Refa: rm useless code. (#11238)

### Type of change

- [x] Refactoring

README.md

@@ -192,9 +192,10 @@ releases! 🌟
 
 ```bash
    $ cd ragflow/docker
-   
+  
    # Optional: use a stable tag (see releases: https://github.com/infiniflow/ragflow/releases), e.g.: git checkout v0.22.0
-
+   # This steps ensures the **entrypoint.sh** file in the code matches the Docker image version.
+   
    # Use CPU for DeepDoc tasks:
    $ docker compose -f docker-compose.yml up -d
 

README_id.md

@@ -192,6 +192,7 @@ Coba demo kami di [https://demo.ragflow.io](https://demo.ragflow.io).
    $ cd ragflow/docker
    
    # Opsional: gunakan tag stabil (lihat releases: https://github.com/infiniflow/ragflow/releases), contoh: git checkout v0.22.0
+   # This steps ensures the **entrypoint.sh** file in the code matches the Docker image version.
 
    # Use CPU for DeepDoc tasks:
    $ docker compose -f docker-compose.yml up -d

README_ja.md

@@ -172,6 +172,7 @@
    $ cd ragflow/docker
 
    # 任意: 安定版タグを利用 (一覧: https://github.com/infiniflow/ragflow/releases) 例: git checkout v0.22.0
+   # この手順は、コード内の entrypoint.sh ファイルが Docker イメージのバージョンと一致していることを確認します。
 
    # Use CPU for DeepDoc tasks:
    $ docker compose -f docker-compose.yml up -d

README_ko.md

@@ -174,6 +174,7 @@
    $ cd ragflow/docker
    
    # Optional: use a stable tag (see releases: https://github.com/infiniflow/ragflow/releases), e.g.: git checkout v0.22.0
+   # 이 단계는 코드의 entrypoint.sh 파일이 Docker 이미지 버전과 일치하도록 보장합니다.
 
    # Use CPU for DeepDoc tasks:
    $ docker compose -f docker-compose.yml up -d

README_pt_br.md

@@ -192,6 +192,7 @@ Experimente nossa demo em [https://demo.ragflow.io](https://demo.ragflow.io).
    $ cd ragflow/docker
    
    # Opcional: use uma tag estável (veja releases: https://github.com/infiniflow/ragflow/releases), ex.: git checkout v0.22.0
+   # Esta etapa garante que o arquivo entrypoint.sh no código corresponda à versão da imagem do Docker.
 
    # Use CPU for DeepDoc tasks:
    $ docker compose -f docker-compose.yml up -d

README_tzh.md

@@ -191,6 +191,7 @@
    $ cd ragflow/docker
    
    # 可選：使用穩定版標籤（查看發佈：https://github.com/infiniflow/ragflow/releases），例：git checkout v0.22.0
+   # 此步驟確保程式碼中的 entrypoint.sh 檔案與 Docker 映像版本一致。
 
    # Use CPU for DeepDoc tasks:
    $ docker compose -f docker-compose.yml up -d

README_zh.md

@@ -192,6 +192,7 @@
    $ cd ragflow/docker
    
    # 可选：使用稳定版本标签（查看发布：https://github.com/infiniflow/ragflow/releases），例如：git checkout v0.22.0
+   # 这一步确保代码中的 entrypoint.sh 文件与 Docker 镜像的版本保持一致。
 
    # Use CPU for DeepDoc tasks:
    $ docker compose -f docker-compose.yml up -d

agent/component/agent_with_tools.py

@@ -368,11 +368,19 @@ Respond immediately with your final comprehensive answer.
 
         return "Error occurred."
 
-    def reset(self, temp=False):
+    def reset(self, only_output=False):
         """
         Reset all tools if they have a reset method. This avoids errors for tools like MCPToolCallSession.
         """
+        for k in self._param.outputs.keys():
+            self._param.outputs[k]["value"] = None
+            
         for k, cpn in self.tools.items():
             if hasattr(cpn, "reset") and callable(cpn.reset):
                 cpn.reset()
+        if only_output:
+            return
+        for k in self._param.inputs.keys():
+            self._param.inputs[k]["value"] = None
+        self._param.debug_inputs = {}
 

agent/component/llm.py

@@ -222,7 +222,7 @@ class LLM(ComponentBase):
             output_structure = self._param.outputs['structured']
         except Exception:
             pass
-        if output_structure:
+        if output_structure and isinstance(output_structure, dict) and output_structure.get("properties"):
             schema=json.dumps(output_structure, ensure_ascii=False, indent=2)
             prompt += structured_output_prompt(schema)
             for _ in range(self._param.max_retries+1):

api/apps/api_app.py

@@ -13,41 +13,16 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 #
-import json
-import os
-import re
 from datetime import datetime, timedelta
-from flask import request, Response
-from api.db.services.llm_service import LLMBundle
+from flask import request
 from flask_login import login_required, current_user
-
-from api.db import VALID_FILE_TYPES, FileType
-from api.db.db_models import APIToken, Task, File
-from api.db.services import duplicate_name
+from api.db.db_models import APIToken
 from api.db.services.api_service import APITokenService, API4ConversationService
-from api.db.services.dialog_service import DialogService, chat
-from api.db.services.document_service import DocumentService, doc_upload_and_parse
-from api.db.services.file2document_service import File2DocumentService
-from api.db.services.file_service import FileService
-from api.db.services.knowledgebase_service import KnowledgebaseService
-from api.db.services.task_service import queue_tasks, TaskService
 from api.db.services.user_service import UserTenantService
-from common.misc_utils import get_uuid
-from common.constants import RetCode, VALID_TASK_STATUS, LLMType, ParserType, FileSource
 from api.utils.api_utils import server_error_response, get_data_error_result, get_json_result, validate_request, \
     generate_confirmation_token
-
-from api.utils.file_utils import filename_type, thumbnail
-from rag.app.tag import label_question
-from rag.prompts.generator import keyword_extraction
 from common.time_utils import current_timestamp, datetime_format
 
-from api.db.services.canvas_service import UserCanvasService
-from agent.canvas import Canvas
-from functools import partial
-from pathlib import Path
-from common import settings
-
 
 @manager.route('/new_token', methods=['POST'])  # noqa: F821
 @login_required
@@ -138,758 +113,3 @@ def stats():
     except Exception as e:
         return server_error_response(e)
 
-
-@manager.route('/new_conversation', methods=['GET'])  # noqa: F821
-def set_conversation():
-    token = request.headers.get('Authorization').split()[1]
-    objs = APIToken.query(token=token)
-    if not objs:
-        return get_json_result(
-            data=False, message='Authentication error: API key is invalid!"', code=RetCode.AUTHENTICATION_ERROR)
-    try:
-        if objs[0].source == "agent":
-            e, cvs = UserCanvasService.get_by_id(objs[0].dialog_id)
-            if not e:
-                return server_error_response("canvas not found.")
