import MemoryBaseSession + from langflow.services.memory_base.task import _advance_cursor + + mb_id = uuid.uuid4() + new_cursor = uuid.uuid4() + task_job_id = uuid.uuid4() + + mbs = MemoryBaseSession( + id=uuid.uuid4(), + memory_base_id=mb_id, + session_id="s1", + cursor_id=None, + total_processed=5, + ) + + mock_db = AsyncMock() + mock_select_result = MagicMock() + mock_select_result.first = MagicMock(return_value=mbs) + # First exec = SELECT, second exec = UPDATE MemoryBaseWorkflowRun + mock_db.exec = AsyncMock(side_effect=[mock_select_result, MagicMock()]) + + await _advance_cursor( + mock_db, + memory_base_id=mb_id, + session_id="s1", + new_cursor_id=new_cursor, + ingested_count=3, + task_job_id=task_job_id, + ) + + assert mbs.cursor_id == new_cursor + assert mbs.total_processed == 8 # 5 + 3 + assert mbs.last_sync_at is not None + mock_db.add.assert_called_once_with(mbs) + mock_db.commit.assert_awaited_once() + + @pytest.mark.asyncio + async def test_vanished_session_does_not_raise(self): + """If MemoryBaseSession is gone, _advance_cursor must log a warning and return.""" + from langflow.services.memory_base.task import _advance_cursor + + mock_db = AsyncMock() + mock_result = MagicMock() + mock_result.first = MagicMock(return_value=None) # session vanished + mock_db.exec = AsyncMock(return_value=mock_result) + + await _advance_cursor( + mock_db, + memory_base_id=uuid.uuid4(), + session_id="gone", + new_cursor_id=uuid.uuid4(), + ingested_count=1, + task_job_id=uuid.uuid4(), + ) + + mock_db.add.assert_not_called() + mock_db.commit.assert_not_awaited() + + +# ------------------------------------------------------------------ # +# _mark_messages_ingested # +# ------------------------------------------------------------------ # + + +class TestMarkMessagesIngested: + @pytest.mark.asyncio + async def test_executes_bulk_update(self): + from langflow.services.memory_base.task import _mark_messages_ingested + + flow_id = uuid.uuid4() + messages = [_make_message(flow_id=flow_id) for _ in range(3)] + job_id = uuid.uuid4() + memory_base_id = uuid.uuid4() + + mock_conn = MagicMock() + mock_conn.dialect.name = "sqlite" + + mock_db = AsyncMock() + mock_db.exec = AsyncMock() + mock_db.connection = AsyncMock(return_value=mock_conn) + + await _mark_messages_ingested(mock_db, messages=messages, job_id=job_id, memory_base_id=memory_base_id) + + mock_db.exec.assert_awaited_once() + mock_db.commit.assert_awaited_once() + + @pytest.mark.asyncio + async def test_update_sets_ingestion_job_id_and_timestamp(self): + """The INSERT statement must include job_id and ingested_at for each message.""" + from langflow.services.memory_base.task import _mark_messages_ingested + + flow_id = uuid.uuid4() + messages = [_make_message(flow_id=flow_id)] + job_id = uuid.uuid4() + memory_base_id = uuid.uuid4() + + captured_stmt = {} + + mock_conn = MagicMock() + mock_conn.dialect.name = "sqlite" + + async def capture_exec(stmt): + captured_stmt["stmt"] = stmt + return MagicMock() + + mock_db = AsyncMock() + mock_db.exec = capture_exec + mock_db.connection = AsyncMock(return_value=mock_conn) + + await _mark_messages_ingested(mock_db, messages=messages, job_id=job_id, memory_base_id=memory_base_id) + + assert "stmt" in captured_stmt, "db.exec was not called" + stmt = captured_stmt["stmt"] + # The INSERT statement must reference job_id and ingested_at columns + stmt_str = str(stmt.compile()) + assert "job_id" in stmt_str + assert "ingested_at" in stmt_str + + +# ------------------------------------------------------------------ # +# TestIngestionLocking — serialization and cursor re-read # +# ------------------------------------------------------------------ # + + +class TestIngestionLocking: + """Tests for the per-session lock, live cursor re-read, and graceful early-exits.""" + + _BASE_KWARGS: dict = { + "session_id": "s1", + "kb_name": "kb", + "kb_username": "user", + "embedding_provider": "OpenAI", + "embedding_model": "text-embedding-3-small", + } + + @pytest.mark.asyncio + async def test_live_cursor_used_not_dispatch_snapshot(self, tmp_path): + """_fetch_pending_messages must receive the live cursor, not the dispatch-time one.""" + import langflow.services.memory_base.task as task_module + + memory_base_id = uuid.uuid4() + flow_id = uuid.uuid4() + dispatch_cursor = uuid.uuid4() # what was captured at dispatch time + live_cursor = uuid.uuid4() # what the DB currently says + + fetch_calls: list = [] + + async def _recording_fetch(db, *, flow_id, session_id, cursor_id): # noqa: ARG001 + fetch_calls.append(cursor_id) + return [] # empty → early exit; no Chroma setup needed + + task_module._session_ingestion_locks.pop((memory_base_id, "s1"), None) + + with ( + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path, + ), + patch( + "langflow.services.memory_base.task._read_live_cursor", + AsyncMock(return_value=live_cursor), + ), + patch( + "langflow.services.memory_base.task._fetch_pending_messages", + side_effect=_recording_fetch, + ), + ): + result = await task_module.ingest_memory_task( + request=task_module.IngestionRequest( + memory_base_id=memory_base_id, + flow_id=flow_id, + user_id=uuid.uuid4(), + cursor_id=dispatch_cursor, + task_job_id=uuid.uuid4(), + job_service=MagicMock(), + **self._BASE_KWARGS, + ), + ) + + assert result == {"message": "No pending messages", "ingested": 0} + assert len(fetch_calls) == 1 + assert fetch_calls[0] == live_cursor, ( + f"Expected fetch called with live_cursor={live_cursor!r}, got {fetch_calls[0]!r}" + ) + + @pytest.mark.asyncio + async def test_lock_released_on_task_exception(self, tmp_path): + """Lock must be released via finally even when the task raises inside the lock body.""" + import langflow.services.memory_base.task as task_module + + memory_base_id = uuid.uuid4() + + task_module._session_ingestion_locks.pop((memory_base_id, "s1"), None) + + with ( + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path, + ), + patch( + "langflow.services.memory_base.task._read_live_cursor", + AsyncMock(return_value=None), + ), + patch( + "langflow.services.memory_base.task._fetch_pending_messages", + AsyncMock(side_effect=RuntimeError("DB exploded inside lock")), + ), + pytest.raises(RuntimeError, match="DB exploded inside lock"), + ): + await task_module.ingest_memory_task( + request=task_module.IngestionRequest( + memory_base_id=memory_base_id, + flow_id=uuid.uuid4(), + user_id=uuid.uuid4(), + cursor_id=None, + task_job_id=uuid.uuid4(), + job_service=MagicMock(), + **self._BASE_KWARGS, + ), + ) + + lock = task_module._session_ingestion_locks[(memory_base_id, "s1")] + assert not lock.locked(), "Lock must be released after exception (finally block must have run)" + + @pytest.mark.asyncio + async def test_lock_timeout_raises_asyncio_timeout_error(self, tmp_path): + """When the lock cannot be acquired within the timeout, asyncio.TimeoutError is raised. + + This allows execute_with_status to record JobStatus.TIMED_OUT for an accurate audit trail. + """ + import langflow.services.memory_base.task as task_module + + memory_base_id = uuid.uuid4() + + task_module._session_ingestion_locks.pop((memory_base_id, "s1"), None) + # Use the factory so the lock is inserted into the WeakValueDictionary; + # holding blocking_lock as a strong reference keeps it alive there so the + # task finds the same lock object and blocks on acquire. + blocking_lock = task_module._get_or_create_session_lock((memory_base_id, "s1")) + await blocking_lock.acquire() # hold it — task will timeout waiting + + try: + with ( + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path, + ), + patch( + "langflow.services.memory_base.task.get_settings_service", + return_value=MagicMock(settings=MagicMock(max_ingestion_timeout_secs=0.01)), + ), + pytest.raises(asyncio.TimeoutError), + ): + await task_module.ingest_memory_task( + request=task_module.IngestionRequest( + memory_base_id=memory_base_id, + flow_id=uuid.uuid4(), + user_id=uuid.uuid4(), + cursor_id=None, + task_job_id=uuid.uuid4(), + job_service=MagicMock(), + **self._BASE_KWARGS, + ), + ) + finally: + blocking_lock.release() + + @pytest.mark.asyncio + async def test_noop_when_cursor_advanced_by_prior_job(self, tmp_path): + """If a prior job already advanced the cursor to msg3, fetch from msg3 finds nothing — graceful exit.""" + import langflow.services.memory_base.task as task_module + + memory_base_id = uuid.uuid4() + msg3_id = uuid.uuid4() + + advance_cursor_mock = AsyncMock() + + task_module._session_ingestion_locks.pop((memory_base_id, "s1"), None) + + with ( + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path, + ), + patch( + "langflow.services.memory_base.task._read_live_cursor", + AsyncMock(return_value=msg3_id), # prior job advanced to msg3 + ), + patch( + "langflow.services.memory_base.task._fetch_pending_messages", + AsyncMock(return_value=[]), # nothing after msg3 + ), + patch( + "langflow.services.memory_base.task._advance_cursor", + advance_cursor_mock, + ), + ): + result = await task_module.ingest_memory_task( + request=task_module.IngestionRequest( + memory_base_id=memory_base_id, + flow_id=uuid.uuid4(), + user_id=uuid.uuid4(), + cursor_id=None, # dispatch-time snapshot before prior job ran + task_job_id=uuid.uuid4(), + job_service=MagicMock(), + **self._BASE_KWARGS, + ), + ) + + assert result == {"message": "No pending messages", "ingested": 0} + advance_cursor_mock.assert_not_awaited() diff --git a/src/backend/tests/unit/test_memory_bases.py b/src/backend/tests/unit/test_memory_bases.py new file mode 100644 index 0000000000..4cac5fc98f --- /dev/null +++ b/src/backend/tests/unit/test_memory_bases.py @@ -0,0 +1,1719 @@ +"""Unit tests for the MemoryBase feature. + +Coverage areas: +- DB model creation and field defaults +- MemoryBaseService CRUD operations +- Concurrency guard (409 on duplicate active job) +- Cursor atomicity (cursor not advanced on ingestion failure) +- Threshold-change deferral +- FS/VectorDB mismatch detection +- Regenerate: cursor reset + re-trigger +- API endpoint routing (happy path + error paths) +- ingest_memory_task: pending message fetch, document building, cursor advance +""" + +from __future__ import annotations + +import asyncio +import contextlib +import uuid +from datetime import datetime, timezone +from unittest.mock import AsyncMock, MagicMock, patch + +import pytest +from langflow.services.database.models.memory_base.model import ( + MemoryBase, + MemoryBaseCreate, + MemoryBaseSession, + MemoryBaseUpdate, +) +from langflow.services.database.models.message.model import MessageTable + +# ------------------------------------------------------------------ # +# Helpers # +# ------------------------------------------------------------------ # + + +def _make_mb( + *, + user_id: uuid.UUID | None = None, + flow_id: uuid.UUID | None = None, + threshold: int = 10, + auto_capture: bool = True, +) -> MemoryBase: + return MemoryBase( + id=uuid.uuid4(), + name="test_mb", + flow_id=flow_id or uuid.uuid4(), + user_id=user_id or uuid.uuid4(), + threshold=threshold, + kb_name="test_kb", + auto_capture=auto_capture, + created_at=datetime.now(timezone.utc), + ) + + +def _make_session( + *, + memory_base_id: uuid.UUID | None = None, + session_id: str = "sess-1", + cursor_id: uuid.UUID | None = None, + total_processed: int = 0, +) -> MemoryBaseSession: + return MemoryBaseSession( + id=uuid.uuid4(), + memory_base_id=memory_base_id or uuid.uuid4(), + session_id=session_id, + cursor_id=cursor_id, + total_processed=total_processed, + ) + + +def _make_message( + *, + flow_id: uuid.UUID, + session_id: str, + is_output: bool = True, + text: str = "Hello from the bot", + run_id: uuid.UUID | None = None, +) -> MessageTable: + return MessageTable( + id=uuid.uuid4(), + sender="AI", + sender_name="Bot", + session_id=session_id, + text=text, + flow_id=flow_id, + is_output=is_output, + run_id=run_id, + timestamp=datetime.now(timezone.utc), + ) + + +# ------------------------------------------------------------------ # +# Model tests # +# ------------------------------------------------------------------ # + + +class TestMemoryBaseModel: + def test_defaults(self): + mb = MemoryBase( + name="mb", + flow_id=uuid.uuid4(), + user_id=uuid.uuid4(), + kb_name="kb", + ) + assert mb.threshold == 50 + assert mb.auto_capture is True + + def test_create_schema(self): + payload = MemoryBaseCreate( + name="mb", + flow_id=uuid.uuid4(), + user_id=uuid.uuid4(), + threshold=25, + kb_name="kb", + ) + assert payload.threshold == 25 + + def test_update_schema_partial(self): + patch = MemoryBaseUpdate(threshold=100) + dumped = patch.model_dump(exclude_unset=True) + assert "threshold" in dumped + assert "name" not in dumped + + def test_memory_base_session_defaults(self): + mbs = MemoryBaseSession( + memory_base_id=uuid.uuid4(), + session_id="s1", + ) + assert mbs.cursor_id is None + assert mbs.total_processed == 0 + assert mbs.last_sync_at is None + + +class TestMessageExtensions: + """Ensure the new fields exist on MessageTable.""" + + def test_run_id_field_exists(self): + msg = _make_message(flow_id=uuid.uuid4(), session_id="s1") + assert hasattr(msg, "run_id") + assert msg.run_id is None + + def test_is_output_field_defaults_false(self): + msg = MessageTable( + sender="Human", + sender_name="User", + session_id="s1", + text="hi", + ) + assert msg.is_output is False + + def test_is_output_can_be_set(self): + msg = _make_message(flow_id=uuid.uuid4(), session_id="s1", is_output=True) + assert msg.is_output is True + + +# ------------------------------------------------------------------ # +# Service tests (mock DB) # +# ------------------------------------------------------------------ # + + +class TestMemoryBaseServiceCRUD: + @pytest.fixture + def service(self): + from langflow.services.memory_base.service import MemoryBaseService + + return MemoryBaseService() + + @pytest.mark.asyncio + async def test_create_stores_user_id(self, service): + user_id = uuid.uuid4() + payload = MemoryBaseCreate( + name="mb", + flow_id=uuid.uuid4(), + user_id=user_id, + kb_name="kb", + ) + + created_mb = _make_mb(user_id=user_id) + + with patch.object(service, "create", AsyncMock(return_value=created_mb)): + result = await service.create(payload, user_id=user_id) + + assert result.user_id == user_id + + @pytest.mark.asyncio + async def test_get_returns_none_for_wrong_user(self, service): + with patch.object(service, "get", AsyncMock(return_value=None)): + result = await service.get(uuid.uuid4(), user_id=uuid.uuid4()) + assert result is None + + @pytest.mark.asyncio + async def test_update_returns_none_for_missing(self, service): + with patch.object(service, "update", AsyncMock(return_value=None)): + result = await service.update(uuid.uuid4(), uuid.uuid4(), MemoryBaseUpdate(threshold=5)) + assert result is None + + @pytest.mark.asyncio + async def test_delete_returns_false_for_missing(self, service): + with patch.object(service, "delete", AsyncMock(return_value=False)): + result = await service.delete(uuid.uuid4(), user_id=uuid.uuid4()) + assert result is False + + +class TestMemoryBaseCreateFlowOwnership: + """Regression tests for cross-user data leak via flow_id at creation time. + + A user must not be able to create a Memory Base pointed at another user's + flow. Without the ownership check, on_flow_output() would capture that + flow's conversation history into the attacker's Memory Base, which they + could then read via /sessions + /messages. + """ + + @pytest.fixture + def service(self): + from langflow.services.memory_base.service import MemoryBaseService + + return MemoryBaseService() + + def _fake_scope(self, mock_db): + class _FakeCtx: + async def __aenter__(self): + return mock_db + + async def __aexit__(self, *a): + pass + + scope = MagicMock() + scope.return_value = _FakeCtx() + return scope + + @pytest.mark.asyncio + async def test_create_rejects_unowned_flow(self, service): + """PermissionError raised when flow_id belongs to a different user.""" + user_id = uuid.uuid4() + payload = MemoryBaseCreate(name="mb", flow_id=uuid.uuid4()) + + mock_db = AsyncMock() + # exec() is async, but .first() on the result is synchronous — use MagicMock + # so that flow_result.first() returns None without returning a coroutine. + exec_result = MagicMock() + exec_result.first.return_value = None + mock_db.exec = AsyncMock(return_value=exec_result) + + with ( + patch("langflow.services.memory_base.service.session_scope", self._fake_scope(mock_db)), + pytest.raises(PermissionError, match="not found"), + ): + await service.create(payload, user_id=user_id) + + @pytest.mark.asyncio + async def test_create_allows_owned_flow(self, service): + """No PermissionError when flow_id is owned by the requesting user.""" + from langflow.services.database.models.flow.model import Flow + + user_id = uuid.uuid4() + flow_id = uuid.uuid4() + payload = MemoryBaseCreate(name="mb", flow_id=flow_id) + + owned_flow = Flow(id=flow_id, user_id=user_id, name="my flow") + created_mb = _make_mb(user_id=user_id, flow_id=flow_id) + + # exec() is async; .first() on the result is synchronous — use MagicMock. + # First call: flow ownership check → returns owned_flow (passes). + # Second call (different session_scope): name-uniqueness check → None. + first_exec = MagicMock() + first_exec.first.return_value = owned_flow + second_exec = MagicMock() + second_exec.first.return_value = None + + mock_db = AsyncMock() + mock_db.exec = AsyncMock(side_effect=[first_exec, second_exec]) + mock_db.refresh = AsyncMock(return_value=created_mb) + + with ( + patch("langflow.services.memory_base.service.session_scope", self._fake_scope(mock_db)), + patch( + "langflow.services.memory_base.kb_path_helpers.resolve_kb_username", AsyncMock(return_value="testuser") + ), + patch("langflow.services.memory_base.kb_path_helpers.initialize_kb", AsyncMock()), + contextlib.suppress(Exception), + ): + # Should not raise PermissionError — ownership check passes + await service.create(payload, user_id=user_id) + + # The flow ownership query must have been executed + mock_db.exec.assert_called() + + @pytest.mark.asyncio + async def test_create_endpoint_returns_404_for_unowned_flow(self): + """POST /memories returns 404 (not 403/409) when flow_id is unowned. + + 404 is intentional: returning 403 would reveal that the flow exists, + which is an information leak in itself. + Tests the try/except mapping in the route handler directly. + """ + from fastapi import HTTPException + from langflow.api.v1.memories import create_memory_base + from langflow.services.database.models.user.model import User + + fake_user = User(id=uuid.uuid4(), username="alice") + mock_service = MagicMock() + mock_service.create = AsyncMock(side_effect=PermissionError("Flow abc not found")) + + with ( + patch("langflow.api.v1.memories.get_memory_base_service", return_value=mock_service), + pytest.raises(HTTPException) as exc_info, + ): + await create_memory_base( + current_user=fake_user, + payload=MemoryBaseCreate(name="mb", flow_id=uuid.uuid4()), + ) + + assert exc_info.value.status_code == 404 + + +class TestMemoryBaseServiceConcurrency: + """409 guard: only one active ingestion per (memory_base_id, session_id).""" + + @pytest.fixture + def service(self): + from langflow.services.memory_base.service import MemoryBaseService + + return MemoryBaseService() + + def _fake_scope(self, mock_db): + class _FakeCtx: + async def __aenter__(self): + return mock_db + + async def __aexit__(self, *a): + pass + + scope = MagicMock() + scope.return_value = _FakeCtx() + return scope + + @pytest.mark.asyncio + async def test_trigger_raises_when_job_active(self, service): + """DuplicateJobError from create_job propagates out of trigger_ingestion.""" + from langflow.services.jobs import DuplicateJobError + + mb = _make_mb() + mbs = _make_session(memory_base_id=mb.id) + mock_db = AsyncMock() + + mock_job_svc = MagicMock() + mock_job_svc.create_job = AsyncMock(side_effect=DuplicateJobError("already running")) + + with ( + patch("langflow.services.memory_base.ingestion.session_scope", self._fake_scope(mock_db)), + patch.object(service, "get_memory_base_or_404", AsyncMock(return_value=mb)), + patch.object(service, "_get_or_create_session", AsyncMock(return_value=mbs)), + patch( + "langflow.services.memory_base.ingestion._get_latest_pending_workflow_job_id", + AsyncMock(return_value=uuid.uuid4()), + ), + patch("langflow.services.memory_base.ingestion.resolve_kb_username", AsyncMock(return_value="testuser")), + patch( + "langflow.services.memory_base.ingestion.resolve_embedding", + return_value=("OpenAI", "text-embedding-3-small"), + ), + patch("langflow.services.memory_base.ingestion.get_job_service", return_value=mock_job_svc), + pytest.raises(DuplicateJobError), + ): + await service.trigger_ingestion(mb.id, mb.user_id, "sess-1") + + @pytest.mark.asyncio + async def test_trigger_succeeds_when_no_active_job(self, service): + mb = _make_mb() + mbs = _make_session(memory_base_id=mb.id) + mock_db = AsyncMock() + + mock_job_svc = MagicMock() + mock_job_svc.create_job = AsyncMock() + mock_task_svc = MagicMock() + mock_task_svc.fire_and_forget_task = AsyncMock() + + with ( + patch("langflow.services.memory_base.ingestion.session_scope", self._fake_scope(mock_db)), + patch.object(service, "get_memory_base_or_404", AsyncMock(return_value=mb)), + patch.object(service, "_get_or_create_session", AsyncMock(return_value=mbs)), + patch( + "langflow.services.memory_base.ingestion._get_latest_pending_workflow_job_id", + AsyncMock(return_value=uuid.uuid4()), + ), + patch("langflow.services.memory_base.ingestion.resolve_kb_username", AsyncMock(return_value="testuser")), + patch( + "langflow.services.memory_base.ingestion.resolve_embedding", + return_value=("OpenAI", "text-embedding-3-small"), + ), + patch("langflow.services.memory_base.ingestion.get_job_service", return_value=mock_job_svc), + patch("langflow.services.memory_base.ingestion.get_task_service", return_value=mock_task_svc), + ): + job_id = await service.trigger_ingestion(mb.id, mb.user_id, "sess-1") + + assert isinstance(job_id, str) + mock_job_svc.create_job.assert_awaited_once() + mock_task_svc.fire_and_forget_task.assert_awaited_once() + + +class TestMemoryBaseServiceThreshold: + """Threshold update should NOT immediately re-evaluate pending count.""" + + @pytest.fixture + def service(self): + from langflow.services.memory_base.service import MemoryBaseService + + return MemoryBaseService() + + @pytest.mark.asyncio + async def test_threshold_update_does_not_trigger_ingestion(self, service): + """Updating threshold via PATCH should never fire a task.""" + mb_updated = _make_mb(threshold=5) + + with patch.object(service, "update", AsyncMock(return_value=mb_updated)): + result = await service.update(mb_updated.id, mb_updated.user_id, MemoryBaseUpdate(threshold=5)) + + assert result.threshold == 5 + # No ingestion task should have been triggered as a side effect + + +class TestMemoryBaseServiceMismatch: + @pytest.fixture + def service(self): + from langflow.services.memory_base.service import MemoryBaseService + + return MemoryBaseService() + + @pytest.mark.asyncio + async def test_mismatch_detected_when_processed_but_empty_store(self, service, tmp_path): + mb = _make_mb() + + with ( + patch.object(service, "get_memory_base_or_404", AsyncMock(return_value=mb)), + patch( + "langflow.services.memory_base.ingestion.resolve_kb_username_by_user_id", + AsyncMock(return_value="testuser"), + ), + patch("langflow.services.memory_base.ingestion.session_scope") as mock_scope, + patch("langflow.services.memory_base.ingestion.KBStorageHelper.get_root_path", return_value=tmp_path), + patch( + "langflow.services.memory_base.ingestion.KBAnalysisHelper.get_metadata", + return_value={"chunks": 0}, + ), + ): + # Simulate session_scope returns total_processed=10 + mock_db = AsyncMock() + mock_db.exec = AsyncMock(return_value=MagicMock(first=MagicMock(return_value=10))) + + class FakeCtx: + async def __aenter__(self): + return mock_db + + async def __aexit__(self, *a): + pass + + mock_scope.return_value = FakeCtx() + + # Create KB path dir so path.exists() is True + kb_path = tmp_path / "testuser" / mb.kb_name + kb_path.mkdir(parents=True) + + result = await service.check_mismatch(mb.id, mb.user_id) + + assert result is True + + async def test_no_mismatch_when_nothing_processed(self, service): + mb = _make_mb() + + with ( + patch.object(service, "get_memory_base_or_404", AsyncMock(return_value=mb)), + patch("langflow.services.memory_base.ingestion.session_scope") as mock_scope, + ): + mock_db = AsyncMock() + mock_db.exec = AsyncMock(return_value=MagicMock(first=MagicMock(return_value=0))) + + class FakeCtx: + async def __aenter__(self): + return mock_db + + async def __aexit__(self, *a): + pass + + mock_scope.return_value = FakeCtx() + + result = await service.check_mismatch(mb.id, mb.user_id) + + assert result is False + + +class TestMemoryBaseServiceRegenerate: + @pytest.fixture + def service(self): + from langflow.services.memory_base.service import MemoryBaseService + + return MemoryBaseService() + + @pytest.mark.asyncio + async def test_regenerate_resets_cursors_and_triggers(self, service): + mb = _make_mb() + mbs1 = _make_session(memory_base_id=mb.id, session_id="s1", cursor_id=uuid.uuid4()) + mbs2 = _make_session(memory_base_id=mb.id, session_id="s2", cursor_id=uuid.uuid4()) + + triggered_sessions: list[str] = [] + + async def fake_trigger(_mb_id, _user_id, session_id): + triggered_sessions.append(session_id) + return str(uuid.uuid4()) + + with ( + patch("langflow.services.memory_base.ingestion.session_scope") as mock_scope, + patch.object(service, "trigger_ingestion", side_effect=fake_trigger), + ): + mock_db = AsyncMock() + mock_mb_result = MagicMock() + mock_mb_result.first = MagicMock(return_value=mb) + mock_session_result = MagicMock() + mock_session_result.all = MagicMock(return_value=[mbs1, mbs2]) + mock_db.exec = AsyncMock(side_effect=[mock_mb_result, mock_session_result, MagicMock()]) + mock_db.add = MagicMock() + mock_db.commit = AsyncMock() + + class FakeCtx: + async def __aenter__(self): + return mock_db + + async def __aexit__(self, *a): + pass + + mock_scope.return_value = FakeCtx() + + job_ids = await service.regenerate(mb.id, mb.user_id) + + assert len(job_ids) == 2 + assert set(triggered_sessions) == {"s1", "s2"} + # Verify cursors were reset + assert mbs1.cursor_id is None + assert mbs2.cursor_id is None + + +# ------------------------------------------------------------------ # +# Task tests # +# ------------------------------------------------------------------ # + + +class TestIngestMemoryTask: + async def test_no_op_when_no_pending_messages(self, tmp_path): + from langflow.services.memory_base.task import IngestionRequest, ingest_memory_task + + job_service = MagicMock() + job_id = uuid.uuid4() + + with ( + patch( + "langflow.services.memory_base.task._acquire_session_lock", + AsyncMock(return_value=asyncio.Lock()), + ), + patch("langflow.services.memory_base.task._release_session_lock", AsyncMock()), + patch("langflow.services.memory_base.task._read_live_cursor", AsyncMock(return_value=None)), + patch( + "langflow.services.memory_base.task._fetch_pending_messages", + AsyncMock(return_value=[]), + ), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path / "kb", + ), + ): + result = await ingest_memory_task( + request=IngestionRequest( + memory_base_id=uuid.uuid4(), + session_id="s1", + flow_id=uuid.uuid4(), + kb_name="kb", + kb_username="user", + user_id=uuid.uuid4(), + embedding_provider="OpenAI", + embedding_model="text-embedding-3-small", + cursor_id=None, + task_job_id=job_id, + job_service=job_service, + ), + ) + + assert result["ingested"] == 0 + + @pytest.mark.asyncio + async def test_cursor_not_advanced_on_ingestion_failure(self, tmp_path): + """Critical: cursor_id must stay unchanged if ingestion fails.""" + from langflow.services.memory_base.task import IngestionRequest, ingest_memory_task + + flow_id = uuid.uuid4() + mb_id = uuid.uuid4() + old_cursor = uuid.uuid4() + + msg = _make_message(flow_id=flow_id, session_id="s1") + job_service = MagicMock() + job_id = uuid.uuid4() + + advance_cursor_called = False + + async def fake_advance_cursor(_db, **_kwargs): + nonlocal advance_cursor_called + advance_cursor_called = True + + with ( + patch( + "langflow.services.memory_base.task._acquire_session_lock", + AsyncMock(return_value=asyncio.Lock()), + ), + patch("langflow.services.memory_base.task._release_session_lock", AsyncMock()), + patch("langflow.services.memory_base.task._read_live_cursor", AsyncMock(return_value=None)), + patch( + "langflow.services.memory_base.task._fetch_pending_messages", + AsyncMock(return_value=[msg]), + ), + patch( + "langflow.services.memory_base.task.build_documents_from_messages", + return_value=[MagicMock()], + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.is_job_cancelled", + AsyncMock(return_value=False), + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.build_embeddings", + AsyncMock(return_value=MagicMock()), + ), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_fresh_chroma_client", + return_value=MagicMock(), + ), + patch("langflow.services.memory_base.task.Chroma"), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.write_documents_to_chroma", + AsyncMock(side_effect=RuntimeError("Chroma exploded")), + ), + patch( + "langflow.services.memory_base.task._advance_cursor", + side_effect=fake_advance_cursor, + ), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path / "kb", + ), + patch("langflow.services.memory_base.task.KBStorageHelper.release_chroma_resources"), + pytest.raises(RuntimeError, match="Chroma exploded"), + ): + await ingest_memory_task( + request=IngestionRequest( + memory_base_id=mb_id, + session_id="s1", + flow_id=flow_id, + kb_name="kb", + kb_username="user", + user_id=uuid.uuid4(), + embedding_provider="OpenAI", + embedding_model="text-embedding-3-small", + cursor_id=old_cursor, + task_job_id=job_id, + job_service=job_service, + ), + ) + + # Cursor