Feat: chats completions API supports metadata filtering (#12023)

### What problem does this PR solve?

Chats completions API supports metadata filtering.

### Type of change

- [x] New Feature (non-breaking change which adds functionality)

api/apps/llm_app.py

@@ -157,7 +157,7 @@ async def add_llm():
     elif factory == "Bedrock":
         # For Bedrock, due to its special authentication method
         # Assemble bedrock_ak, bedrock_sk, bedrock_region
-        api_key = apikey_json(["bedrock_ak", "bedrock_sk", "bedrock_region"])
+        api_key = apikey_json(["auth_mode", "bedrock_ak", "bedrock_sk", "bedrock_region", "aws_role_arn"])
 
     elif factory == "LocalAI":
         llm_name += "___LocalAI"

api/apps/sdk/session.py

@@ -33,7 +33,7 @@ from api.db.services.dialog_service import DialogService, async_ask, async_chat,
 from api.db.services.document_service import DocumentService
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.llm_service import LLMBundle
-from common.metadata_utils import apply_meta_data_filter
+from common.metadata_utils import apply_meta_data_filter, convert_conditions, meta_filter
 from api.db.services.search_service import SearchService
 from api.db.services.user_service import UserTenantService
 from common.misc_utils import get_uuid
@@ -129,11 +129,33 @@ async def chat_completion(tenant_id, chat_id):
         req = {"question": ""}
     if not req.get("session_id"):
         req["question"] = ""
-    if not DialogService.query(tenant_id=tenant_id, id=chat_id, status=StatusEnum.VALID.value):
+    dia = DialogService.query(tenant_id=tenant_id, id=chat_id, status=StatusEnum.VALID.value)
+    if not dia:
         return get_error_data_result(f"You don't own the chat {chat_id}")
+    dia = dia[0]
     if req.get("session_id"):
         if not ConversationService.query(id=req["session_id"], dialog_id=chat_id):
             return get_error_data_result(f"You don't own the session {req['session_id']}")
+
+    metadata_condition = req.get("metadata_condition") or {}
+    if metadata_condition and not isinstance(metadata_condition, dict):
+        return get_error_data_result(message="metadata_condition must be an object.")
+
+    if metadata_condition and req.get("question"):
+        metas = DocumentService.get_meta_by_kbs(dia.kb_ids or [])
+        filtered_doc_ids = meta_filter(
+            metas,
+            convert_conditions(metadata_condition),
+            metadata_condition.get("logic", "and"),
+        )
+        if metadata_condition.get("conditions") and not filtered_doc_ids:
+            filtered_doc_ids = ["-999"]
+
+        if filtered_doc_ids:
+            req["doc_ids"] = ",".join(filtered_doc_ids)
+        else:
+            req.pop("doc_ids", None)
+
     if req.get("stream", True):
         resp = Response(rag_completion(tenant_id, chat_id, **req), mimetype="text/event-stream")
         resp.headers.add_header("Cache-control", "no-cache")
@@ -196,7 +218,19 @@ async def chat_completion_openai_like(tenant_id, chat_id):
             {"role": "user", "content": "Can you tell me how to install neovim"},
         ],
         stream=stream,
-        extra_body={"reference": reference}
+        extra_body={
+            "reference": reference,
+            "metadata_condition": {
+                "logic": "and",
+                "conditions": [
+                    {
+                        "name": "author",
+                        "comparison_operator": "is",
+                        "value": "bob"
+                    }
+                ]
+            }
+        }
     )
 
     if stream:
@@ -212,7 +246,11 @@ async def chat_completion_openai_like(tenant_id, chat_id):
     """
     req = await get_request_json()
 
-    need_reference = bool(req.get("reference", False))
+    extra_body = req.get("extra_body") or {}
+    if extra_body and not isinstance(extra_body, dict):
+        return get_error_data_result("extra_body must be an object.")
+
+    need_reference = bool(extra_body.get("reference", False))
 
     messages = req.get("messages", [])
     # To prevent empty [] input
@@ -230,6 +268,22 @@ async def chat_completion_openai_like(tenant_id, chat_id):
         return get_error_data_result(f"You don't own the chat {chat_id}")
     dia = dia[0]
 
+    metadata_condition = extra_body.get("metadata_condition") or {}
+    if metadata_condition and not isinstance(metadata_condition, dict):
+        return get_error_data_result(message="metadata_condition must be an object.")
+
+    doc_ids_str = None
+    if metadata_condition:
+        metas = DocumentService.get_meta_by_kbs(dia.kb_ids or [])
+        filtered_doc_ids = meta_filter(
+            metas,
+            convert_conditions(metadata_condition),
+            metadata_condition.get("logic", "and"),
+        )
+        if metadata_condition.get("conditions") and not filtered_doc_ids:
+            filtered_doc_ids = ["-999"]
+        doc_ids_str = ",".join(filtered_doc_ids) if filtered_doc_ids else None
+
     # Filter system and non-sense assistant messages
     msg = []
     for m in messages:
@@ -277,14 +331,17 @@ async def chat_completion_openai_like(tenant_id, chat_id):
             }
 
             try:
-                async for ans in async_chat(dia, msg, True, toolcall_session=toolcall_session, tools=tools, quote=need_reference):
+                chat_kwargs = {"toolcall_session": toolcall_session, "tools": tools, "quote": need_reference}
+                if doc_ids_str:
+                    chat_kwargs["doc_ids"] = doc_ids_str
+                async for ans in async_chat(dia, msg, True, **chat_kwargs):
                     last_ans = ans
                     answer = ans["answer"]
 