-            if not isinstance(cvs.dsl, str):
-                cvs.dsl = json.dumps(cvs.dsl, ensure_ascii=False)
-            canvas = Canvas(cvs.dsl, objs[0].tenant_id)
-            conv = {
-                "id": get_uuid(),
-                "dialog_id": cvs.id,
-                "user_id": request.args.get("user_id", ""),
-                "message": [{"role": "assistant", "content": canvas.get_prologue()}],
-                "source": "agent"
-            }
-            API4ConversationService.save(**conv)
-            return get_json_result(data=conv)
-        else:
-            e, dia = DialogService.get_by_id(objs[0].dialog_id)
-            if not e:
-                return get_data_error_result(message="Dialog not found")
-            conv = {
-                "id": get_uuid(),
-                "dialog_id": dia.id,
-                "user_id": request.args.get("user_id", ""),
-                "message": [{"role": "assistant", "content": dia.prompt_config["prologue"]}]
-            }
-            API4ConversationService.save(**conv)
-            return get_json_result(data=conv)
-    except Exception as e:
-        return server_error_response(e)
-
-
-@manager.route('/completion', methods=['POST'])  # noqa: F821
-@validate_request("conversation_id", "messages")
-def completion():
-    token = request.headers.get('Authorization').split()[1]
-    objs = APIToken.query(token=token)
-    if not objs:
-        return get_json_result(
-            data=False, message='Authentication error: API key is invalid!"', code=RetCode.AUTHENTICATION_ERROR)
-    req = request.json
-    e, conv = API4ConversationService.get_by_id(req["conversation_id"])
-    if not e:
-        return get_data_error_result(message="Conversation not found!")
-    if "quote" not in req:
-        req["quote"] = False
-
-    msg = []
-    for m in req["messages"]:
-        if m["role"] == "system":
-            continue
-        if m["role"] == "assistant" and not msg:
-            continue
-        msg.append(m)
-    if not msg[-1].get("id"):
-        msg[-1]["id"] = get_uuid()
-    message_id = msg[-1]["id"]
-
-    def fillin_conv(ans):
-        nonlocal conv, message_id
-        if not conv.reference:
-            conv.reference.append(ans["reference"])
-        else:
-            conv.reference[-1] = ans["reference"]
-        conv.message[-1] = {"role": "assistant", "content": ans["answer"], "id": message_id}
-        ans["id"] = message_id
-
-    def rename_field(ans):
-        reference = ans['reference']
-        if not isinstance(reference, dict):
-            return
-        for chunk_i in reference.get('chunks', []):
-            if 'docnm_kwd' in chunk_i:
-                chunk_i['doc_name'] = chunk_i['docnm_kwd']
-                chunk_i.pop('docnm_kwd')
-
-    try:
-        if conv.source == "agent":
-            stream = req.get("stream", True)
-            conv.message.append(msg[-1])
-            e, cvs = UserCanvasService.get_by_id(conv.dialog_id)
-            if not e:
-                return server_error_response("canvas not found.")
-            del req["conversation_id"]
-            del req["messages"]
-
-            if not isinstance(cvs.dsl, str):
-                cvs.dsl = json.dumps(cvs.dsl, ensure_ascii=False)
-
-            if not conv.reference:
-                conv.reference = []
-            conv.message.append({"role": "assistant", "content": "", "id": message_id})
-            conv.reference.append({"chunks": [], "doc_aggs": []})
-
-            final_ans = {"reference": [], "content": ""}
-            canvas = Canvas(cvs.dsl, objs[0].tenant_id)
-
-            canvas.messages.append(msg[-1])
-            canvas.add_user_input(msg[-1]["content"])
-            answer = canvas.run(stream=stream)
-
-            assert answer is not None, "Nothing. Is it over?"
-
-            if stream:
-                assert isinstance(answer, partial), "Nothing. Is it over?"
-
-                def sse():
-                    nonlocal answer, cvs, conv
-                    try:
-                        for ans in answer():
-                            for k in ans.keys():
-                                final_ans[k] = ans[k]
-                            ans = {"answer": ans["content"], "reference": ans.get("reference", [])}
-                            fillin_conv(ans)
-                            rename_field(ans)
-                            yield "data:" + json.dumps({"code": 0, "message": "", "data": ans},
-                                                       ensure_ascii=False) + "\n\n"
-
-                        canvas.messages.append({"role": "assistant", "content": final_ans["content"], "id": message_id})
-                        canvas.history.append(("assistant", final_ans["content"]))
-                        if final_ans.get("reference"):
-                            canvas.reference.append(final_ans["reference"])
-                        cvs.dsl = json.loads(str(canvas))
-                        API4ConversationService.append_message(conv.id, conv.to_dict())
-                    except Exception as e:
-                        yield "data:" + json.dumps({"code": 500, "message": str(e),
-                                                    "data": {"answer": "**ERROR**: " + str(e), "reference": []}},
-                                                   ensure_ascii=False) + "\n\n"
-                    yield "data:" + json.dumps({"code": 0, "message": "", "data": True}, ensure_ascii=False) + "\n\n"
-
-                resp = Response(sse(), mimetype="text/event-stream")
-                resp.headers.add_header("Cache-control", "no-cache")
-                resp.headers.add_header("Connection", "keep-alive")
-                resp.headers.add_header("X-Accel-Buffering", "no")
-                resp.headers.add_header("Content-Type", "text/event-stream; charset=utf-8")
-                return resp
-
-            final_ans["content"] = "\n".join(answer["content"]) if "content" in answer else ""
-            canvas.messages.append({"role": "assistant", "content": final_ans["content"], "id": message_id})
-            if final_ans.get("reference"):
-                canvas.reference.append(final_ans["reference"])
-            cvs.dsl = json.loads(str(canvas))
-
-            result = {"answer": final_ans["content"], "reference": final_ans.get("reference", [])}
-            fillin_conv(result)
-            API4ConversationService.append_message(conv.id, conv.to_dict())
-            rename_field(result)
-            return get_json_result(data=result)
-
-        # ******************For dialog******************
-        conv.message.append(msg[-1])
-        e, dia = DialogService.get_by_id(conv.dialog_id)
-        if not e:
-            return get_data_error_result(message="Dialog not found!")
-        del req["conversation_id"]
-        del req["messages"]
-
-        if not conv.reference:
-            conv.reference = []
-        conv.message.append({"role": "assistant", "content": "", "id": message_id})
-        conv.reference.append({"chunks": [], "doc_aggs": []})
-
-        def stream():
-            nonlocal dia, msg, req, conv
-            try:
-                for ans in chat(dia, msg, True, **req):
-                    fillin_conv(ans)
-                    rename_field(ans)
-                    yield "data:" + json.dumps({"code": 0, "message": "", "data": ans},
-                                               ensure_ascii=False) + "\n\n"
-                API4ConversationService.append_message(conv.id, conv.to_dict())
-            except Exception as e:
-                yield "data:" + json.dumps({"code": 500, "message": str(e),
-                                            "data": {"answer": "**ERROR**: " + str(e), "reference": []}},
-                                           ensure_ascii=False) + "\n\n"
-            yield "data:" + json.dumps({"code": 0, "message": "", "data": True}, ensure_ascii=False) + "\n\n"
-
-        if req.get("stream", True):
-            resp = Response(stream(), mimetype="text/event-stream")
-            resp.headers.add_header("Cache-control", "no-cache")
-            resp.headers.add_header("Connection", "keep-alive")
-            resp.headers.add_header("X-Accel-Buffering", "no")
-            resp.headers.add_header("Content-Type", "text/event-stream; charset=utf-8")
-            return resp
-
-        answer = None
-        for ans in chat(dia, msg, **req):
-            answer = ans
-            fillin_conv(ans)
-            API4ConversationService.append_message(conv.id, conv.to_dict())
-            break
-        rename_field(answer)
-        return get_json_result(data=answer)
-
-    except Exception as e:
-        return server_error_response(e)
-
-
-@manager.route('/conversation/<conversation_id>', methods=['GET'])  # noqa: F821
-# @login_required
-def get_conversation(conversation_id):
-    token = request.headers.get('Authorization').split()[1]
-    objs = APIToken.query(token=token)
-    if not objs:
-        return get_json_result(
-            data=False, message='Authentication error: API key is invalid!"', code=RetCode.AUTHENTICATION_ERROR)
-
-    try:
-        e, conv = API4ConversationService.get_by_id(conversation_id)
-        if not e:
-            return get_data_error_result(message="Conversation not found!")