must NOT have been advanced + assert not advance_cursor_called, "cursor_id must not advance when ingestion fails" + + @pytest.mark.asyncio + async def test_metadata_synced_on_success(self, tmp_path): + """embedding_metadata.json must be updated after a successful ingestion.""" + from langflow.services.memory_base.task import IngestionRequest, ingest_memory_task + + flow_id = uuid.uuid4() + msg = _make_message(flow_id=flow_id, session_id="s1") + job_service = MagicMock() + job_id = uuid.uuid4() + + sync_called_with: dict = {} + + def fake_sync_kb_metadata(*, kb_path, chroma): + sync_called_with["kb_path"] = kb_path + sync_called_with["chroma"] = chroma + + with ( + patch( + "langflow.services.memory_base.task._acquire_session_lock", + AsyncMock(return_value=asyncio.Lock()), + ), + patch("langflow.services.memory_base.task._release_session_lock", AsyncMock()), + patch("langflow.services.memory_base.task._read_live_cursor", AsyncMock(return_value=None)), + patch( + "langflow.services.memory_base.task._fetch_pending_messages", + AsyncMock(return_value=[msg]), + ), + patch( + "langflow.services.memory_base.task.build_documents_from_messages", + return_value=[MagicMock()], + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.is_job_cancelled", + AsyncMock(return_value=False), + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.build_embeddings", + AsyncMock(return_value=MagicMock()), + ), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_fresh_chroma_client", + return_value=MagicMock(), + ), + patch("langflow.services.memory_base.task.Chroma"), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.write_documents_to_chroma", + AsyncMock(return_value=1), + ), + patch("langflow.services.memory_base.task.sync_kb_metadata", side_effect=fake_sync_kb_metadata), + patch("langflow.services.memory_base.task._mark_messages_ingested", AsyncMock()), + patch("langflow.services.memory_base.task._advance_cursor", AsyncMock()), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path / "kb", + ), + patch("langflow.services.memory_base.task.KBStorageHelper.release_chroma_resources"), + ): + await ingest_memory_task( + request=IngestionRequest( + memory_base_id=uuid.uuid4(), + session_id="s1", + flow_id=flow_id, + kb_name="kb", + kb_username="user", + user_id=uuid.uuid4(), + embedding_provider="OpenAI", + embedding_model="text-embedding-3-small", + cursor_id=None, + task_job_id=job_id, + job_service=job_service, + ), + ) + + assert "kb_path" in sync_called_with, "sync_kb_metadata was not called on success" + + @pytest.mark.asyncio + async def test_metadata_not_synced_when_cancelled(self, tmp_path): + """embedding_metadata.json must NOT be updated when ingestion is cancelled.""" + from langflow.services.memory_base.task import IngestionRequest, ingest_memory_task + + flow_id = uuid.uuid4() + msg = _make_message(flow_id=flow_id, session_id="s1") + job_service = MagicMock() + job_id = uuid.uuid4() + sync_called = False + + def fake_sync(*_args, **_kwargs): + nonlocal sync_called + sync_called = True + + with ( + patch( + "langflow.services.memory_base.task._acquire_session_lock", + AsyncMock(return_value=asyncio.Lock()), + ), + patch("langflow.services.memory_base.task._release_session_lock", AsyncMock()), + patch("langflow.services.memory_base.task._read_live_cursor", AsyncMock(return_value=None)), + patch( + "langflow.services.memory_base.task._fetch_pending_messages", + AsyncMock(return_value=[msg]), + ), + patch( + "langflow.services.memory_base.task.build_documents_from_messages", + return_value=[MagicMock()], + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.is_job_cancelled", + AsyncMock(return_value=False), + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.build_embeddings", + AsyncMock(return_value=MagicMock()), + ), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_fresh_chroma_client", + return_value=MagicMock(), + ), + patch("langflow.services.memory_base.task.Chroma"), + # write_documents_to_chroma returns fewer docs than sent → cancelled + patch( + "langflow.services.memory_base.task.KBIngestionHelper.write_documents_to_chroma", + AsyncMock(return_value=0), + ), + patch("langflow.services.memory_base.task.sync_kb_metadata", side_effect=fake_sync), + patch("langflow.services.memory_base.task._advance_cursor", AsyncMock()), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path / "kb", + ), + patch("langflow.services.memory_base.task.KBStorageHelper.release_chroma_resources"), + ): + result = await ingest_memory_task( + request=IngestionRequest( + memory_base_id=uuid.uuid4(), + session_id="s1", + flow_id=flow_id, + kb_name="kb", + kb_username="user", + user_id=uuid.uuid4(), + embedding_provider="OpenAI", + embedding_model="text-embedding-3-small", + cursor_id=None, + task_job_id=job_id, + job_service=job_service, + ), + ) + + assert not sync_called, "sync_kb_metadata must not be called when ingestion is cancelled" + assert "cancelled" in result["message"].lower() + + @pytest.mark.asyncio + async def test_cursor_advanced_on_success(self, tmp_path): + from langflow.services.memory_base.task import IngestionRequest, ingest_memory_task + + flow_id = uuid.uuid4() + mb_id = uuid.uuid4() + + msg = _make_message(flow_id=flow_id, session_id="s1") + job_service = MagicMock() + job_id = uuid.uuid4() + + advance_kwargs: dict = {} + + async def fake_advance_cursor(_db, **kwargs): + advance_kwargs.update(kwargs) + + with ( + patch( + "langflow.services.memory_base.task._acquire_session_lock", + AsyncMock(return_value=asyncio.Lock()), + ), + patch("langflow.services.memory_base.task._release_session_lock", AsyncMock()), + patch("langflow.services.memory_base.task._read_live_cursor", AsyncMock(return_value=None)), + patch( + "langflow.services.memory_base.task._fetch_pending_messages", + AsyncMock(return_value=[msg]), + ), + patch( + "langflow.services.memory_base.task.build_documents_from_messages", + return_value=[MagicMock()], + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.is_job_cancelled", + AsyncMock(return_value=False), + ), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.build_embeddings", + AsyncMock(return_value=MagicMock()), + ), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_fresh_chroma_client", + return_value=MagicMock(), + ), + patch("langflow.services.memory_base.task.Chroma"), + patch( + "langflow.services.memory_base.task.KBIngestionHelper.write_documents_to_chroma", + AsyncMock(return_value=1), + ), + patch("langflow.services.memory_base.task.sync_kb_metadata"), + patch("langflow.services.memory_base.task._mark_messages_ingested", AsyncMock()), + patch("langflow.services.memory_base.task._advance_cursor", AsyncMock(side_effect=fake_advance_cursor)), + patch( + "langflow.services.memory_base.task.KBStorageHelper.get_root_path", + return_value=tmp_path / "kb", + ), + patch("langflow.services.memory_base.task.KBStorageHelper.release_chroma_resources"), + ): + result = await ingest_memory_task( + request=IngestionRequest( + memory_base_id=mb_id, + session_id="s1", + flow_id=flow_id, + kb_name="kb", + kb_username="user", + user_id=uuid.uuid4(), + embedding_provider="OpenAI", + embedding_model="text-embedding-3-small", + cursor_id=None, + task_job_id=job_id, + job_service=job_service, + ), + ) + + assert result["ingested"] == 1 + assert advance_kwargs["new_cursor_id"] == msg.id + assert advance_kwargs["ingested_count"] == 1 + + def test_sync_kb_metadata_stamps_is_memory_base(self, tmp_path): + """sync_kb_metadata must write is_memory_base: true to the metadata file.""" + import json + + from langflow.services.memory_base.document_builders import sync_kb_metadata as _sync_kb_metadata + + kb_path = tmp_path / "test_kb" + kb_path.mkdir() + + mock_chroma = MagicMock() + + with ( + patch( + "langflow.services.memory_base.document_builders.KBAnalysisHelper.get_metadata", + return_value={"chunks": 0, "embedding_provider": "OpenAI"}, + ), + patch("langflow.services.memory_base.document_builders.KBAnalysisHelper.update_text_metrics"), + patch( + "langflow.services.memory_base.document_builders.KBStorageHelper.get_directory_size", return_value=1024 + ), + ): + _sync_kb_metadata(kb_path=kb_path, chroma=mock_chroma) + + written = json.loads((kb_path / "embedding_metadata.json").read_text()) + assert written["is_memory_base"] is True + assert "memory" in written.get("source_types", []) + + def test_sync_kb_metadata_failure_does_not_raise(self, tmp_path): + """Metadata sync errors must be swallowed so the cursor can still advance.""" + from langflow.services.memory_base.document_builders import sync_kb_metadata as _sync_kb_metadata + + kb_path = tmp_path / "no_such_dir" # does not exist + + with patch( + "langflow.services.memory_base.document_builders.KBAnalysisHelper.get_metadata", + side_effect=OSError("disk full"), + ): + # Must not raise + _sync_kb_metadata(kb_path=kb_path, chroma=MagicMock()) + + def test_build_documents_skips_empty_messages(self): + from langflow.services.memory_base.document_builders import ( + build_documents_from_messages as _build_documents_from_messages, + ) + + flow_id = uuid.uuid4() + messages = [ + _make_message(flow_id=flow_id, session_id="s1", text=""), + _make_message(flow_id=flow_id, session_id="s1", text=" "), + _make_message(flow_id=flow_id, session_id="s1", text="Valid content here."), + ] + docs = _build_documents_from_messages(messages, session_id="s1", flow_id=str(flow_id)) + assert len(docs) == 1 + assert docs[0].page_content == "Valid content here." + + def test_build_documents_metadata(self): + from langflow.services.memory_base.document_builders import ( + build_documents_from_messages as _build_documents_from_messages, + ) + + flow_id = uuid.uuid4() + run_id = uuid.uuid4() + msg = _make_message(flow_id=flow_id, session_id="s1", text="Test output.", run_id=run_id) + docs = _build_documents_from_messages([msg], session_id="s1", flow_id=str(flow_id)) + assert docs[0].metadata["message_id"] == str(msg.id) + assert docs[0].metadata["run_id"] == str(run_id) + assert docs[0].metadata["session_id"] == "s1" + + +# ------------------------------------------------------------------ # +# on_flow_output hook and threshold-trigger tests # +# ------------------------------------------------------------------ # + + +class TestOnFlowOutputHook: + """Tests for on_flow_output, _maybe_trigger threshold logic, and hook wiring.""" + + @pytest.fixture + def service(self): + from langflow.services.memory_base.service import MemoryBaseService + + return MemoryBaseService() + + def _fake_scope(self, mock_db): + """Return a mock session_scope context manager backed by mock_db.""" + + class _FakeCtx: + async def __aenter__(self): + return mock_db + + async def __aexit__(self, *a): + pass + + scope = MagicMock() + scope.return_value = _FakeCtx() + return scope + + @pytest.mark.asyncio + async def test_on_flow_output_skips_when_below_threshold(self, service): + """No job must be created when pending message count is below the threshold.""" + from langflow.services.memory_base.ingestion import _maybe_trigger + + mb = _make_mb(threshold=5) + mbs = _make_session(memory_base_id=mb.id) + mock_db = AsyncMock() + + with ( + patch("langflow.services.memory_base.ingestion.session_scope", self._fake_scope(mock_db)), + patch.object(service, "_get_or_create_session", AsyncMock(return_value=mbs)), + patch("langflow.services.memory_base.ingestion._insert_workflow_run", AsyncMock()), + patch( + "langflow.services.memory_base.ingestion.count_pending_messages", AsyncMock(return_value=3) + ), # 3 < threshold 5 + patch("langflow.services.memory_base.ingestion.get_job_service") as mock_jsc, + ): + await _maybe_trigger( + mb=mb, session_id="s1", job_id=None, get_or_create_session=service._get_or_create_session + ) + + mock_jsc.return_value.create_job.assert_not_called() + + @pytest.mark.asyncio + async def test_on_flow_output_triggers_when_threshold_met(self, service): + """A job must be created and dispatched when pending message count meets threshold.""" + from langflow.services.memory_base.ingestion import _maybe_trigger + + mb = _make_mb(threshold=3) + mbs = _make_session(memory_base_id=mb.id) + mock_db = AsyncMock() + + mock_job_svc = MagicMock() + mock_job_svc.create_job = AsyncMock() + mock_task_svc = MagicMock() + mock_task_svc.fire_and_forget_task = AsyncMock() + + with ( + patch("langflow.services.memory_base.ingestion.session_scope", self._fake_scope(mock_db)), + patch.object(service, "_get_or_create_session", AsyncMock(return_value=mbs)), + patch("langflow.services.memory_base.ingestion._insert_workflow_run", AsyncMock()), + patch( + "langflow.services.memory_base.ingestion.count_pending_messages", AsyncMock(return_value=5) + ), # 5 >= threshold 3 + patch( + "langflow.services.memory_base.ingestion._get_latest_pending_workflow_job_id", + AsyncMock(return_value=uuid.uuid4()), + ), + patch( + "langflow.services.memory_base.kb_path_helpers.resolve_kb_username", AsyncMock(return_value="testuser") + ), + patch( + "langflow.services.memory_base.ingestion.resolve_embedding", + return_value=("OpenAI", "text-embedding-3-small"), + ), + patch("langflow.services.memory_base.ingestion.get_job_service", return_value=mock_job_svc), + patch("langflow.services.memory_base.ingestion.get_task_service", return_value=mock_task_svc), + ): + await _maybe_trigger( + mb=mb, session_id="s1", job_id=None, get_or_create_session=service._get_or_create_session + ) + + mock_job_svc.create_job.assert_awaited_once() + mock_task_svc.fire_and_forget_task.assert_awaited_once() + + @pytest.mark.asyncio + async def test_on_flow_output_is_silent_on_error(self, service): + """on_flow_output must swallow _maybe_trigger exceptions without propagating them. + + This guarantees memory-base failures never cause regressions in flow execution. + """ + flow_id = uuid.uuid4() + mb = _make_mb(flow_id=flow_id, auto_capture=True) + mock_db = AsyncMock() + result_mock = MagicMock() + result_mock.all = MagicMock(return_value=[mb]) + mock_db.exec = AsyncMock(return_value=result_mock) + + with ( + patch("langflow.services.memory_base.ingestion.session_scope", self._fake_scope(mock_db)), + patch( + "langflow.services.memory_base.ingestion._maybe_trigger", AsyncMock(side_effect=RuntimeError("boom")) + ), + ): + # Must not raise even though _maybe_trigger blows up + await service.on_flow_output(flow_id=flow_id, session_id="s1", job_id=uuid.uuid4()) + + @pytest.mark.asyncio + async def test_hook_wiring_playground(self): + """Playground path: background_tasks.add_task dispatches on_flow_output with correct kwargs. + + Verifies the contract of the hook-dispatch block added to generate_flow_events + in api/build.py after end_all_traces(). + """ + from starlette.background import BackgroundTasks + + flow_id = uuid.uuid4() + run_id = uuid.uuid4() + + mb_service = MagicMock() + mb_service.on_flow_output = AsyncMock() + + bg_tasks = MagicMock(spec=BackgroundTasks) + mock_graph = MagicMock() + mock_graph.run_id = str(run_id) # graph.run_id is always a str + mock_graph.session_id = "test-session" + + with patch("langflow.api.build.get_memory_base_service", return_value=mb_service): + import langflow.api.build as build_module + + # Confirm the import is wired at module level + assert hasattr(build_module, "get_memory_base_service") + + # Execute the same hook-dispatch block as in generate_flow_events + _run_id_uuid = uuid.UUID(mock_graph.run_id) if mock_graph.run_id else None + bg_tasks.add_task( + mb_service.on_flow_output, + flow_id=flow_id, + session_id=mock_graph.session_id or str(flow_id), + run_id=_run_id_uuid, + ) + + bg_tasks.add_task.assert_called_once_with( + mb_service.on_flow_output, + flow_id=flow_id, + session_id="test-session", + run_id=run_id, # UUID, not str — type-cast from graph.run_id + ) + + @pytest.mark.asyncio + async def test_hook_wiring_v2_async_wrapper(self): + """V2 async path: _run_and_notify preserves run_graph_internal result and dispatches hook. + + Verifies the behavioral contract of the closure added to execute_workflow_background + in api/v2/workflow.py: the wrapper must be transparent to execute_with_status + (return value unchanged) while also firing the memory-base hook. + """ + expected_result = (MagicMock(), "effective-session-42") + run_graph_mock = AsyncMock(return_value=expected_result) + hook_mock = AsyncMock() + task_service_mock = MagicMock() + task_service_mock.fire_and_forget_task = AsyncMock() + + hook_flow_id = uuid.uuid4() + hook_run_id = uuid.uuid4() + + # Mirror the _run_and_notify closure from workflow.py execute_workflow_background + async def _run_and_notify(**kwargs): + result = await run_graph_mock(**kwargs) + _, _effective_session_id = result + with contextlib.suppress(Exception): + await task_service_mock.fire_and_forget_task( + hook_mock, + flow_id=hook_flow_id, + session_id=_effective_session_id, + run_id=hook_run_id, + ) + return result + + result = await _run_and_notify(graph=MagicMock()) + + # Return value must be identical — execute_with_status depends on this + assert result == expected_result + # Hook must be dispatched with the session_id extracted from run_graph_internal + task_service_mock.fire_and_forget_task.assert_awaited_once_with( + hook_mock, + flow_id=hook_flow_id, + session_id="effective-session-42", + run_id=hook_run_id, + ) + + @pytest.mark.asyncio + async def test_hook_failure_does_not_affect_wrapper_return(self): + """If fire_and_forget_task raises inside _run_and_notify, the return value is still correct.""" + expected_result = (MagicMock(), "some-session") + run_graph_mock = AsyncMock(return_value=expected_result) + task_service_mock = MagicMock() + task_service_mock.fire_and_forget_task = AsyncMock(side_effect=RuntimeError("dispatch failed")) + + async def _run_and_notify(**kwargs): + result = await run_graph_mock(**kwargs) + _, _effective_session_id = result + with contextlib.suppress(Exception): + await task_service_mock.fire_and_forget_task( + AsyncMock(), + flow_id=uuid.uuid4(), + session_id=_effective_session_id, + run_id=uuid.uuid4(), + ) + return result + + result = await _run_and_notify(graph=MagicMock()) + assert result == expected_result + + +# ------------------------------------------------------------------ # +# API endpoint routing tests # +# ------------------------------------------------------------------ # + + +class TestMemoriesAPIRouting: + """Verify routing and response codes without hitting the DB.""" + + @pytest.fixture + def patched_service(self): + """Patch get_memory_base_service in memories.py.""" + mock_svc = MagicMock() + with patch("langflow.api.v1.memories.get_memory_base_service", return_value=mock_svc): + yield mock_svc + + @pytest.mark.asyncio + async def test_get_not_found_returns_404(self, patched_service): + """Handler returns 404 when service.get returns None (covered via direct call).""" + from fastapi import HTTPException + from langflow.api.v1.memories import get_memory_base + + patched_service.get = AsyncMock(return_value=None) + + mock_user = MagicMock() + mock_user.id = uuid.uuid4() + + with pytest.raises(HTTPException) as exc_info: + await get_memory_base(memory_base_id=uuid.uuid4(), current_user=mock_user) + + assert exc_info.value.status_code == 404 + + @pytest.mark.asyncio + async def test_flush_conflict_returns_409(self, patched_service): + """trigger_ingestion raising RuntimeError should map to HTTP 409.""" + from langflow.api.v1.memories import flush_memory_base + + patched_service.trigger_ingestion = AsyncMock(side_effect=RuntimeError("already in progress")) + + # We call the handler directly to test the error mapping + mock_user = MagicMock() + mock_user.id = uuid.uuid4() + + from fastapi import HTTPException + from langflow.api.v1.memories import FlushRequest + + with pytest.raises(HTTPException) as exc_info: + await flush_memory_base( + memory_base_id=uuid.uuid4(), + body=FlushRequest(session_id="s1"), + current_user=mock_user, + ) + from fastapi import HTTPException + + assert isinstance(exc_info.value, HTTPException) + assert exc_info.value.status_code == 409 + + +# ------------------------------------------------------------------ # +# API handler unit tests (direct invocation, no HTTP stack) # +# ------------------------------------------------------------------ # + + +class TestMemoriesAPIHandlers: + """Call endpoint handlers directly, mocking _service, to test all status-code branches.""" + + @pytest.fixture + def mock_user(self): + user = MagicMock() + user.id = uuid.uuid4() + return user + + # ---------------------------------------------------------------- # + # create_memory_base # + # ---------------------------------------------------------------- # + + @pytest.mark.asyncio + async def test_create_success_returns_memory_base_read(self, mock_user): + from langflow.api.v1.memories import create_memory_base + + mb = _make_mb(user_id=mock_user.id) + payload = MemoryBaseCreate(name="mb", flow_id=mb.flow_id, user_id=mock_user.id, kb_name="kb") + + svc = MagicMock() + svc.create = AsyncMock(return_value=mb) + with patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc): + result = await create_memory_base(current_user=mock_user, payload=payload) + + assert result.id == mb.id + assert result.user_id == mock_user.id + + @pytest.mark.asyncio + async def test_create_duplicate_name_returns_409(self, mock_user): + from fastapi import HTTPException + from langflow.api.v1.memories import create_memory_base + + payload = MemoryBaseCreate(name="dup", flow_id=uuid.uuid4(), user_id=mock_user.id, kb_name="kb") + + svc = MagicMock() + svc.create = AsyncMock(side_effect=ValueError("name already in use")) + with ( + patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc), + pytest.raises(HTTPException) as exc_info, + ): + await create_memory_base(current_user=mock_user, payload=payload) + + assert exc_info.value.status_code == 409 + + # ---------------------------------------------------------------- # + # get_memory_base # + # ---------------------------------------------------------------- # + + @pytest.mark.asyncio + async def test_get_success_returns_memory_base_read(self, mock_user): + from langflow.api.v1.memories import get_memory_base + + mb = _make_mb(user_id=mock_user.id) + + svc = MagicMock() + svc.get = AsyncMock(return_value=mb) + with patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc): + result = await get_memory_base(memory_base_id=mb.id, current_user=mock_user) + + assert result.id == mb.id + + @pytest.mark.asyncio + async def test_get_not_found_raises_404(self, mock_user): + from fastapi import HTTPException + from langflow.api.v1.memories import get_memory_base + + svc = MagicMock() + svc.get = AsyncMock(return_value=None) + with ( + patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc), + pytest.raises(HTTPException) as exc_info, + ): + await get_memory_base(memory_base_id=uuid.uuid4(), current_user=mock_user) + + assert exc_info.value.status_code == 404 + + # ---------------------------------------------------------------- # + # update_memory_base # + # ---------------------------------------------------------------- # + + @pytest.mark.asyncio + async def test_update_success_returns_updated_record(self, mock_user): + from langflow.api.v1.memories import update_memory_base + + mb = _make_mb(user_id=mock_user.id, threshold=99) + + svc = MagicMock() + svc.update = AsyncMock(return_value=mb) + with patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc): + result = await update_memory_base( + memory_base_id=mb.id, + current_user=mock_user, + patch=MemoryBaseUpdate(threshold=99), + ) + + assert result.threshold == 99 + + @pytest.mark.asyncio + async def test_update_not_found_raises_404(self, mock_user): + from fastapi import HTTPException + from langflow.api.v1.memories import update_memory_base + + svc = MagicMock() + svc.update = AsyncMock(return_value=None) + with ( + patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc), + pytest.raises(HTTPException) as exc_info, + ): + await update_memory_base( + memory_base_id=uuid.uuid4(), + current_user=mock_user, + patch=MemoryBaseUpdate(threshold=5), + ) + + assert exc_info.value.status_code == 404 + + # ---------------------------------------------------------------- # + # delete_memory_base # + # ---------------------------------------------------------------- # + + @pytest.mark.asyncio + async def test_delete_success_returns_none(self, mock_user): + from langflow.api.v1.memories import delete_memory_base + + svc = MagicMock() + svc.delete = AsyncMock(return_value=True) + with patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc): + result = await delete_memory_base(memory_base_id=uuid.uuid4(), current_user=mock_user) + + assert result is None + + @pytest.mark.asyncio + async def test_delete_not_found_raises_404(self, mock_user): + from fastapi import HTTPException + from langflow.api.v1.memories import delete_memory_base + + svc = MagicMock() + svc.delete = AsyncMock(return_value=False) + with ( + patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc), + pytest.raises(HTTPException) as exc_info, + ): + await delete_memory_base(memory_base_id=uuid.uuid4(), current_user=mock_user) + + assert exc_info.value.status_code == 404 + + # ---------------------------------------------------------------- # + # flush_memory_base # + # ---------------------------------------------------------------- # + + @pytest.mark.asyncio + async def test_flush_success_returns_job_id(self, mock_user): + from langflow.api.v1.memories import FlushRequest, flush_memory_base + + job_id = str(uuid.uuid4()) + + svc = MagicMock() + svc.trigger_ingestion = AsyncMock(return_value=job_id) + with patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc): + result = await flush_memory_base( + memory_base_id=uuid.uuid4(), + current_user=mock_user, + body=FlushRequest(session_id="s1"), + ) + + assert result == {"job_id": job_id} + + @pytest.mark.asyncio + async def test_flush_value_error_raises_404(self, mock_user): + from fastapi import HTTPException + from langflow.api.v1.memories import FlushRequest, flush_memory_base + + svc = MagicMock() + svc.trigger_ingestion = AsyncMock(side_effect=ValueError("memory base not found")) + with ( + patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc), + pytest.raises(HTTPException) as exc_info, + ): + await flush_memory_base( + memory_base_id=uuid.uuid4(), + current_user=mock_user, + body=FlushRequest(session_id="s1"), + ) + + assert exc_info.value.status_code == 404 + + @pytest.mark.asyncio + async def test_flush_duplicate_job_error_raises_409(self, mock_user): + from fastapi import HTTPException + from langflow.api.v1.memories import FlushRequest, flush_memory_base + from langflow.services.jobs import DuplicateJobError + + svc = MagicMock() + svc.trigger_ingestion = AsyncMock(side_effect=DuplicateJobError("already running")) + with ( + patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc), + pytest.raises(HTTPException) as exc_info, + ): + await flush_memory_base( + memory_base_id=uuid.uuid4(), + current_user=mock_user, + body=FlushRequest(session_id="s1"), + ) + + assert exc_info.value.status_code == 409 + + # ---------------------------------------------------------------- # + # check_mismatch # + # ---------------------------------------------------------------- # + + @pytest.mark.asyncio + async def test_check_mismatch_detected_returns_true(self, mock_user): + from langflow.api.v1.memories import check_mismatch + + svc = MagicMock() + svc.check_mismatch = AsyncMock(return_value=True) + with patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc): + result = await check_mismatch(memory_base_id=uuid.uuid4(), current_user=mock_user) + + assert result.mismatch_detected is True + + @pytest.mark.asyncio + async def test_check_mismatch_not_detected_returns_false(self, mock_user): + from langflow.api.v1.memories import check_mismatch + + svc = MagicMock() + svc.check_mismatch = AsyncMock(return_value=False) + with patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc): + result = await check_mismatch(memory_base_id=uuid.uuid4(), current_user=mock_user) + + assert result.mismatch_detected is False + + @pytest.mark.asyncio + async def test_check_mismatch_not_found_raises_404(self, mock_user): + from fastapi import HTTPException + from langflow.api.v1.memories import check_mismatch + + svc = MagicMock() + svc.check_mismatch = AsyncMock(side_effect=ValueError("not found")) + with ( + patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc), + pytest.raises(HTTPException) as exc_info, + ): + await check_mismatch(memory_base_id=uuid.uuid4(), current_user=mock_user) + + assert exc_info.value.status_code == 404 + + # ---------------------------------------------------------------- # + # regenerate_memory_base # + # ---------------------------------------------------------------- # + + @pytest.mark.asyncio + async def test_regenerate_success_returns_job_ids(self, mock_user): + from langflow.api.v1.memories import regenerate_memory_base + + job_ids = [str(uuid.uuid4()), str(uuid.uuid4())] + + svc = MagicMock() + svc.regenerate = AsyncMock(return_value=job_ids) + with patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc): + result = await regenerate_memory_base(memory_base_id=uuid.uuid4(), current_user=mock_user) + + assert result.job_ids == job_ids + + @pytest.mark.asyncio + async def test_regenerate_not_found_raises_404(self, mock_user): + from fastapi import HTTPException + from langflow.api.v1.memories import regenerate_memory_base + + svc = MagicMock() + svc.regenerate = AsyncMock(side_effect=ValueError("not found")) + with ( + patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc), + pytest.raises(HTTPException) as exc_info, + ): + await