-
-        conv = conv.to_dict()
-        if token != APIToken.query(dialog_id=conv['dialog_id'])[0].token:
-            return get_json_result(data=False, message='Authentication error: API key is invalid for this conversation_id!"',
-                                   code=RetCode.AUTHENTICATION_ERROR)
-
-        for referenct_i in conv['reference']:
-            if referenct_i is None or len(referenct_i) == 0:
-                continue
-            for chunk_i in referenct_i['chunks']:
-                if 'docnm_kwd' in chunk_i.keys():
-                    chunk_i['doc_name'] = chunk_i['docnm_kwd']
-                    chunk_i.pop('docnm_kwd')
-        return get_json_result(data=conv)
-    except Exception as e:
-        return server_error_response(e)
-
-
-@manager.route('/document/upload', methods=['POST'])  # noqa: F821
-@validate_request("kb_name")
-def upload():
-    token = request.headers.get('Authorization').split()[1]
-    objs = APIToken.query(token=token)
-    if not objs:
-        return get_json_result(
-            data=False, message='Authentication error: API key is invalid!"', code=RetCode.AUTHENTICATION_ERROR)
-
-    kb_name = request.form.get("kb_name").strip()
-    tenant_id = objs[0].tenant_id
-
-    try:
-        e, kb = KnowledgebaseService.get_by_name(kb_name, tenant_id)
-        if not e:
-            return get_data_error_result(
-                message="Can't find this knowledgebase!")
-        kb_id = kb.id
-    except Exception as e:
-        return server_error_response(e)
-
-    if 'file' not in request.files:
-        return get_json_result(
-            data=False, message='No file part!', code=RetCode.ARGUMENT_ERROR)
-
-    file = request.files['file']
-    if file.filename == '':
-        return get_json_result(
-            data=False, message='No file selected!', code=RetCode.ARGUMENT_ERROR)
-
-    root_folder = FileService.get_root_folder(tenant_id)
-    pf_id = root_folder["id"]
-    FileService.init_knowledgebase_docs(pf_id, tenant_id)
-    kb_root_folder = FileService.get_kb_folder(tenant_id)
-    kb_folder = FileService.new_a_file_from_kb(kb.tenant_id, kb.name, kb_root_folder["id"])
-
-    try:
-        if DocumentService.get_doc_count(kb.tenant_id) >= int(os.environ.get('MAX_FILE_NUM_PER_USER', 8192)):
-            return get_data_error_result(
-                message="Exceed the maximum file number of a free user!")
-
-        filename = duplicate_name(
-            DocumentService.query,
-            name=file.filename,
-            kb_id=kb_id)
-        filetype = filename_type(filename)
-        if not filetype:
-            return get_data_error_result(
-                message="This type of file has not been supported yet!")
-
-        location = filename
-        while settings.STORAGE_IMPL.obj_exist(kb_id, location):
-            location += "_"
-        blob = request.files['file'].read()
-        settings.STORAGE_IMPL.put(kb_id, location, blob)
-        doc = {
-            "id": get_uuid(),
-            "kb_id": kb.id,
-            "parser_id": kb.parser_id,
-            "parser_config": kb.parser_config,
-            "created_by": kb.tenant_id,
-            "type": filetype,
-            "name": filename,
-            "location": location,
-            "size": len(blob),
-            "thumbnail": thumbnail(filename, blob),
-            "suffix": Path(filename).suffix.lstrip("."),
-        }
-
-        form_data = request.form
-        if "parser_id" in form_data.keys():
-            if request.form.get("parser_id").strip() in list(vars(ParserType).values())[1:-3]:
-                doc["parser_id"] = request.form.get("parser_id").strip()
-        if doc["type"] == FileType.VISUAL:
-            doc["parser_id"] = ParserType.PICTURE.value
-        if doc["type"] == FileType.AURAL:
-            doc["parser_id"] = ParserType.AUDIO.value
-        if re.search(r"\.(ppt|pptx|pages)$", filename):
-            doc["parser_id"] = ParserType.PRESENTATION.value
-        if re.search(r"\.(eml)$", filename):
-            doc["parser_id"] = ParserType.EMAIL.value
-
-        doc_result = DocumentService.insert(doc)
-        FileService.add_file_from_kb(doc, kb_folder["id"], kb.tenant_id)
-    except Exception as e:
-        return server_error_response(e)
-
-    if "run" in form_data.keys():
-        if request.form.get("run").strip() == "1":
-            try:
-                info = {"run": 1, "progress": 0, "progress_msg": "", "chunk_num": 0, "token_num": 0}
-                DocumentService.update_by_id(doc["id"], info)
-                # if str(req["run"]) == TaskStatus.CANCEL.value:
-                tenant_id = DocumentService.get_tenant_id(doc["id"])
-                if not tenant_id:
-                    return get_data_error_result(message="Tenant not found!")
-
-                # e, doc = DocumentService.get_by_id(doc["id"])
-                TaskService.filter_delete([Task.doc_id == doc["id"]])
-                e, doc = DocumentService.get_by_id(doc["id"])
-                doc = doc.to_dict()
-                doc["tenant_id"] = tenant_id
-                bucket, name = File2DocumentService.get_storage_address(doc_id=doc["id"])
-                queue_tasks(doc, bucket, name, 0)
-            except Exception as e:
-                return server_error_response(e)
-
-    return get_json_result(data=doc_result.to_json())
-
-
-@manager.route('/document/upload_and_parse', methods=['POST'])  # noqa: F821
-@validate_request("conversation_id")
-def upload_parse():
-    token = request.headers.get('Authorization').split()[1]
-    objs = APIToken.query(token=token)
-    if not objs:
-        return get_json_result(
-            data=False, message='Authentication error: API key is invalid!"', code=RetCode.AUTHENTICATION_ERROR)
-
-    if 'file' not in request.files:
-        return get_json_result(
-            data=False, message='No file part!', code=RetCode.ARGUMENT_ERROR)
-
-    file_objs = request.files.getlist('file')
-    for file_obj in file_objs:
-        if file_obj.filename == '':
-            return get_json_result(
-                data=False, message='No file selected!', code=RetCode.ARGUMENT_ERROR)
-
-    doc_ids = doc_upload_and_parse(request.form.get("conversation_id"), file_objs, objs[0].tenant_id)
-    return get_json_result(data=doc_ids)
-
-
-@manager.route('/list_chunks', methods=['POST'])  # noqa: F821
-# @login_required
-def list_chunks():
-    token = request.headers.get('Authorization').split()[1]
-    objs = APIToken.query(token=token)
-    if not objs:
-        return get_json_result(
-            data=False, message='Authentication error: API key is invalid!"', code=RetCode.AUTHENTICATION_ERROR)
-
-    req = request.json
-
-    try:
-        if "doc_name" in req.keys():
-            tenant_id = DocumentService.get_tenant_id_by_name(req['doc_name'])
-            doc_id = DocumentService.get_doc_id_by_doc_name(req['doc_name'])
-
-        elif "doc_id" in req.keys():
-            tenant_id = DocumentService.get_tenant_id(req['doc_id'])
-            doc_id = req['doc_id']
-        else:
-            return get_json_result(
-                data=False, message="Can't find doc_name or doc_id"
-            )
-        kb_ids = KnowledgebaseService.get_kb_ids(tenant_id)
-
-        res = settings.retriever.chunk_list(doc_id, tenant_id, kb_ids)
-        res = [
-            {
-                "content": res_item["content_with_weight"],
-                "doc_name": res_item["docnm_kwd"],
-                "image_id": res_item["img_id"]
-            } for res_item in res
-        ]
-
-    except Exception as e:
-        return server_error_response(e)
-
-    return get_json_result(data=res)
-
-@manager.route('/get_chunk/<chunk_id>', methods=['GET'])  # noqa: F821
-# @login_required
-def get_chunk(chunk_id):
-    from rag.nlp import search
-    token = request.headers.get('Authorization').split()[1]
-    objs = APIToken.query(token=token)
-    if not objs:
-        return get_json_result(
-            data=False, message='Authentication error: API key is invalid!"', code=RetCode.AUTHENTICATION_ERROR)
-    try:
-        tenant_id = objs[0].tenant_id
-        kb_ids = KnowledgebaseService.get_kb_ids(tenant_id)
-        chunk = settings.docStoreConn.get(chunk_id, search.index_name(tenant_id), kb_ids)
-        if chunk is None:
-            return server_error_response(Exception("Chunk not found"))
-        k = []
-        for n in chunk.keys():
-            if re.search(r"(_vec$|_sm_|_tks|_ltks)", n):
-                k.append(n)
-        for n in k:
-            del chunk[n]
-
-        return get_json_result(data=chunk)
-    except Exception as e:
-        return server_error_response(e)
-
-@manager.route('/list_kb_docs', methods=['POST'])  # noqa: F821
-# @login_required
-def list_kb_docs():
-    token = request.headers.get('Authorization').split()[1]
-    objs = APIToken.query(token=token)
-    if not objs:
-        return get_json_result(
-            data=False, message='Authentication error: API key is invalid!"', code=RetCode.AUTHENTICATION_ERROR)
-
-    req = request.json
-    tenant_id = objs[0].tenant_id
-    kb_name = req.get("kb_name", "").strip()
-
-    try:
-        e, kb = KnowledgebaseService.get_by_name(kb_name, tenant_id)
-        if not e:
-            return get_data_error_result(
-                message="Can't find this knowledgebase!")
-        kb_id = kb.id
-
-    except Exception as e:
-        return server_error_response(e)
-
-    page_number = int(req.get("page", 1))
-    items_per_page = int(req.get("page_size", 15))
-    orderby = req.get("orderby", "create_time")
-    desc = req.get("desc", True)
-    keywords = req.get("keywords", "")
-    status = req.get("status", [])
-    if status:
-        invalid_status = {s for s in status if s not in VALID_TASK_STATUS}
-        if invalid_status:
-            return get_data_error_result(
-                message=f"Invalid filter status conditions: {', '.join(invalid_status)}"
-            )
-    types = req.get("types", [])
-    if types:
-        invalid_types = {t for t in types if t not in VALID_FILE_TYPES}
-        if invalid_types:
-            return get_data_error_result(
-                message=f"Invalid filter conditions: {', '.join(invalid_types)} type{'s' if len(invalid_types) > 1 else ''}"
-            )
-    try:
-        docs, tol = DocumentService.get_by_kb_id(
-            kb_id, page_number, items_per_page, orderby, desc, keywords, status, types)
-        docs = [{"doc_id": doc['id'], "doc_name": doc['name']} for doc in docs]
-
-        return get_json_result(data={"total": tol, "docs": docs})
-
-    except Exception as e:
-        return server_error_response(e)
-
-
-@manager.route('/document/infos', methods=['POST'])  # noqa: F821
-@validate_request("doc_ids")
-def docinfos():
-    token = request.headers.get('Authorization').split()[1]
-    objs = APIToken.query(token=token)
-    if not objs:
-        return get_json_result(
-            data=False, message='Authentication error: API key is invalid!"', code=RetCode.AUTHENTICATION_ERROR)
-    req = request.json
-    doc_ids = req["doc_ids"]
-    docs = DocumentService.get_by_ids(doc_ids)
-    return get_json_result(data=list(docs.dicts()))
-
-
-@manager.route('/document', methods=['DELETE'])  # noqa: F821
-# @login_required
-def document_rm():
-    token = request.headers.get('Authorization').split()[1]
-    objs = APIToken.query(token=token)
-    if not objs:
-        return get_json_result(
-            data=False, message='Authentication error: API key is invalid!"', code=RetCode.AUTHENTICATION_ERROR)
-
-    tenant_id = objs[0].tenant_id
-    req = request.json
-    try:
-        doc_ids = DocumentService.get_doc_ids_by_doc_names(req.get("doc_names", []))
-        for doc_id in req.get("doc_ids", []):
-            if doc_id not in doc_ids:
-                doc_ids.append(doc_id)
-
-        if not doc_ids:
-            return get_json_result(
-                data=False, message="Can't find doc_names or doc_ids"
-            )
-
-    except Exception as e:
-        return server_error_response(e)
-
-    root_folder = FileService.get_root_folder(tenant_id)
-    pf_id = root_folder["id"]
-    FileService.init_knowledgebase_docs(pf_id, tenant_id)
-
-    errors = ""
-    docs = DocumentService.get_by_ids(doc_ids)
-    doc_dic = {}
-    for doc in docs:
-        doc_dic[doc.id] = doc
-
-    for doc_id in doc_ids:
-        try:
-            if doc_id not in doc_dic:
-                return get_data_error_result(message="Document not found!")
-            doc = doc_dic[doc_id]
-            tenant_id = DocumentService.get_tenant_id(doc_id)
-            if not tenant_id:
-                return get_data_error_result(message="Tenant not found!")
-
-            b, n = File2DocumentService.get_storage_address(doc_id=doc_id)
-
-            if not DocumentService.remove_document(doc, tenant_id):
-                return get_data_error_result(
-                    message="Database error (Document removal)!")
-
-            f2d = File2DocumentService.get_by_document_id(doc_id)
-            FileService.filter_delete([File.source_type == FileSource.KNOWLEDGEBASE, File.id == f2d[0].file_id])
-            File2DocumentService.delete_by_document_id(doc_id)
-
-            settings.STORAGE_IMPL.rm(b, n)
-        except Exception as e:
-            errors += str(e)
-
-    if errors:
-        return get_json_result(data=False, message=errors, code=RetCode.SERVER_ERROR)
-
-    return get_json_result(data=True)
-
-
-@manager.route('/completion_aibotk', methods=['POST'])  # noqa: F821
-@validate_request("Authorization", "conversation_id", "word")
-def completion_faq():
-    import base64
-    req = request.json
-
-    token = req["Authorization"]
-    objs = APIToken.query(token=token)
-    if not objs:
-        return get_json_result(
-            data=False, message='Authentication error: API key is invalid!"', code=RetCode.AUTHENTICATION_ERROR)
-
-    e, conv = API4ConversationService.get_by_id(req["conversation_id"])
-    if not e:
-        return get_data_error_result(message="Conversation not found!")