regenerate_memory_base(memory_base_id=uuid.uuid4(), current_user=mock_user) + + assert exc_info.value.status_code == 404 + + # ---------------------------------------------------------------- # + # list_sessions # + # ---------------------------------------------------------------- # + + @pytest.mark.asyncio + async def test_list_sessions_ownership_failure_raises_404(self, mock_user): + from fastapi import HTTPException + from fastapi_pagination import Params + from langflow.api.v1.memories import list_sessions + + mock_db = AsyncMock() + + class FakeCtx: + async def __aenter__(self): + return mock_db + + async def __aexit__(self, *a): + pass + + svc = MagicMock() + svc.get_memory_base_or_404 = AsyncMock(side_effect=ValueError("not found")) + with ( + patch("langflow.api.v1.memories.session_scope", return_value=FakeCtx()), + patch("langflow.api.v1.memories.get_memory_base_service", return_value=svc), + pytest.raises(HTTPException) as exc_info, + ): + await list_sessions( + memory_base_id=uuid.uuid4(), + current_user=mock_user, + params=Params(), + ) + + assert exc_info.value.status_code == 404 + + # ---------------------------------------------------------------- # + # list_session_messages # + # ---------------------------------------------------------------- # + + @pytest.mark.asyncio + async def test_list_session_messages_not_found_raises_404(self, mock_user): + from fastapi import HTTPException + from fastapi_pagination import Params + from langflow.api.v1.memories import list_session_messages + + mock_db = AsyncMock() + result_mock = MagicMock() + result_mock.first = MagicMock(return_value=None) + mock_db.exec = AsyncMock(return_value=result_mock) + + class FakeCtx: + async def __aenter__(self): + return mock_db + + async def __aexit__(self, *a): + pass + + with ( + patch("langflow.api.v1.memories.session_scope", return_value=FakeCtx()), + pytest.raises(HTTPException) as exc_info, + ): + await list_session_messages( + memory_base_id=uuid.uuid4(), + session_id="s1", + current_user=mock_user, + params=Params(), + ) + + assert exc_info.value.status_code == 404 + + # ---------------------------------------------------------------- # + # MessageReadResponse schema # + # ---------------------------------------------------------------- # + + def test_message_read_response_from_attributes(self): + from langflow.api.v1.memories import MessageReadResponse + + msg = _make_message(flow_id=uuid.uuid4(), session_id="s1", text="hello") + response = MessageReadResponse.model_validate(msg, from_attributes=True) + + assert response.text == "hello" + assert response.session_id == "s1" + assert response.sender == "AI" + assert response.content_blocks == [] + + def test_flush_request_schema(self): + from langflow.api.v1.memories import FlushRequest + + req = FlushRequest(session_id="my-session") + assert req.session_id == "my-session" + + def test_mismatch_response_schema(self): + from langflow.api.v1.memories import MismatchResponse + + assert MismatchResponse(mismatch_detected=True).mismatch_detected is True + assert MismatchResponse(mismatch_detected=False).mismatch_detected is False + + def test_regenerate_response_schema(self): + from langflow.api.v1.memories import RegenerateResponse + + ids = ["a", "b", "c"] + assert RegenerateResponse(job_ids=ids).job_ids == ids + + +# ------------------------------------------------------------------ # +# Security adversarial tests # +# ------------------------------------------------------------------ # + + +class TestMemoryBaseSecurityAdversarial: + """Pin authorization invariants that the rest of the code depends on.""" + + @pytest.fixture + def service(self): + from langflow.services.memory_base.service import MemoryBaseService + + return MemoryBaseService() + + def _fake_scope(self, mock_db): + class _FakeCtx: + async def __aenter__(self): + return mock_db + + async def __aexit__(self, *a): + pass + + scope = MagicMock() + scope.return_value = _FakeCtx() + return scope + + @pytest.mark.asyncio + async def test_update_rejects_when_flow_changed_to_unowned_flow(self, service): + user_a = uuid.uuid4() + user_b = uuid.uuid4() + mb = _make_mb(user_id=user_a) + mock_db = AsyncMock() + exec_result = MagicMock() + exec_result.first.return_value = None + mock_db.exec = AsyncMock(return_value=exec_result) + with patch("langflow.services.memory_base.service.session_scope", self._fake_scope(mock_db)): + result = await service.update(mb.id, user_b, MemoryBaseUpdate(threshold=5)) + assert result is None + + @pytest.mark.asyncio + async def test_on_flow_output_refuses_cross_user_flow(self, service): + unrelated_flow_id = uuid.uuid4() + mock_db = AsyncMock() + exec_result = MagicMock() + exec_result.all.return_value = [] + mock_db.exec = AsyncMock(return_value=exec_result) + with patch("langflow.services.memory_base.ingestion.session_scope", self._fake_scope(mock_db)): + await service.on_flow_output(flow_id=unrelated_flow_id, session_id="sess-1", job_id=uuid.uuid4()) + mock_db.exec.assert_called_once() + + @pytest.mark.asyncio + async def test_regenerate_rejects_unowned_memory_base(self, service): + user_b = uuid.uuid4() + mb_id = uuid.uuid4() + mock_db = AsyncMock() + exec_result = MagicMock() + exec_result.first.return_value = None + mock_db.exec = AsyncMock(return_value=exec_result) + with ( + patch("langflow.services.memory_base.ingestion.session_scope", self._fake_scope(mock_db)), + pytest.raises(ValueError, match="not found"), + ): + await service.regenerate(mb_id, user_b) + + @pytest.mark.asyncio + async def test_check_mismatch_rejects_unowned_memory_base(self, service): + user_b = uuid.uuid4() + mb_id = uuid.uuid4() + mock_db = AsyncMock() + exec_result = MagicMock() + exec_result.first.return_value = None + mock_db.exec = AsyncMock(return_value=exec_result) + with ( + patch("langflow.services.memory_base.ingestion.session_scope", self._fake_scope(mock_db)), + pytest.raises(ValueError, match="not found"), + ): + await service.check_mismatch(mb_id, user_b) + + @pytest.mark.asyncio + async def test_sessions_stmt_enforces_user_id(self, service): + user_id = uuid.uuid4() + mb_id = uuid.uuid4() + stmt = service.sessions_stmt(mb_id, user_id) + compiled = str(stmt) + assert "user_id" in compiled + assert "memory_base" in compiled.lower() diff --git a/src/backend/tests/unit/test_session_metadata.py b/src/backend/tests/unit/test_session_metadata.py index a08cdc8b23..f245e6a0b5 100644 --- a/src/backend/tests/unit/test_session_metadata.py +++ b/src/backend/tests/unit/test_session_metadata.py @@ -322,3 +322,15 @@ async def test_session_metadata_retrieval(): messages = await aget_messages(sender="User", session_id=session_id) assert len(messages) == 1 assert messages[0].session_metadata == initial_metadata + + +@pytest.mark.usefixtures("client") +async def test_messageupdate_with_session_metadata(sample_session_metadata): + """Test MessageUpdate schema with session_metadata.""" + from langflow.services.database.models.message.model import MessageUpdate + + message_update = MessageUpdate( + text="Updated text", + session_metadata=sample_session_metadata, + ) + assert message_update.session_metadata == sample_session_metadata diff --git a/src/frontend/src/controllers/API/queries/deployment-provider-accounts/use-delete-provider-account.ts b/src/frontend/src/controllers/API/queries/deployment-provider-accounts/use-delete-provider-account.ts index e23d33c527..b0741fedcd 100644 --- a/src/frontend/src/controllers/API/queries/deployment-provider-accounts/use-delete-provider-account.ts +++ b/src/frontend/src/controllers/API/queries/deployment-provider-accounts/use-delete-provider-account.ts @@ -19,10 +19,8 @@ export const useDeleteProviderAccount: useMutationFunctionType< ); }; - // TODO: Add retries for transient server-side errors (5xx, timeouts). return mutate(["useDeleteProviderAccount"], fn, { ...options, - retry: false, onSuccess: (...args) => { queryClient.refetchQueries({ queryKey: ["useGetProviderAccounts"] }); options?.onSuccess?.(...args); diff --git a/src/frontend/src/controllers/API/queries/deployments/use-delete-deployment.ts b/src/frontend/src/controllers/API/queries/deployments/use-delete-deployment.ts index a7a679eeb0..598b137572 100644 --- a/src/frontend/src/controllers/API/queries/deployments/use-delete-deployment.ts +++ b/src/frontend/src/controllers/API/queries/deployments/use-delete-deployment.ts @@ -17,10 +17,8 @@ export const useDeleteDeployment: useMutationFunctionType< await api.delete(`${getURL("DEPLOYMENTS")}/${deployment_id}`); }; - // TODO: Add retries for transient server-side errors (5xx, timeouts). return mutate(["useDeleteDeployment"], fn, { ...options, - retry: false, onSuccess: (...args) => { queryClient.refetchQueries({ queryKey: ["useGetDeployments"] }); options?.onSuccess?.(...args); diff --git a/src/frontend/src/controllers/API/queries/deployments/use-patch-deployment.ts b/src/frontend/src/controllers/API/queries/deployments/use-patch-deployment.ts index 968e7c9c1c..586dc86c94 100644 --- a/src/frontend/src/controllers/API/queries/deployments/use-patch-deployment.ts +++ b/src/frontend/src/controllers/API/queries/deployments/use-patch-deployment.ts @@ -50,10 +50,8 @@ export const usePatchDeployment: useMutationFunctionType< return res.data; }; - // TODO: Add retries for transient server-side errors (5xx, timeouts). return mutate(["usePatchDeployment"], fn, { ...options, - retry: false, onSuccess: (...args) => { queryClient.refetchQueries({ queryKey: ["useGetDeployments"] }); queryClient.removeQueries({ diff --git a/src/frontend/src/controllers/API/queries/deployments/use-patch-snapshot.ts b/src/frontend/src/controllers/API/queries/deployments/use-patch-snapshot.ts index 22a024797c..55810f7a04 100644 --- a/src/frontend/src/controllers/API/queries/deployments/use-patch-snapshot.ts +++ b/src/frontend/src/controllers/API/queries/deployments/use-patch-snapshot.ts @@ -26,10 +26,8 @@ export const usePatchSnapshot: useMutationFunctionType< return data; }; - // TODO: Add retries for transient server-side errors (5xx, timeouts). return mutate(["usePatchSnapshot"], fn, { ...options, - retry: false, onSuccess: (...args) => { queryClient.refetchQueries({ queryKey: ["useGetDeployments"] }); options?.onSuccess?.(...args); diff --git a/src/frontend/src/controllers/API/queries/deployments/use-post-deployment-run.ts b/src/frontend/src/controllers/API/queries/deployments/use-post-deployment-run.ts index f26422407c..f11852967c 100644 --- a/src/frontend/src/controllers/API/queries/deployments/use-post-deployment-run.ts +++ b/src/frontend/src/controllers/API/queries/deployments/use-post-deployment-run.ts @@ -47,6 +47,5 @@ export const usePostDeploymentRun: useMutationFunctionType< return res.data; }; - // TODO: Add retries for transient server-side errors (5xx, timeouts). - return mutate(["usePostDeploymentRun"], fn, { ...options, retry: false }); + return mutate(["usePostDeploymentRun"], fn, options); }; diff --git a/src/frontend/src/controllers/API/queries/deployments/use-post-deployment.ts b/src/frontend/src/controllers/API/queries/deployments/use-post-deployment.ts index 52e2fed23b..ef32797c08 100644 --- a/src/frontend/src/controllers/API/queries/deployments/use-post-deployment.ts +++ b/src/frontend/src/controllers/API/queries/deployments/use-post-deployment.ts @@ -53,10 +53,8 @@ export const usePostDeployment: useMutationFunctionType< return res.data; }; - // TODO: Add retries for transient server-side errors (5xx, timeouts). return mutate(["usePostDeployment"], fn, { ...options, - retry: false, onSuccess: () => { return queryClient.refetchQueries({ queryKey: ["useGetDeployments"] }); }, diff --git a/src/frontend/src/controllers/API/services/__tests__/request-processor.test.ts b/src/frontend/src/controllers/API/services/__tests__/request-processor.test.ts new file mode 100644 index 0000000000..00a401723b --- /dev/null +++ b/src/frontend/src/controllers/API/services/__tests__/request-processor.test.ts @@ -0,0 +1,189 @@ +// biome-ignore-all lint/suspicious/noExplicitAny: test mocks +const mockQueryClient = { + invalidateQueries: jest.fn(), +}; + +let mockCapturedQueryOptions: any = null; +let mockCapturedMutationOptions: any = null; + +jest.mock("@tanstack/react-query", () => ({ + useQueryClient: jest.fn(() => mockQueryClient), + useQuery: jest.fn((options: any) => { + mockCapturedQueryOptions = options; + return { data: undefined, isLoading: false }; + }), + useMutation: jest.fn((options: any) => { + mockCapturedMutationOptions = options; + return { mutate: jest.fn(), mutateAsync: jest.fn() }; + }), +})); + +import { UseRequestProcessor } from "../request-processor"; + +// --------------------------------------------------------------------------- +// Helpers +// --------------------------------------------------------------------------- + +// `axios.isAxiosError` checks `error.isAxiosError === true`, so fixtures must +// set that flag to be classified as HTTP errors rather than transient. +const axiosError = (status: number) => ({ + isAxiosError: true, + response: { status }, +}); +const axiosNetworkError = () => ({ isAxiosError: true, response: undefined }); +const nonAxiosError = () => new Error("Network Error"); + +beforeEach(() => { + jest.clearAllMocks(); + mockCapturedQueryOptions = null; + mockCapturedMutationOptions = null; +}); + +// --------------------------------------------------------------------------- +// Queries: retry policy +// --------------------------------------------------------------------------- + +describe("UseRequestProcessor.query retry policy", () => { + const setup = (options: any = {}) => { + const { query } = UseRequestProcessor(); + query(["k"], async () => ({}), options); + if (mockCapturedQueryOptions == null) { + throw new Error("query was not called by UseRequestProcessor"); + } + return mockCapturedQueryOptions; + }; + + it("does not retry on any 4xx response", () => { + const { retry } = setup(); + for (const status of [400, 401, 403, 404, 409, 422, 429, 499]) { + expect(retry(0, axiosError(status))).toBe(false); + } + }); + + it("retries up to 5 times on 5xx responses", () => { + const { retry } = setup(); + for (const status of [500, 502, 503, 504]) { + expect(retry(0, axiosError(status))).toBe(true); + expect(retry(4, axiosError(status))).toBe(true); + expect(retry(5, axiosError(status))).toBe(false); + } + }); + + it("retries up to 5 times on axios errors with no response", () => { + const { retry } = setup(); + expect(retry(0, axiosNetworkError())).toBe(true); + expect(retry(4, axiosNetworkError())).toBe(true); + expect(retry(5, axiosNetworkError())).toBe(false); + }); + + it("treats non-axios / unknown errors as transient (retries)", () => { + const { retry } = setup(); + expect(retry(0, undefined)).toBe(true); + expect(retry(0, {})).toBe(true); + expect(retry(0, { response: {} })).toBe(true); + expect(retry(0, nonAxiosError())).toBe(true); + }); + + it("allows per-call