-    if "quote" not in req:
-        req["quote"] = True
-
-    msg = [{"role": "user", "content": req["word"]}]
-    if not msg[-1].get("id"):
-        msg[-1]["id"] = get_uuid()
-    message_id = msg[-1]["id"]
-
-    def fillin_conv(ans):
-        nonlocal conv, message_id
-        if not conv.reference:
-            conv.reference.append(ans["reference"])
-        else:
-            conv.reference[-1] = ans["reference"]
-        conv.message[-1] = {"role": "assistant", "content": ans["answer"], "id": message_id}
-        ans["id"] = message_id
-
-    try:
-        if conv.source == "agent":
-            conv.message.append(msg[-1])
-            e, cvs = UserCanvasService.get_by_id(conv.dialog_id)
-            if not e:
-                return server_error_response("canvas not found.")
-
-            if not isinstance(cvs.dsl, str):
-                cvs.dsl = json.dumps(cvs.dsl, ensure_ascii=False)
-
-            if not conv.reference:
-                conv.reference = []
-            conv.message.append({"role": "assistant", "content": "", "id": message_id})
-            conv.reference.append({"chunks": [], "doc_aggs": []})
-
-            final_ans = {"reference": [], "doc_aggs": []}
-            canvas = Canvas(cvs.dsl, objs[0].tenant_id)
-
-            canvas.messages.append(msg[-1])
-            canvas.add_user_input(msg[-1]["content"])
-            answer = canvas.run(stream=False)
-
-            assert answer is not None, "Nothing. Is it over?"
-
-            data_type_picture = {
-                "type": 3,
-                "url": "base64 content"
-            }
-            data = [
-                {
-                    "type": 1,
-                    "content": ""
-                }
-            ]
-            final_ans["content"] = "\n".join(answer["content"]) if "content" in answer else ""
-            canvas.messages.append({"role": "assistant", "content": final_ans["content"], "id": message_id})
-            if final_ans.get("reference"):
-                canvas.reference.append(final_ans["reference"])
-            cvs.dsl = json.loads(str(canvas))
-
-            ans = {"answer": final_ans["content"], "reference": final_ans.get("reference", [])}
-            data[0]["content"] += re.sub(r'##\d\$\$', '', ans["answer"])
-            fillin_conv(ans)
-            API4ConversationService.append_message(conv.id, conv.to_dict())
-
-            chunk_idxs = [int(match[2]) for match in re.findall(r'##\d\$\$', ans["answer"])]
-            for chunk_idx in chunk_idxs[:1]:
-                if ans["reference"]["chunks"][chunk_idx]["img_id"]:
-                    try:
-                        bkt, nm = ans["reference"]["chunks"][chunk_idx]["img_id"].split("-")
-                        response = settings.STORAGE_IMPL.get(bkt, nm)
-                        data_type_picture["url"] = base64.b64encode(response).decode('utf-8')
-                        data.append(data_type_picture)
-                        break
-                    except Exception as e:
-                        return server_error_response(e)
-
-            response = {"code": 200, "msg": "success", "data": data}
-            return response
-
-        # ******************For dialog******************
-        conv.message.append(msg[-1])
-        e, dia = DialogService.get_by_id(conv.dialog_id)
-        if not e:
-            return get_data_error_result(message="Dialog not found!")
-        del req["conversation_id"]
-
-        if not conv.reference:
-            conv.reference = []
-        conv.message.append({"role": "assistant", "content": "", "id": message_id})
-        conv.reference.append({"chunks": [], "doc_aggs": []})
-
-        data_type_picture = {
-            "type": 3,
-            "url": "base64 content"
-        }
-        data = [
-            {
-                "type": 1,
-                "content": ""
-            }
-        ]
-        ans = ""
-        for a in chat(dia, msg, stream=False, **req):
-            ans = a
-            break
-        data[0]["content"] += re.sub(r'##\d\$\$', '', ans["answer"])
-        fillin_conv(ans)
-        API4ConversationService.append_message(conv.id, conv.to_dict())
-
-        chunk_idxs = [int(match[2]) for match in re.findall(r'##\d\$\$', ans["answer"])]
-        for chunk_idx in chunk_idxs[:1]:
-            if ans["reference"]["chunks"][chunk_idx]["img_id"]:
-                try:
-                    bkt, nm = ans["reference"]["chunks"][chunk_idx]["img_id"].split("-")
-                    response = settings.STORAGE_IMPL.get(bkt, nm)
-                    data_type_picture["url"] = base64.b64encode(response).decode('utf-8')
-                    data.append(data_type_picture)
-                    break
-                except Exception as e:
-                    return server_error_response(e)
-
-        response = {"code": 200, "msg": "success", "data": data}
-        return response
-
-    except Exception as e:
-        return server_error_response(e)
-
-
-@manager.route('/retrieval', methods=['POST'])  # noqa: F821
-@validate_request("kb_id", "question")
-def retrieval():
-    token = request.headers.get('Authorization').split()[1]
-    objs = APIToken.query(token=token)
-    if not objs:
-        return get_json_result(
-            data=False, message='Authentication error: API key is invalid!"', code=RetCode.AUTHENTICATION_ERROR)
-
-    req = request.json
-    kb_ids = req.get("kb_id", [])
-    doc_ids = req.get("doc_ids", [])
-    question = req.get("question")
-    page = int(req.get("page", 1))
-    size = int(req.get("page_size", 30))
-    similarity_threshold = float(req.get("similarity_threshold", 0.2))
-    vector_similarity_weight = float(req.get("vector_similarity_weight", 0.3))
-    top = int(req.get("top_k", 1024))
-    highlight = bool(req.get("highlight", False))
-
-    try:
-        kbs = KnowledgebaseService.get_by_ids(kb_ids)
-        embd_nms = list(set([kb.embd_id for kb in kbs]))
-        if len(embd_nms) != 1:
-            return get_json_result(
-                data=False, message='Knowledge bases use different embedding models or does not exist."',
-                code=RetCode.AUTHENTICATION_ERROR)
-
-        embd_mdl = LLMBundle(kbs[0].tenant_id, LLMType.EMBEDDING, llm_name=kbs[0].embd_id)
-        rerank_mdl = None
-        if req.get("rerank_id"):
-            rerank_mdl = LLMBundle(kbs[0].tenant_id, LLMType.RERANK, llm_name=req["rerank_id"])
-        if req.get("keyword", False):
-            chat_mdl = LLMBundle(kbs[0].tenant_id, LLMType.CHAT)
-            question += keyword_extraction(chat_mdl, question)
-        ranks = settings.retriever.retrieval(question, embd_mdl, kbs[0].tenant_id, kb_ids, page, size,
-                                               similarity_threshold, vector_similarity_weight, top,
-                                               doc_ids, rerank_mdl=rerank_mdl, highlight= highlight,
-                                               rank_feature=label_question(question, kbs))
-        for c in ranks["chunks"]:
-            c.pop("vector", None)
-        return get_json_result(data=ranks)
-    except Exception as e:
-        if str(e).find("not_found") > 0:
-            return get_json_result(data=False, message='No chunk found! Check the chunk status please!',
-                                   code=RetCode.DATA_ERROR)
-        return server_error_response(e)

api/apps/sdk/doc.py

@@ -93,6 +93,10 @@ def upload(dataset_id, tenant_id):
         type: file
         required: true
         description: Document files to upload.
+      - in: formData
+        name: parent_path
+        type: string
+        description: Optional nested path under the parent folder. Uses '/' separators.
     responses:
       200:
         description: Successfully uploaded documents.
@@ -151,7 +155,7 @@ def upload(dataset_id, tenant_id):
     e, kb = KnowledgebaseService.get_by_id(dataset_id)
     if not e:
         raise LookupError(f"Can't find the dataset with ID {dataset_id}!")
-    err, files = FileService.upload_document(kb, file_objs, tenant_id)
+    err, files = FileService.upload_document(kb, file_objs, tenant_id, parent_path=request.form.get("parent_path"))
     if err:
         return get_result(message="\n".join(err), code=RetCode.SERVER_ERROR)
     # rename key's name

api/db/services/connector_service.py

@@ -242,7 +242,7 @@ class Connector2KbService(CommonService):
                 "id": get_uuid(),
                 "connector_id": conn_id,
                 "kb_id": kb_id,
-                "auto_parse": conn.get("auto_parse", "1")   
+                "auto_parse": conn.get("auto_parse", "1")
             })
             SyncLogsService.schedule(conn_id, kb_id, reindex=True)
 

api/db/services/file_service.py

@@ -31,7 +31,7 @@ from common.misc_utils import get_uuid
 from common.constants import TaskStatus, FileSource, ParserType
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.task_service import TaskService
-from api.utils.file_utils import filename_type, read_potential_broken_pdf, thumbnail_img
+from api.utils.file_utils import filename_type, read_potential_broken_pdf, thumbnail_img, sanitize_path
 from rag.llm.cv_model import GptV4
 from common import settings
 
@@ -329,7 +329,7 @@ class FileService(CommonService):
         current_id = start_id
         while current_id:
             e, file = cls.get_by_id(current_id)
-            if file.parent_id != file.id and e:
+            if e and file.parent_id != file.id:
                 parent_folders.append(file)
                 current_id = file.parent_id
             else:
@@ -423,13 +423,15 @@ class FileService(CommonService):
 
     @classmethod
     @DB.connection_context()
-    def upload_document(self, kb, file_objs, user_id, src="local"):
+    def upload_document(self, kb, file_objs, user_id, src="local", parent_path: str | None = None):
         root_folder = self.get_root_folder(user_id)
         pf_id = root_folder["id"]
         self.init_knowledgebase_docs(pf_id, user_id)
         kb_root_folder = self.get_kb_folder(user_id)
         kb_folder = self.new_a_file_from_kb(kb.tenant_id, kb.name, kb_root_folder["id"])
 
+        safe_parent_path = sanitize_path(parent_path)
+
         err, files = [], []
         for file in file_objs:
             try:
@@ -439,7 +441,7 @@ class FileService(CommonService):
                 if filetype == FileType.OTHER.value:
                     raise RuntimeError("This type of file has not been supported yet!")
 
-                location = filename
+                location = filename if not safe_parent_path else f"{safe_parent_path}/{filename}"
                 while settings.STORAGE_IMPL.obj_exist(kb.id, location):
                     location += "_"
 

api/utils/file_utils.py

@@ -164,3 +164,23 @@ def read_potential_broken_pdf(blob):
         return repaired
 
     return blob
+
+
+def sanitize_path(raw_path: str | None) -> str:
+    """Normalize and sanitize a user-provided path segment.
+
+    - Converts backslashes to forward slashes
+    - Strips leading/trailing slashes
+    - Removes '.' and '..' segments
+    - Restricts characters to A-Za-z0-9, underscore, dash, and '/'
+    """
+    if not raw_path:
+        return ""
+    backslash_re = re.compile(r"[\\]+")
+    unsafe_re = re.compile(r"[^A-Za-z0-9_\-/]")
+    normalized = backslash_re.sub("/", raw_path)
+    normalized = normalized.strip("/")
+    parts = [seg for seg in normalized.split("/") if seg and seg not in (".", "..")]
+    sanitized = "/".join(parts)
+    sanitized = unsafe_re.sub("", sanitized)
+    return sanitized

deepdoc/parser/docling_parser.py

@@ -186,9 +186,6 @@ class DoclingParser(RAGFlowPdfParser):
                 yield (DoclingContentType.EQUATION.value, text, bbox)
 
     def _transfer_to_sections(self, doc) -> list[tuple[str, str]]:
-        """
-        和 MinerUParser 保持一致：返回 [(section_text, line_tag), ...]
-        """
         sections: list[tuple[str, str]] = []
         for typ, payload, bbox in self._iter_doc_items(doc):
             if typ == DoclingContentType.TEXT.value:

deepdoc/parser/figure_parser.py

@@ -34,6 +34,7 @@ def vision_figure_parser_figure_data_wrapper(figures_data_without_positions):
         if isinstance(figure_data[1], Image.Image)
     ]
 
+
 def vision_figure_parser_docx_wrapper(sections,tbls,callback=None,**kwargs):
     try:
         vision_model = LLMBundle(kwargs["tenant_id"], LLMType.IMAGE2TEXT)
@@ -50,7 +51,8 @@ def vision_figure_parser_docx_wrapper(sections,tbls,callback=None,**kwargs):
             callback(0.8, f"Visual model error: {e}. Skipping figure parsing enhancement.")
     return tbls
 
-def vision_figure_parser_pdf_wrapper(tbls,callback=None,**kwargs):
+
+def vision_figure_parser_pdf_wrapper(tbls, callback=None, **kwargs):
     try:
         vision_model = LLMBundle(kwargs["tenant_id"], LLMType.IMAGE2TEXT)
         callback(0.7, "Visual model detected. Attempting to enhance figure extraction...")