options.retry to override the default", () => { + const { retry } = setup({ retry: false }); + expect(retry).toBe(false); + }); + + it("allows per-call options.retry as a number to override the default", () => { + const { retry } = setup({ retry: 10 }); + expect(retry).toBe(10); + }); + + it("allows per-call options.retryDelay to override the default", () => { + const customDelay = jest.fn(() => 42); + const { retryDelay } = setup({ retryDelay: customDelay }); + expect(retryDelay).toBe(customDelay); + }); +}); + +// --------------------------------------------------------------------------- +// Mutations: retry policy +// --------------------------------------------------------------------------- + +describe("UseRequestProcessor.mutate retry policy", () => { + const setup = (options: any = {}) => { + const { mutate } = UseRequestProcessor(); + mutate(["k"], async () => ({}), options); + if (mockCapturedMutationOptions == null) { + throw new Error("mutate was not called by UseRequestProcessor"); + } + return mockCapturedMutationOptions; + }; + + it("does not retry on any 4xx response", () => { + const { retry } = setup(); + for (const status of [400, 401, 403, 404, 409, 422, 429, 499]) { + expect(retry(0, axiosError(status))).toBe(false); + } + }); + + it("retries up to 3 times on 5xx responses", () => { + const { retry } = setup(); + expect(retry(0, axiosError(500))).toBe(true); + expect(retry(2, axiosError(500))).toBe(true); + expect(retry(3, axiosError(500))).toBe(false); + }); + + it("retries up to 3 times on axios errors with no response", () => { + const { retry } = setup(); + expect(retry(0, axiosNetworkError())).toBe(true); + expect(retry(2, axiosNetworkError())).toBe(true); + expect(retry(3, axiosNetworkError())).toBe(false); + }); + + it("respects options.retry === false", () => { + const { retry } = setup({ retry: false }); + expect(retry).toBe(false); + }); + + it("respects options.retry === 0 (nullish coalescing, not falsy)", () => { + const { retry } = setup({ retry: 0 }); + expect(retry).toBe(0); + }); + + it("respects a custom options.retry callback", () => { + const customRetry = jest.fn(() => true); + const { retry } = setup({ retry: customRetry }); + expect(retry).toBe(customRetry); + }); + + it("respects a custom options.retryDelay", () => { + const customDelay = jest.fn(() => 42); + const { retryDelay } = setup({ retryDelay: customDelay }); + expect(retryDelay).toBe(customDelay); + }); +}); + +// --------------------------------------------------------------------------- +// retryDelay: exponential backoff capped at 30s +// --------------------------------------------------------------------------- + +describe("UseRequestProcessor retryDelay", () => { + it("uses exponential backoff capped at 30s for queries", () => { + const { query } = UseRequestProcessor(); + query(["k"], async () => ({})); + const { retryDelay } = mockCapturedQueryOptions; + expect(retryDelay(0)).toBe(1000); + expect(retryDelay(1)).toBe(2000); + expect(retryDelay(2)).toBe(4000); + expect(retryDelay(3)).toBe(8000); + expect(retryDelay(4)).toBe(16000); + expect(retryDelay(5)).toBe(30000); + expect(retryDelay(10)).toBe(30000); + }); + + it("uses exponential backoff capped at 30s for mutations", () => { + const { mutate } = UseRequestProcessor(); + mutate(["k"], async () => ({})); + const { retryDelay } = mockCapturedMutationOptions; + expect(retryDelay(0)).toBe(1000); + expect(retryDelay(1)).toBe(2000); + expect(retryDelay(2)).toBe(4000); + expect(retryDelay(10)).toBe(30000); + }); +}); diff --git a/src/frontend/src/controllers/API/services/request-processor.ts b/src/frontend/src/controllers/API/services/request-processor.ts index 41e9c19a88..6b54d9ef1e 100644 --- a/src/frontend/src/controllers/API/services/request-processor.ts +++ b/src/frontend/src/controllers/API/services/request-processor.ts @@ -6,11 +6,33 @@ import { useQuery, useQueryClient, } from "@tanstack/react-query"; +import axios from "axios"; import type { MutationFunctionType, QueryFunctionType, } from "../../../types/api"; +// 4xx responses are intentional client-side rejections (auth, validation, +// deployment guards, etc.) and won't change on retry. +function isClientError(error: unknown): boolean { + if (!axios.isAxiosError(error)) return false; + const status = error.response?.status; + return typeof status === "number" && status >= 400 && status < 500; +} + +function makeRetry(maxRetries: number) { + return (failureCount: number, error: unknown) => { + if (isClientError(error)) return false; + return failureCount < maxRetries; + }; +} + +const queryRetry = makeRetry(5); +const mutationRetry = makeRetry(3); + +const retryDelay = (attemptIndex: number) => + Math.min(1000 * 2 ** attemptIndex, 30000); + export function UseRequestProcessor(): { query: QueryFunctionType; mutate: MutationFunctionType; @@ -26,8 +48,8 @@ export function UseRequestProcessor(): { return useQuery({ queryKey, queryFn, - retry: 5, - retryDelay: (attemptIndex) => Math.min(1000 * 2 ** attemptIndex, 30000), + retry: queryRetry, + retryDelay, ...options, }); } @@ -45,8 +67,8 @@ export function UseRequestProcessor(): { options.onSettled && options.onSettled(data, error, variables, context); }, ...options, - retry: options.retry ?? 3, - retryDelay: (attemptIndex) => Math.min(1000 * 2 ** attemptIndex, 30000), + retry: options.retry ?? mutationRetry, + retryDelay: options.retryDelay ?? retryDelay, }); } diff --git a/src/frontend/src/modals/IOModal/components/chatView/chatInput/components/input-wrapper.tsx b/src/frontend/src/modals/IOModal/components/chatView/chatInput/components/input-wrapper.tsx index dcd86ab347..db83845685 100644 --- a/src/frontend/src/modals/IOModal/components/chatView/chatInput/components/input-wrapper.tsx +++ b/src/frontend/src/modals/IOModal/components/chatView/chatInput/components/input-wrapper.tsx @@ -94,7 +94,7 @@ const InputWrapper: React.FC = ({
diff --git a/src/frontend/src/modals/confirmationModal/index.tsx b/src/frontend/src/modals/confirmationModal/index.tsx index e7a510313b..694012cfe1 100644 --- a/src/frontend/src/modals/confirmationModal/index.tsx +++ b/src/frontend/src/modals/confirmationModal/index.tsx @@ -50,7 +50,7 @@ function ConfirmationModal({ const [flag, setFlag] = useState(false); useEffect(() => { - if (open) setModalOpen(open); + if (open !== undefined) setModalOpen(open); }, [open]); useEffect(() => { @@ -81,7 +81,7 @@ function ConfirmationModal({ return ( diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/create-mode.test.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/create-mode.test.tsx index c0d7f1e73f..074de71ac4 100644 --- a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/create-mode.test.tsx +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/create-mode.test.tsx @@ -218,6 +218,42 @@ describe("Create mode — canGoNext validation", () => { expect(result.current.canGoNext).toBe(true); }); + + it("blocks when trimmed name does not start with a letter", () => { + const { result } = renderCreateHook({ + initialProvider: mockProvider, + initialInstance: mockInstance, + }); + + act(() => result.current.handleNext()); // → step 2 + + act(() => { + result.current.setDeploymentName(" 1 Agent"); + result.current.setSelectedLlm("gpt-4"); + }); + + expect(result.current.canGoNext).toBe(false); + expect(result.current.isDeploymentNameValid).toBe(false); + expect(result.current.hasDeploymentNameFormatError).toBe(true); + }); + + it("allows when trimmed name starts with a unicode letter", () => { + const { result } = renderCreateHook({ + initialProvider: mockProvider, + initialInstance: mockInstance, + }); + + act(() => result.current.handleNext()); // → step 2 + + act(() => { + result.current.setDeploymentName(" Ágent"); + result.current.setSelectedLlm("gpt-4"); + }); + + expect(result.current.canGoNext).toBe(true); + expect(result.current.isDeploymentNameValid).toBe(true); + expect(result.current.hasDeploymentNameFormatError).toBe(false); + }); }); describe("Step 3 (Attach Flows)", () => { diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/custom-tool-naming.test.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/custom-tool-naming.test.tsx index 3acfacc87e..51328a4efe 100644 --- a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/custom-tool-naming.test.tsx +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/custom-tool-naming.test.tsx @@ -180,4 +180,18 @@ describe("Custom tool naming", () => { "Tool Beta", ); }); + + it("rejects create payload when agent name does not start with a letter", () => { + const { result } = renderStepperHook(); + + act(() => { + result.current.setDeploymentName("1 Agent"); + result.current.setSelectedLlm("test-model"); + result.current.handleSelectVersion("flow-1", "ver-1", "v1"); + }); + + expect(() => result.current.buildDeploymentPayload("provider-1")).toThrow( + "Deployment name must start with a letter", + ); + }); }); diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/deployment-stepper-footer.test.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/deployment-stepper-footer.test.tsx new file mode 100644 index 0000000000..7e721a9c74 --- /dev/null +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/deployment-stepper-footer.test.tsx @@ -0,0 +1,75 @@ +import { render, screen } from "@testing-library/react"; +import userEvent from "@testing-library/user-event"; +import DeploymentStepperFooter from "../components/deployment-stepper-footer"; + +jest.mock( + "@/components/common/genericIconComponent", + () => + function MockIcon({ name }: { name: string }) { + return ; + }, +); + +describe("DeploymentStepperFooter", () => { + it("shows done state controls when deployed", () => { + render( + , + ); + + expect(screen.getByRole("button", { name: "Done" })).toBeInTheDocument(); + expect( + screen.queryByRole("button", { name: "Cancel" }), + ).not.toBeInTheDocument(); + expect( + screen.queryByRole("button", { name: "Back" }), + ).not.toBeInTheDocument(); + expect( + screen.queryByTestId("deployment-stepper-next"), + ).not.toBeInTheDocument(); + }); + + it("calls onClose from done button", async () => { + const user = userEvent.setup(); + const onClose = jest.fn(); + + render( + , + ); + + await user.click(screen.getByRole("button", { name: "Done" })); + + expect(onClose).toHaveBeenCalled(); + }); +}); diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/deployment-stepper-modal.test.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/deployment-stepper-modal.test.tsx index 4640affa27..a719aa06d8 100644 --- a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/deployment-stepper-modal.test.tsx +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/deployment-stepper-modal.test.tsx @@ -345,7 +345,7 @@ describe("Step labels", () => { renderModal(); expect(screen.getByText("Provider")).toBeInTheDocument(); expect(screen.getByText("Type")).toBeInTheDocument(); - expect(screen.getByText("Attach Flows")).toBeInTheDocument(); + expect(screen.getByText("Flows")).toBeInTheDocument(); expect(screen.getByText("Review")).toBeInTheDocument(); }); @@ -355,7 +355,7 @@ describe("Step labels", () => { renderModal({ editingDeployment: makeDeployment() }); expect(screen.queryByText("Provider")).not.toBeInTheDocument(); expect(screen.getByText("Type")).toBeInTheDocument(); - expect(screen.getByText("Attach Flows")).toBeInTheDocument(); + expect(screen.getByText("Flows")).toBeInTheDocument(); expect(screen.getByText("Review")).toBeInTheDocument(); }); }); diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/deployment-success-content.test.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/deployment-success-content.test.tsx new file mode 100644 index 0000000000..2fab3d6881 --- /dev/null +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/deployment-success-content.test.tsx @@ -0,0 +1,86 @@ +import { render, screen } from "@testing-library/react"; +import userEvent from "@testing-library/user-event"; +import DeploymentSuccessContent from "../components/deployment-success-content"; + +jest.mock( + "@/components/common/genericIconComponent", + () => + function MockIcon({ name }: { name: string }) { + return ; + }, +); + +describe("DeploymentSuccessContent", () => { + it("renders success copy and provider link", () => { + render( + , + ); + + expect(screen.getByText("Deployment successful")).toBeInTheDocument(); + expect( + screen.getByText("Deployed to watsonx Orchestrate as draft"), + ).toBeInTheDocument(); + expect( + screen.getByRole("link", { name: /watsonx Orchestrate/i }), + ).toHaveAttribute( + "href", + "https://www.ibm.com/products/watsonx-orchestrate", + ); + }); + + it("shows test button and calls handler", async () => { + const user = userEvent.setup(); + const onTest = jest.fn(); + + render( + , + ); + + await user.click(screen.getByRole("button", { name: "Test Deployment" })); + + expect(onTest).toHaveBeenCalled(); + }); + + it("hides test button without deployment name", () => { + render( + , + ); + + expect( + screen.queryByRole("button", { name: "Test Deployment" }), + ).not.toBeInTheDocument(); + }); + + it("hides test button when showTestButton is false", () => { + render( + , + ); + + expect( + screen.queryByRole("button", { name: "Test Deployment" }), + ).not.toBeInTheDocument(); + }); +}); diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/edit-mode.test.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/edit-mode.test.tsx index 86ca32849d..fd86cf3dfa 100644 --- a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/edit-mode.test.tsx +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/edit-mode.test.tsx @@ -92,6 +92,31 @@ describe("Edit mode — basic state", () => { expect(result.current.canGoNext).toBe(true); }); + it("blocks update flow when existing name does not start with a letter", () => { + const invalidDeployment = { ...mockDeployment, name: "1 Agent" }; + const wrapper = ({ children }: { children: React.ReactNode }) => ( + + {children} + + ); + const { result } = renderHook(() => useDeploymentStepper(), { wrapper }); + + expect(result.current.canGoNext).toBe(false); + expect(result.current.isDeploymentNameValid).toBe(false); + expect(result.current.hasDeploymentNameFormatError).toBe(true); + expect(() => result.current.buildDeploymentUpdatePayload()).toThrow( + "Deployment name must start with a letter", + ); + }); + it("canGoNext on step 2 (Attach) allows proceeding in edit mode", () => { const { result } = renderEditHook(); act(() => result.current.handleNext()); // step 2 diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/step-attach-flows.test.