@@ -72,6 +74,7 @@ def vision_figure_parser_pdf_wrapper(tbls,callback=None,**kwargs):
             callback(0.8, f"Visual model error: {e}. Skipping figure parsing enhancement.")
     return tbls
 
+
 shared_executor = ThreadPoolExecutor(max_workers=10)
 
 

deepdoc/vision/ocr.py

@@ -117,7 +117,6 @@ def load_model(model_dir, nm, device_id: int | None = None):
             providers=['CUDAExecutionProvider'],
             provider_options=[cuda_provider_options]
             )
-        run_options.add_run_config_entry("memory.enable_memory_arena_shrinkage", "gpu:" + str(provider_device_id))
         logging.info(f"load_model {model_file_path} uses GPU (device {provider_device_id}, gpu_mem_limit={cuda_provider_options['gpu_mem_limit']}, arena_strategy={arena_strategy})")
     else:
         sess = ort.InferenceSession(

web/src/components/avatar-upload.tsx

@@ -5,12 +5,14 @@ import {
   forwardRef,
   useCallback,
   useEffect,
+  useRef,
   useState,
 } from 'react';
 import { useTranslation } from 'react-i18next';
 import { Avatar, AvatarFallback, AvatarImage } from './ui/avatar';
 import { Button } from './ui/button';
 import { Input } from './ui/input';
+import { Modal } from './ui/modal/modal';
 
 type AvatarUploadProps = {
   value?: string;
@@ -22,14 +24,24 @@ export const AvatarUpload = forwardRef<HTMLInputElement, AvatarUploadProps>(
   function AvatarUpload({ value, onChange, tips }, ref) {
     const { t } = useTranslation();
     const [avatarBase64Str, setAvatarBase64Str] = useState(''); // Avatar Image base64
+    const [isCropModalOpen, setIsCropModalOpen] = useState(false);
+    const [imageToCrop, setImageToCrop] = useState<string | null>(null);
+    const [cropArea, setCropArea] = useState({ x: 0, y: 0, size: 200 });
+    const imageRef = useRef<HTMLImageElement>(null);
+    const canvasRef = useRef<HTMLCanvasElement>(null);
+    const containerRef = useRef<HTMLDivElement>(null);
+    const isDraggingRef = useRef(false);
+    const dragStartRef = useRef({ x: 0, y: 0 });
+    const [imageScale, setImageScale] = useState(1);
+    const [imageOffset, setImageOffset] = useState({ x: 0, y: 0 });
 
     const handleChange: ChangeEventHandler<HTMLInputElement> = useCallback(
       async (ev) => {
         const file = ev.target?.files?.[0];
         if (/\.(jpg|jpeg|png|webp|bmp)$/i.test(file?.name ?? '')) {
-          const str = await transformFile2Base64(file!);
-          setAvatarBase64Str(str);
-          onChange?.(str);
+          const str = await transformFile2Base64(file!, 1000);
+          setImageToCrop(str);
+          setIsCropModalOpen(true);
         }
         ev.target.value = '';
       },
@@ -41,17 +53,209 @@ export const AvatarUpload = forwardRef<HTMLInputElement, AvatarUploadProps>(
       onChange?.('');
     }, [onChange]);
 
+    const handleCrop = useCallback(() => {
+      if (!imageRef.current || !canvasRef.current) return;
+
+      const canvas = canvasRef.current;
+      const ctx = canvas.getContext('2d');
+      const image = imageRef.current;
+
+      if (!ctx) return;
+
+      // Set canvas size to 64x64 (avatar size)
+      canvas.width = 64;
+      canvas.height = 64;
+
+      // Draw cropped image on canvas
+      ctx.drawImage(
+        image,
+        cropArea.x,
+        cropArea.y,
+        cropArea.size,
+        cropArea.size,
+        0,
+        0,
+        64,
+        64,
+      );
+
+      // Convert to base64
+      const croppedImageBase64 = canvas.toDataURL('image/png');
+      setAvatarBase64Str(croppedImageBase64);
+      onChange?.(croppedImageBase64);
+      setIsCropModalOpen(false);
+    }, [cropArea, onChange]);
+
+    const handleCancelCrop = useCallback(() => {
+      setIsCropModalOpen(false);
+      setImageToCrop(null);
+    }, []);
+
+    const initCropArea = useCallback(() => {
+      if (!imageRef.current || !containerRef.current) return;
+
+      const image = imageRef.current;
+      const container = containerRef.current;
+
+      // Calculate image scale to fit container
+      const scale = Math.min(
+        container.clientWidth / image.width,
+        container.clientHeight / image.height,
+      );
+      setImageScale(scale);
+
+      // Calculate image offset to center it
+      const scaledWidth = image.width * scale;
+      const scaledHeight = image.height * scale;
+      const offsetX = (container.clientWidth - scaledWidth) / 2;
+      const offsetY = (container.clientHeight - scaledHeight) / 2;
+      setImageOffset({ x: offsetX, y: offsetY });
+
+      // Initialize crop area to center of image
+      const size = Math.min(scaledWidth, scaledHeight) * 0.8; // 80% of the smaller dimension
+      const x = (image.width - size / scale) / 2;
+      const y = (image.height - size / scale) / 2;
+
+      setCropArea({ x, y, size: size / scale });
+    }, []);
+
+    const handleMouseMove = useCallback(
+      (e: MouseEvent) => {
+        if (
+          !isDraggingRef.current ||
+          !imageRef.current ||
+          !containerRef.current
+        )
+          return;
+
+        const image = imageRef.current;
+        const container = containerRef.current;
+        const containerRect = container.getBoundingClientRect();
+
+        // Calculate mouse position relative to container
+        const mouseX = e.clientX - containerRect.left;
+        const mouseY = e.clientY - containerRect.top;
+
+        // Calculate mouse position relative to image
+        const imageX = (mouseX - imageOffset.x) / imageScale;
+        const imageY = (mouseY - imageOffset.y) / imageScale;
+
+        // Calculate new crop area position based on mouse movement
+        let newX = imageX - dragStartRef.current.x;
+        let newY = imageY - dragStartRef.current.y;
+
+        // Boundary checks
+        newX = Math.max(0, Math.min(newX, image.width - cropArea.size));
+        newY = Math.max(0, Math.min(newY, image.height - cropArea.size));
+
+        setCropArea((prev) => ({
+          ...prev,
+          x: newX,
+          y: newY,
+        }));
+      },
+      [cropArea.size, imageScale, imageOffset],
+    );
+
+    const handleMouseUp = useCallback(() => {
+      isDraggingRef.current = false;
+      document.removeEventListener('mousemove', handleMouseMove);
+      document.removeEventListener('mouseup', handleMouseUp);
+    }, [handleMouseMove]);
+
+    const handleMouseDown = useCallback(
+      (e: React.MouseEvent) => {
+        e.preventDefault();
+        e.stopPropagation();
+        isDraggingRef.current = true;
+        if (imageRef.current && containerRef.current) {
+          const container = containerRef.current;
+          const containerRect = container.getBoundingClientRect();
+
+          // Calculate mouse position relative to container
+          const mouseX = e.clientX - containerRect.left;
+          const mouseY = e.clientY - containerRect.top;
+
+          // Calculate mouse position relative to image
+          const imageX = (mouseX - imageOffset.x) / imageScale;
+          const imageY = (mouseY - imageOffset.y) / imageScale;
+
+          // Store the offset between mouse position and crop area position
+          dragStartRef.current = {
+            x: imageX - cropArea.x,
+            y: imageY - cropArea.y,
+          };
+        }
+        document.addEventListener('mousemove', handleMouseMove);
+        document.addEventListener('mouseup', handleMouseUp);
+      },
+      [cropArea, imageScale, imageOffset],
+    );
+
+    const handleWheel = useCallback((e: React.WheelEvent) => {
+      if (!imageRef.current) return;
+
+      e.preventDefault();
+      const image = imageRef.current;
+      const delta = e.deltaY > 0 ? 0.9 : 1.1; // Zoom factor
+
+      setCropArea((prev) => {
+        const newSize = Math.max(
+          20,
+          Math.min(prev.size * delta, Math.min(image.width, image.height)),
+        );
+
+        // Adjust position to keep crop area centered
+        const centerRatioX = (prev.x + prev.size / 2) / image.width;
+        const centerRatioY = (prev.y + prev.size / 2) / image.height;
+
+        const newX = centerRatioX * image.width - newSize / 2;
+        const newY = centerRatioY * image.height - newSize / 2;
+
+        // Boundary checks
+        const boundedX = Math.max(0, Math.min(newX, image.width - newSize));
+        const boundedY = Math.max(0, Math.min(newY, image.height - newSize));
+
+        return {
+          x: boundedX,
+          y: boundedY,
+          size: newSize,
+        };
+      });
+    }, []);
+
     useEffect(() => {
       if (value) {
         setAvatarBase64Str(value);
       }
     }, [value]);
 
+    useEffect(() => {
+      const container = containerRef.current;
+      setTimeout(() => {
+        console.log('container', container);
+        // initCropArea();
+        if (imageToCrop && container && isCropModalOpen) {
+          container.addEventListener(
+            'wheel',
+            handleWheel as unknown as EventListener,
+            { passive: false },
+          );
+          return () => {
+            container.removeEventListener(
+              'wheel',
+              handleWheel as unknown as EventListener,
+            );
+          };
+        }
+      }, 100);
+    }, [handleWheel, containerRef.current]);
+
     return (
       <div className="flex justify-start items-end space-x-2">
         <div className="relative group">
           {!avatarBase64Str ? (
-            <div className="w-[64px] h-[64px] grid place-content-center border border-dashed bg-bg-input	rounded-md">
+            <div className="w-[64px] h-[64px] grid place-content-center border border-dashed bg-bg-input rounded-md">
               <div className="flex flex-col items-center">
                 <Plus />
                 <p>{t('common.upload')}</p>
@@ -60,7 +264,7 @@ export const AvatarUpload = forwardRef<HTMLInputElement, AvatarUploadProps>(
           ) : (
             <div className="w-[64px] h-[64px] relative grid place-content-center">
               <Avatar className="w-[64px] h-[64px] rounded-md">
-                <AvatarImage className=" block" src={avatarBase64Str} alt="" />
+                <AvatarImage className="block" src={avatarBase64Str} alt="" />
                 <AvatarFallback></AvatarFallback>
               </Avatar>
               <div className="absolute inset-0 bg-[#000]/20 group-hover:bg-[#000]/60">
@@ -93,6 +297,79 @@ export const AvatarUpload = forwardRef<HTMLInputElement, AvatarUploadProps>(
         <div className="margin-1 text-text-secondary">
           {tips ?? t('knowledgeConfiguration.photoTip')}
         </div>
+
+        {/* Crop Modal */}
+        <Modal
+          open={isCropModalOpen}
+          onOpenChange={(open) => {
+            setIsCropModalOpen(open);
+            if (!open) {
+              setImageToCrop(null);
+            }
+          }}
+          title={t('setting.cropImage')}
+          size="small"
+          onCancel={handleCancelCrop}
+          onOk={handleCrop}
+          // footer={
+          //   <div className="flex justify-end space-x-2">
+          //     <Button variant="secondary" onClick={handleCancelCrop}>
+          //       {t('common.cancel')}
+          //     </Button>
+          //     <Button onClick={handleCrop}>{t('common.confirm')}</Button>
+          //   </div>
+          // }
+        >
+          <div className="flex flex-col items-center p-4">
+            {imageToCrop && (
+              <div className="w-full">
+                <div
+                  ref={containerRef}
+                  className="relative overflow-hidden border border-border rounded-md mx-auto bg-bg-card"
+                  style={{
+                    width: '300px',
+                    height: '300px',
+                    touchAction: 'none',
+                  }}
+                  // onWheel={handleWheel}
+                >
+                  <img
+                    ref={imageRef}
+                    src={imageToCrop}
+                    alt="To crop"
+                    className="absolute block"
+                    style={{
+                      transform: `scale(${imageScale})`,
+                      transformOrigin: 'top left',
+                      left: `${imageOffset.x}px`,
+                      top: `${imageOffset.y}px`,
+                    }}
+                    onLoad={initCropArea}
+                  />
+                  {imageRef.current && (
+                    <div
+                      className="absolute border-2 border-white border-dashed cursor-move"
+                      style={{
+                        left: `${imageOffset.x + cropArea.x * imageScale}px`,
+                        top: `${imageOffset.y + cropArea.y * imageScale}px`,
+                        width: `${cropArea.size * imageScale}px`,
+                        height: `${cropArea.size * imageScale}px`,
+                        boxShadow: '0 0 0 9999px rgba(0, 0, 0, 0.5)',
+                      }}
+                      onMouseDown={handleMouseDown}
+                    />
+                  )}
+                </div>
+                <div className="flex justify-center mt-4">
+                  <p className="text-sm text-text-secondary">
+                    {t('setting.cropTip')}
+                  </p>
+                </div>
+                <canvas ref={canvasRef} className="hidden" />
+              </div>
+            )}
+          </div>
+        </Modal>
       </div>
     );
   },

web/src/locales/en.ts

@@ -694,6 +694,9 @@ This auto-tagging feature enhances retrieval by adding another layer of domain-s
       tocEnhanceTip: ` During the parsing of the document, table of contents information was generated (see the 'Enable Table of Contents Extraction' option in the General method). This allows the large model to return table of contents items relevant to the user's query, thereby using these items to retrieve related chunks and apply weighting to these chunks during the sorting process. This approach is derived from mimicking the behavioral logic of how humans search for knowledge in books.`,
     },
     setting: {
+      cropTip:
+        'Drag the selection area to choose the cropping position of the image, and scroll to zoom in/out',
+      cropImage: 'Crop image',
       selectModelPlaceholder: 'Select model',
       configureModelTitle: 'Configure model',
       confluenceIsCloudTip:

web/src/locales/zh.ts

@@ -684,6 +684,8 @@ General：实体和关系提取提示来自 GitHub - microsoft/graphrag：基于
       tocEnhanceTip: `解析文档时生成了目录信息（见General方法的‘启用目录抽取’），让大模型返回和用户问题相关的目录项，从而利用目录项拿到相关chunk，对这些chunk在排序中进行加权。这种方法来源于模仿人类查询书本中知识的行为逻辑`,
     },
     setting: {
+      cropTip: '拖动选区可以选择要图片的裁剪位置，滚动可以放大/缩小选区',
+      cropImage: '剪裁图片',
       selectModelPlaceholder: '请选择模型',
       configureModelTitle: '配置模型',
       confluenceIsCloudTip:

web/src/utils/file-util.ts

@@ -2,7 +2,10 @@ import { FileMimeType } from '@/constants/common';
 import fileManagerService from '@/services/file-manager-service';
 import { UploadFile } from 'antd';
 
-export const transformFile2Base64 = (val: any): Promise<any> => {
+export const transformFile2Base64 = (
+  val: any,
+  imgSize?: number,
+): Promise<any> => {
   return new Promise((resolve, reject) => {
     const reader = new FileReader();
     reader.readAsDataURL(val);
@@ -19,7 +22,7 @@ export const transformFile2Base64 = (val: any): Promise<any> => {
         // Calculate compressed dimensions, set max width/height to 800px
         let width = img.width;
         let height = img.height;
-        const maxSize = 100;
+        const maxSize = imgSize ?? 100;
 
         if (width > height && width > maxSize) {
           height = (height * maxSize) / width;