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/step-attach-flows.test.tsx index 49284ece5f..aa0375dd3a 100644 --- a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/step-attach-flows.test.tsx +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/step-attach-flows.test.tsx @@ -230,14 +230,14 @@ beforeEach(() => { // --------------------------------------------------------------------------- describe("Three-panel layout rendering", () => { - it("renders the Attach Flows heading", () => { + it("renders the Flows heading", () => { render(); - expect(screen.getByText("Attach Flows")).toBeInTheDocument(); + expect(screen.getByText("Flows")).toBeInTheDocument(); }); - it("renders the Available Flows panel header", () => { + it("renders the Available panel header", () => { render(); - expect(screen.getByText("Available Flows")).toBeInTheDocument(); + expect(screen.getByText("Available")).toBeInTheDocument(); }); it("renders flow items in the list", () => { diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/step-type.test.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/step-type.test.tsx index aafd885939..1cd2f07022 100644 --- a/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/step-type.test.tsx +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/__tests__/step-type.test.tsx @@ -13,6 +13,8 @@ const mockSetSelectedLlm = jest.fn(); let mockIsEditMode = false; let mockDeploymentType = "agent"; let mockDeploymentName = ""; +let mockIsDeploymentNameValid = false; +let mockHasDeploymentNameFormatError = false; let mockDeploymentDescription = ""; let mockSelectedLlm = ""; let mockSelectedInstance: { id: string } | null = { id: "inst-1" }; @@ -26,6 +28,8 @@ jest.mock("../contexts/deployment-stepper-context", () => ({ setDeploymentType: mockSetDeploymentType, deploymentName: mockDeploymentName, setDeploymentName: mockSetDeploymentName, + isDeploymentNameValid: mockIsDeploymentNameValid, + hasDeploymentNameFormatError: mockHasDeploymentNameFormatError, deploymentDescription: mockDeploymentDescription, setDeploymentDescription: mockSetDeploymentDescription, selectedLlm: mockSelectedLlm, @@ -61,6 +65,8 @@ beforeEach(() => { mockIsEditMode = false; mockDeploymentType = "agent"; mockDeploymentName = ""; + mockIsDeploymentNameValid = false; + mockHasDeploymentNameFormatError = false; mockDeploymentDescription = ""; mockSelectedLlm = ""; mockSelectedInstance = { id: "inst-1" }; @@ -143,6 +149,30 @@ describe("Name input", () => { screen.queryByText("Name cannot be changed after creation."), ).not.toBeInTheDocument(); }); + + it("shows validation error when name does not start with a letter", () => { + mockDeploymentName = "1 Agent"; + mockHasDeploymentNameFormatError = true; + render(); + expect( + screen.getByText("Agent name must start with a letter."), + ).toBeInTheDocument(); + expect(screen.getByPlaceholderText("e.g., Sales Bot")).toHaveAttribute( + "aria-invalid", + "true", + ); + }); + + it("does not show validation error for empty name", () => { + render(); + expect( + screen.queryByText("Agent name must start with a letter."), + ).not.toBeInTheDocument(); + expect(screen.getByPlaceholderText("e.g., Sales Bot")).toHaveAttribute( + "aria-invalid", + "false", + ); + }); }); // --------------------------------------------------------------------------- diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-stepper-footer.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-stepper-footer.tsx new file mode 100644 index 0000000000..ef0a056d99 --- /dev/null +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-stepper-footer.tsx @@ -0,0 +1,89 @@ +import ForwardedIconComponent from "@/components/common/genericIconComponent"; +import { Button } from "@/components/ui/button"; + +interface DeploymentStepperFooterProps { + canGoNext: boolean; + currentStep: number; + isCreatingAccount: boolean; + isDeployed: boolean; + isDeploying: boolean; + isInDeployPhase: boolean; + isFinalStep: boolean; + minStep: number; + actionIcon: string; + actionLabel: string; + progressLabel: string; + onBack: () => void; + onCancel: () => void; + onClose: () => void; + onPrimaryAction: () => void; +} + +export default function DeploymentStepperFooter({ + canGoNext, + currentStep, + isCreatingAccount, + isDeployed, + isDeploying, + isInDeployPhase, + isFinalStep, + minStep, + actionIcon, + actionLabel, + progressLabel, + onBack, + onCancel, + onClose, + onPrimaryAction, +}: DeploymentStepperFooterProps) { + return ( +
+ {isDeployed ? ( +
+ ) : ( + + )} +
+ {!isDeployed && ( + + )} + {!isInDeployPhase && ( + + )} + {isDeploying && ( + + )} + {isDeployed && } +
+
+ ); +} diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-stepper-modal.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-stepper-modal.tsx index fc4d5c99ee..7bfabc79a3 100644 --- a/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-stepper-modal.tsx +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-stepper-modal.tsx @@ -1,7 +1,5 @@ import { useMemo, useState } from "react"; import { useParams } from "react-router-dom"; -import ForwardedIconComponent from "@/components/common/genericIconComponent"; -import { Button } from "@/components/ui/button"; import { Dialog, DialogContent, @@ -21,7 +19,9 @@ import { } from "../contexts/deployment-stepper-context"; import { useErrorAlert } from "../hooks/use-error-alert"; import type { Deployment, DeploymentProvider, ProviderAccount } from "../types"; -import DeploymentStepper from "./deployment-stepper"; +import DeploymentStepper, { CREATE_DEPLOYED_STEPS } from "./deployment-stepper"; +import DeploymentStepperFooter from "./deployment-stepper-footer"; +import DeploymentSuccessContent from "./deployment-success-content"; import StepAttachFlows from "./step-attach-flows"; import StepDeployStatus from "./step-deploy-status"; import StepProvider from "./step-provider"; @@ -206,6 +206,7 @@ function DeploymentStepperModalContent({ canGoNext, handleNext, handleBack, + selectedProvider, selectedInstance, setSelectedInstance, needsProviderAccountCreation, @@ -225,6 +226,8 @@ function DeploymentStepperModalContent({ const isDeploying = deploymentPhase === "deploying"; const isDeployed = deploymentPhase === "deployed"; const isInDeployPhase = isDeploying || isDeployed; + const providerConsoleUrl = "https://www.ibm.com/products/watsonx-orchestrate"; + const providerDisplayName = selectedProvider?.name ?? "watsonx Orchestrate"; // In edit mode, steps are shifted: 1=Type, 2=Attach, 3=Review. const logicalStep = isEditMode ? currentStep + 1 : currentStep; @@ -316,19 +319,40 @@ function DeploymentStepperModalContent({ className="text-center text-2xl font-semibold" data-testid="stepper-modal-title" > - {isEditMode ? "Update Deployment" : "Create New Deployment"} + {isDeployed && !isEditMode + ? "Deployed" + : isEditMode + ? "Update Deployment" + : "Create New Deployment"} - +
{/* Content box: step content + footer */}
{isInDeployPhase ? ( - + isDeployed && !isEditMode ? ( + + ) : ( + + ) ) : ( <> {logicalStep === 1 && } @@ -339,61 +363,23 @@ function DeploymentStepperModalContent({ )}
- {/* Footer */} -
- -
- {!isDeployed && ( - - )} - {!isInDeployPhase && ( - - )} - {isDeploying && ( - - )} - {isDeployed && onTestDeployment && !isEditMode && ( - - )} -
-
+ setOpen(false)} + onClose={() => setOpen(false)} + onPrimaryAction={isFinalStep ? handleDeploy : handleStepNext} + />
); diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-stepper.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-stepper.tsx index 903abf6144..3abc44b849 100644 --- a/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-stepper.tsx +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-stepper.tsx @@ -1,25 +1,41 @@ import { cn } from "@/utils/utils"; import { useDeploymentStepper } from "../contexts/deployment-stepper-context"; -const CREATE_STEPS = [ +export const CREATE_STEPS = [ { number: 1, label: "Provider" }, { number: 2, label: "Type" }, - { number: 3, label: "Attach Flows" }, + { number: 3, label: "Flows" }, { number: 4, label: "Review" }, ] as const; +export const CREATE_DEPLOYED_STEPS = [ + { number: 1, label: "Provider" }, + { number: 2, label: "Type" }, + { number: 3, label: "Flows" }, + { number: 4, label: "Deployed" }, +] as const; + const EDIT_STEPS = [ { number: 1, label: "Type" }, - { number: 2, label: "Attach Flows" }, + { number: 2, label: "Flows" }, { number: 3, label: "Review" }, ] as const; export const DEPLOYMENT_STEPS = CREATE_STEPS; -export default function DeploymentStepper() { +interface DeploymentStepperProps { + steps?: readonly { number: number; label: string }[]; + currentStepOverride?: number; +} + +export default function DeploymentStepper({ + steps: stepsProp, + currentStepOverride, +}: DeploymentStepperProps) { const { currentStep, isEditMode } = useDeploymentStepper(); - const steps = isEditMode ? EDIT_STEPS : CREATE_STEPS; - const progressPercent = ((currentStep - 1) / (steps.length - 1)) * 100; + const steps = stepsProp ?? (isEditMode ? EDIT_STEPS : CREATE_STEPS); + const activeStep = currentStepOverride ?? currentStep; + const progressPercent = ((activeStep - 1) / (steps.length - 1)) * 100; return (
@@ -35,7 +51,7 @@ export default function DeploymentStepper() {
= step.number + activeStep >= step.number ? "bg-foreground text-background" : "bg-muted text-muted-foreground", )} @@ -45,7 +61,7 @@ export default function DeploymentStepper() { = step.number && "font-medium", + activeStep >= step.number && "font-medium", )} > {step.label} diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-success-content.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-success-content.tsx new file mode 100644 index 0000000000..1cf6280749 --- /dev/null +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/deployment-success-content.tsx @@ -0,0 +1,63 @@ +import ForwardedIconComponent from "@/components/common/genericIconComponent"; +import { Button } from "@/components/ui/button"; + +interface DeploymentSuccessContentProps { + deploymentName?: string; + providerName: string; + providerUrl: string; + showTestButton: boolean; + onTest: () => void; +} + +export default function DeploymentSuccessContent({ + deploymentName, + providerName, + providerUrl, + showTestButton, + onTest, +}: DeploymentSuccessContentProps) { + return ( +
+
+ +
+ +
+

+ Deployment successful +

+

+ Deployed to {providerName} as draft +

+
+ +
+ Publish from Draft to Live in + + {providerName} + + +
+ + {showTestButton && deploymentName && ( + + )} +
+ ); +} diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/step-attach-flows-flow-list-panel.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/step-attach-flows-flow-list-panel.tsx index 6303ed25d3..6a5ba993bd 100644 --- a/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/step-attach-flows-flow-list-panel.tsx +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/step-attach-flows-flow-list-panel.tsx @@ -29,7 +29,7 @@ export const FlowListPanel = memo(function FlowListPanel({ return (
- Available Flows + Available
{flows.map((flow) => { diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/step-attach-flows.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/step-attach-flows.tsx index 1beac14672..07141246e8 100644 --- a/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/step-attach-flows.tsx +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/components/step-attach-flows.tsx @@ -259,7 +259,7 @@ export default function StepAttachFlows() { return (
-

Attach Flows

+

Flows

setDeploymentName(e.target.value)} disabled={isEditMode} + aria-invalid={hasDeploymentNameFormatError} /> + {hasDeploymentNameFormatError && ( + + Agent name must start with a letter. + + )} {isEditMode && ( Name cannot be changed after creation. diff --git a/src/frontend/src/pages/MainPage/pages/deploymentsPage/contexts/deployment-stepper-context.tsx b/src/frontend/src/pages/MainPage/pages/deploymentsPage/contexts/deployment-stepper-context.tsx index 79341c3be5..3a46d79cd4 100644 --- a/src/frontend/src/pages/MainPage/pages/deploymentsPage/contexts/deployment-stepper-context.tsx +++ b/src/frontend/src/pages/MainPage/pages/deploymentsPage/contexts/deployment-stepper-context.tsx @@ -73,6 +73,8 @@ interface DeploymentStepperContextType { setDeploymentType: (type: DeploymentType) => void; deploymentName: string; setDeploymentName: (name: string) => void; + isDeploymentNameValid: boolean; + hasDeploymentNameFormatError: boolean; deploymentDescription: string; setDeploymentDescription: (description: string) => void; selectedLlm: string; @@ -168,6 +170,11 @@ export function DeploymentStepperProvider({ >(initialState?.initialConnectionsByFlow ?? new Map()); const [hasToolNameErrors, setHasToolNameErrors] = useState(false); + const trimmedDeploymentName = deploymentName.trim(); + const hasDeploymentNameFormatError = + trimmedDeploymentName !== "" && !/^\p{L}/u.test(trimmedDeploymentName); + const isDeploymentNameValid = + trimmedDeploymentName !== "" && !hasDeploymentNameFormatError; // Edit mode: track which pre-existing flows the user wants to detach. const [removedFlowIds, setRemovedFlowIds] = useState>(new Set()); @@ -253,7 +260,7 @@ export function DeploymentStepperProvider({ ); } if (logical === 2) { - return deploymentName.trim() !== "" && selectedLlm.trim() !== ""; + return isDeploymentNameValid && selectedLlm.trim() !== ""; } if (logical === 3) { // In edit mode, user can proceed without new attachments (may just change desc/LLM). @@ -270,6 +277,7 @@ export function DeploymentStepperProvider({ selectedInstance, hasValidCredentials, deploymentName, + isDeploymentNameValid, selectedLlm, selectedVersionByFlow, isEditMode, @@ -357,6 +365,9 @@ export function DeploymentStepperProvider({ const buildDeploymentPayload = useCallback( (providerId: string): DeploymentCreateRequest => { + if (!isDeploymentNameValid) { + throw new Error("Deployment name must start with a letter"); + } const allConnectionIds = new Set(); Array.from(attachedConnectionByFlow.values()).forEach((ids) => { ids.forEach((id) => allConnectionIds.add(id)); @@ -381,7 +392,7 @@ export function DeploymentStepperProvider({ ...(initialState?.projectId ? { project_id: initialState.projectId } : {}), - name: deploymentName, + name: trimmedDeploymentName, description: deploymentDescription, type: deploymentType, provider_data: { @@ -396,10 +407,11 @@ export function DeploymentStepperProvider({ buildConnectionPayloads, initialState?.projectId, deploymentDescription, - deploymentName, deploymentType, + isDeploymentNameValid, selectedLlm, selectedVersionByFlow, + trimmedDeploymentName, toolNameByFlow, ], ); @@ -411,6 +423,9 @@ export function DeploymentStepperProvider({ "buildDeploymentUpdatePayload called outside edit mode", ); } + if (!isDeploymentNameValid) { + throw new Error("Deployment name must start with a letter"); + } const result: DeploymentUpdateRequest = { deployment_id: editingDeployment.id, @@ -510,6 +525,7 @@ export function DeploymentStepperProvider({ }, [ editingDeployment, deploymentDescription, + isDeploymentNameValid, selectedLlm, initialVersionByFlow, initialToolNameByFlow, @@ -541,6 +557,8 @@ export function DeploymentStepperProvider({ setDeploymentType, deploymentName, setDeploymentName, + isDeploymentNameValid, + hasDeploymentNameFormatError, deploymentDescription, setDeploymentDescription, selectedLlm, @@ -581,6 +599,8 @@ export function DeploymentStepperProvider({ credentials, deploymentType, deploymentName, + isDeploymentNameValid, + hasDeploymentNameFormatError, deploymentDescription, selectedLlm, connections, diff --git a/src/frontend/tests/core/features/deployment-create.spec.ts b/src/frontend/tests/core/features/deployment-create.spec.ts index c1a02ee7e3..67d0bf111f 100644 --- a/src/frontend/tests/core/features/deployment-create.spec.ts +++ b/src/frontend/tests/core/features/deployment-create.spec.ts @@ -150,12 +150,13 @@ async function selectProvider(page: Page) { async function goToStepType(page: Page) { await selectProvider(page); await page.getByTestId("deployment-stepper-next").click(); - // Wait for the Type step heading - await page.waitForSelector("text=Deployment Type"); + await expect( + page.getByRole("heading", { name: /Deployment Type/i }), + ).toBeVisible(); } // --------------------------------------------------------------------------- -// Helper: navigate steps 1 → 2 → 3 (provider → type → attach flows) +// Helper: navigate steps 1 → 2 → 3 (provider → type → flows) // --------------------------------------------------------------------------- async function goToStepAttachFlows(page: Page) { await goToStepType(page); @@ -165,13 +166,13 @@ async function goToStepAttachFlows(page: Page) { // Select the LLM model await page.getByRole("combobox").click(); await page.getByRole("option", { name: "ibm/granite-13b-chat" }).click(); - // Advance to attach flows + // Advance to flows await page.getByTestId("deployment-stepper-next").click(); - await page.waitForSelector("text=Attach Flows"); + await expect(page.getByRole("heading", { name: /^Flows$/i })).toBeVisible(); } // --------------------------------------------------------------------------- -// Helper: navigate all steps through attach flows and select a flow+version +// Helper: navigate all steps through flows and select a flow+version // --------------------------------------------------------------------------- async function goToStepReview(page: Page) { await goToStepAttachFlows(page); @@ -189,7 +190,9 @@ async function goToStepReview(page: Page) { } // Advance to review await page.getByTestId("deployment-stepper-next").click(); - await page.waitForSelector("text=Review & Confirm"); + await expect( + page.getByRole("heading", { name: /Review & Confirm/i }), + ).toBeVisible(); } // --------------------------------------------------------------------------- @@ -197,7 +200,9 @@ async function goToStepReview(page: Page) { // --------------------------------------------------------------------------- test( "deployment-create: opens stepper on New Deployment click", - { tag: ["@deployment", "@workspace"] }, + { + tag: ["@deployment", "@workspace"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -220,7 +225,9 @@ test( // --------------------------------------------------------------------------- test( "deployment-create: step 1 provider - Next disabled without selection, enabled after selecting", - { tag: ["@deployment", "@workspace"] }, + { + tag: ["@deployment", "@workspace"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -245,7 +252,9 @@ test( // --------------------------------------------------------------------------- test( "deployment-create: step 2 type - fill name and select type to enable Next", - { tag: ["@deployment", "@workspace"] }, + { + tag: ["@deployment", "@workspace"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -280,11 +289,13 @@ test( ); // --------------------------------------------------------------------------- -// Test 4: Step 3 (Attach Flows) — select a flow and version, Next enables +// Test 4: Step 3 (Flows) — select a flow and version, Next enables // --------------------------------------------------------------------------- test( "deployment-create: step 3 attach flows - select flow and version enables Next", - { tag: ["@deployment", "@workspace"] }, + { + tag: ["@deployment", "@workspace"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -331,7 +342,9 @@ test( // --------------------------------------------------------------------------- test( "deployment-create: step 4 review - shows review content and Deploy button text", - { tag: ["@deployment", "@workspace"] }, + { + tag: ["@deployment", "@workspace"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -361,7 +374,9 @@ test( // --------------------------------------------------------------------------- test( "deployment-create: clicking Deploy triggers POST and shows deploy status", - { tag: ["@deployment", "@workspace"] }, + { + tag: ["@deployment", "@workspace"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -414,7 +429,9 @@ test( // --------------------------------------------------------------------------- test( "deployment-create: user can change tool name on review step", - { tag: ["@deployment", "@workspace"] }, + { + tag: ["@deployment", "@workspace"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -451,7 +468,9 @@ test( // --------------------------------------------------------------------------- test( "deployment-create: review step shows error when tool name already exists in provider", - { tag: ["@deployment", "@workspace"] }, + { + tag: ["@deployment", "@workspace"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -476,7 +495,9 @@ test( // --------------------------------------------------------------------------- test( "deployment-create: review step shows no error when tool name is unique", - { tag: ["@deployment", "@workspace"] }, + { + tag: ["@deployment", "@workspace"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -501,7 +522,9 @@ test( // --------------------------------------------------------------------------- test( "deployment-create: editing tool name to unique value clears duplicate error", - { tag: ["@deployment", "@workspace"] }, + { + tag: ["@deployment", "@workspace"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", diff --git a/src/frontend/tests/core/features/deployment-edit.spec.ts b/src/frontend/tests/core/features/deployment-edit.spec.ts index fbec55b3f2..ba8062ccae 100644 --- a/src/frontend/tests/core/features/deployment-edit.spec.ts +++ b/src/frontend/tests/core/features/deployment-edit.spec.ts @@ -83,9 +83,23 @@ async function openEditDialog(page: Parameters[2]["page"]) { await page.waitForSelector('[data-testid="stepper-modal-title"]'); } +async function expectDeploymentTypeStep( + page: Parameters[2]["page"], +) { + await expect( + page.getByRole("heading", { name: /Deployment Type/i }), + ).toBeVisible(); +} + +async function expectFlowsStep(page: Parameters[2]["page"]) { + await expect(page.getByRole("heading", { name: /^Flows$/i })).toBeVisible(); +} + test( "Opens edit stepper from actions menu", - { tag: ["@release", "@workspace", "@api"] }, + { + tag: ["@release", "@workspace", "@api"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -110,7 +124,9 @@ test( test( "Edit mode skips provider step — starts at Type", - { tag: ["@release", "@workspace", "@api"] }, + { + tag: ["@release", "@workspace", "@api"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -126,7 +142,7 @@ test( await openEditDialog(page); // Wait for stepper body to render (parallel fetches must complete) - await page.waitForSelector('h2:has-text("Deployment Type")'); + await expectDeploymentTypeStep(page); await expect(page.getByText("Deployment Type")).toBeVisible(); @@ -141,7 +157,9 @@ test( test( "Name field pre-populated in edit mode", - { tag: ["@release", "@workspace", "@api"] }, + { + tag: ["@release", "@workspace", "@api"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -157,7 +175,7 @@ test( await openEditDialog(page); // Wait for stepper body to render - await page.waitForSelector('h2:has-text("Deployment Type")'); + await expectDeploymentTypeStep(page); const nameInput = page.getByPlaceholder("e.g., Sales Bot"); await expect(nameInput).toHaveValue("Test Deployment"); @@ -166,7 +184,9 @@ test( test( "Submitting PATCH closes modal", - { tag: ["@release", "@workspace", "@api"] }, + { + tag: ["@release", "@workspace", "@api"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -182,13 +202,13 @@ test( await openEditDialog(page); // Wait for stepper body to render (parallel fetches must complete) - await page.waitForSelector('h2:has-text("Deployment Type")'); + await expectDeploymentTypeStep(page); // Navigate through the stepper steps to reach Review // Step: Type → click Next await page.getByTestId("deployment-stepper-next").click(); - // Step: Attach Flows → click Next + // Step: Flows → click Next await page.getByTestId("deployment-stepper-next").click(); // Step: Review → click Update (final step) @@ -209,7 +229,9 @@ test( test( "Cancel during edit closes modal without calling PATCH", - { tag: ["@release", "@workspace", "@api"] }, + { + tag: ["@release", "@workspace", "@api"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -225,7 +247,7 @@ test( await openEditDialog(page); // Wait for stepper body to render - await page.waitForSelector('h2:has-text("Deployment Type")'); + await expectDeploymentTypeStep(page); let patchCalled = false; page.on("request", (req) => { @@ -356,7 +378,9 @@ async function setupRoutesWithConnections( // --------------------------------------------------------------------------- test( "Edit mode includes new connections in PATCH request", - { tag: ["@release", "@workspace", "@api"] }, + { + tag: ["@release", "@workspace", "@api"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -399,11 +423,11 @@ test( await page.waitForSelector('[data-testid="stepper-modal-title"]'); // Step 1 (Type) → Next - await page.waitForSelector('h2:has-text("Deployment Type")'); + await expectDeploymentTypeStep(page); await page.getByTestId("deployment-stepper-next").click(); - // Step 2 (Attach Flows) — flow "f1" should already be attached - await page.waitForSelector("text=Attach Flows"); + // Step 2 (Flows) — flow "f1" should already be attached + await expectFlowsStep(page); await page.waitForSelector('[data-testid="flow-item-f1"]'); // Click the pre-attached flow and its version to open the connection panel @@ -455,7 +479,9 @@ test( // --------------------------------------------------------------------------- test( "Edit mode includes removed connections in PATCH request", - { tag: ["@release", "@workspace", "@api"] }, + { + tag: ["@release", "@workspace", "@api"], + }, async ({ page }) => { test.skip( process.env.LANGFLOW_FEATURE_WXO_DEPLOYMENTS !== "true", @@ -497,11 +523,11 @@ test( await page.waitForSelector('[data-testid="stepper-modal-title"]'); // Step 1 (Type) → Next - await page.waitForSelector('h2:has-text("Deployment Type")'); + await expectDeploymentTypeStep(page); await page.getByTestId("deployment-stepper-next").click(); - // Step 2 (Attach Flows) - await page.waitForSelector("text=Attach Flows"); + // Step 2 (Flows) + await expectFlowsStep(page); await page.waitForSelector('[data-testid="flow-item-f1"]'); await page.getByTestId("flow-item-f1").click(); await page.waitForSelector('[data-testid="version-item-fv1"]'); diff --git a/src/frontend/tests/core/unit/agentDefaultToolsComponent.spec.ts b/src/frontend/tests/core/unit/agentDefaultToolsComponent.spec.ts new file mode 100644 index 0000000000..4ee5fe6de6 --- /dev/null +++ b/src/frontend/tests/core/unit/agentDefaultToolsComponent.spec.ts @@ -0,0 +1,103 @@ +import { expect, test } from "../../fixtures"; +import { adjustScreenView } from "../../utils/adjust-screen-view"; +import { awaitBootstrapTest } from "../../utils/await-bootstrap-test"; +import { + closeAdvancedOptions, + openAdvancedOptions, +} from "../../utils/open-advanced-options"; + +/** + * Covers the user-facing contract of the "Default Agent Tools" feature: + * see CZL/MANUAL_TEST_DEFAULT_AGENT_TOOLS.md scenarios S1, S3, S5. + * + * Backing unit tests (pytest) already cover runtime behaviour; these tests + * validate the UI wiring so a regression in the inputs list or default + * prompt value is caught in the release gate. + */ + +async function dragAgentOntoCanvas(page: import("@playwright/test").Page) { + await awaitBootstrapTest(page); + + await page.waitForSelector('[data-testid="blank-flow"]', { timeout: 30000 }); + await page.getByTestId("blank-flow").click(); + + await page.waitForSelector('[data-testid="sidebar-search-input"]', { + timeout: 30000, + }); + + await page.getByTestId("sidebar-search-input").click(); + await page.getByTestId("sidebar-search-input").fill("agent"); + + await page.waitForSelector('[data-testid="models_and_agentsAgent"]', { + timeout: 30000, + }); + + await page + .getByTestId("models_and_agentsAgent") + .dragTo(page.locator('//*[@id="react-flow-id"]')); + + await adjustScreenView(page); +} + +test( + "Agent ships with Calculator and Current Date toggles enabled by default (S1/S3)", + { tag: ["@release", "@workspace", "@components"] }, + async ({ page }) => { + await dragAgentOntoCanvas(page); + + // Focus the Agent node so its advanced-field drawer is reachable. + await page.getByTestId("div-generic-node").click(); + + await openAdvancedOptions(page); + + // Both advanced toggles exist as show-on-canvas checkboxes. + // Their default `value=True` is validated by the pytest suite + // (`test_should_have_placeholders_in_default_system_prompt` covers the + // default contract of the inputs list). + await expect( + page.locator('//*[@id="showadd_current_date_tool"]'), + ).toBeVisible({ timeout: 10000 }); + await expect( + page.locator('//*[@id="showadd_calculator_tool"]'), + ).toBeVisible({ timeout: 10000 }); + + // Flip the Calculator field visible on canvas so we can assert the toggle + // is active and can be switched off and on (S3). + await page.locator('//*[@id="showadd_calculator_tool"]').click(); + await closeAdvancedOptions(page); + + await adjustScreenView(page); + + const calculatorToggle = page.getByTestId( + "toggle_bool_add_calculator_tool", + ); + await expect(calculatorToggle).toBeVisible({ timeout: 10000 }); + expect(await calculatorToggle.isChecked()).toBeTruthy(); + + // S3 — user disables the toggle. + await calculatorToggle.click(); + expect(await calculatorToggle.isChecked()).toBeFalsy(); + + // Re-enable to confirm the control is bi-directional. + await calculatorToggle.click(); + expect(await calculatorToggle.isChecked()).toBeTruthy(); + }, +); + +test( + "Agent default system prompt contains {current_date} and {model_name} placeholders (S5)", + { tag: ["@release", "@workspace", "@components"] }, + async ({ page }) => { + await dragAgentOntoCanvas(page); + + // The placeholders must be present inside the Agent Instructions textarea + // so the dynamic injection has a discoverable effect out of the box. + const instructionsTextarea = page.getByTestId("textarea_str_system_prompt"); + await expect(instructionsTextarea).toBeVisible({ timeout: 10000 }); + + //