                     reasoning_match = re.search(r"<think>(.*?)</think>", answer, flags=re.DOTALL)
                     if reasoning_match:
                         reasoning_part = reasoning_match.group(1)
-                        content_part = answer[reasoning_match.end():]
+                        content_part = answer[reasoning_match.end() :]
                     else:
                         reasoning_part = ""
                         content_part = answer
@@ -329,8 +386,7 @@ async def chat_completion_openai_like(tenant_id, chat_id):
             response["choices"][0]["delta"]["content"] = None
             response["choices"][0]["delta"]["reasoning_content"] = None
             response["choices"][0]["finish_reason"] = "stop"
-            response["usage"] = {"prompt_tokens": len(prompt), "completion_tokens": token_used,
-                                 "total_tokens": len(prompt) + token_used}
+            response["usage"] = {"prompt_tokens": len(prompt), "completion_tokens": token_used, "total_tokens": len(prompt) + token_used}
             if need_reference:
                 response["choices"][0]["delta"]["reference"] = chunks_format(last_ans.get("reference", []))
                 response["choices"][0]["delta"]["final_content"] = last_ans.get("answer", "")
@@ -345,7 +401,10 @@ async def chat_completion_openai_like(tenant_id, chat_id):
         return resp
     else:
         answer = None
-        async for ans in async_chat(dia, msg, False, toolcall_session=toolcall_session, tools=tools, quote=need_reference):
+        chat_kwargs = {"toolcall_session": toolcall_session, "tools": tools, "quote": need_reference}
+        if doc_ids_str:
+            chat_kwargs["doc_ids"] = doc_ids_str
+        async for ans in async_chat(dia, msg, False, **chat_kwargs):
             # focus answer content only
             answer = ans
             break

docs/references/http_api_reference.md

@@ -48,6 +48,7 @@ This API follows the same request and response format as OpenAI's API. It allows
   - `"model"`: `string`
   - `"messages"`: `object list`
   - `"stream"`: `boolean`
+  - `"extra_body"`: `object` (optional)
 
 ##### Request example
 
@@ -59,7 +60,20 @@ curl --request POST \
      --data '{
         "model": "model",
         "messages": [{"role": "user", "content": "Say this is a test!"}],
-        "stream": true
+        "stream": true,
+        "extra_body": {
+          "reference": true,
+          "metadata_condition": {
+            "logic": "and",
+            "conditions": [
+              {
+                "name": "author",
+                "comparison_operator": "is",
+                "value": "bob"
+              }
+            ]
+          }
+        }
       }'
 ```
 
@@ -74,6 +88,11 @@ curl --request POST \
 - `stream` (*Body parameter*) `boolean`  
   Whether to receive the response as a stream. Set this to `false` explicitly if you prefer to receive the entire response in one go instead of as a stream.
 
+- `extra_body` (*Body parameter*) `object`  
+  Extra request parameters:  
+  - `reference`: `boolean` - include reference in the final chunk (stream) or in the final message (non-stream).
+  - `metadata_condition`: `object` - metadata filter conditions applied to retrieval results.
+
 #### Response
 
 Stream:
@@ -3185,6 +3204,7 @@ Asks a specified chat assistant a question to start an AI-powered conversation.
   - `"stream"`: `boolean`
   - `"session_id"`: `string` (optional)
   - `"user_id`: `string` (optional)
+  - `"metadata_condition"`: `object` (optional)
 
 ##### Request example
 
@@ -3207,7 +3227,17 @@ curl --request POST \
      {
           "question": "Who are you",
           "stream": true,
-          "session_id":"9fa7691cb85c11ef9c5f0242ac120005"
+          "session_id":"9fa7691cb85c11ef9c5f0242ac120005",
+          "metadata_condition": {
+            "logic": "and",
+            "conditions": [
+              {
+                "name": "author",
+                "comparison_operator": "is",
+                "value": "bob"
+              }
+            ]
+          }
      }'
 ```
 
@@ -3225,6 +3255,13 @@ curl --request POST \
   The ID of session. If it is not provided, a new session will be generated.
 - `"user_id"`: (*Body parameter*), `string`  
   The optional user-defined ID. Valid *only* when no `session_id` is provided.
+- `"metadata_condition"`: (*Body parameter*), `object`  
+  Optional metadata filter conditions applied to retrieval results.  
+  - `logic`: `string`, one of `and` / `or`
+  - `conditions`: `list[object]` where each condition contains:
+    - `name`: `string` metadata key
+    - `comparison_operator`: `string` (e.g. `is`, `not is`, `contains`, `not contains`, `start with`, `end with`, `empty`, `not empty`, `>`, `<`, `≥`, `≤`)
+    - `value`: `string|number|boolean` (optional for `empty`/`not empty`)
 
 #### Response
 

rag/llm/chat_model.py

@@ -1217,11 +1217,7 @@ class LiteLLMBase(ABC):
         self.toolcall_sessions = {}
 
         # Factory specific fields
-        if self.provider == SupportedLiteLLMProvider.Bedrock:
-            self.bedrock_ak = json.loads(key).get("bedrock_ak", "")
-            self.bedrock_sk = json.loads(key).get("bedrock_sk", "")
-            self.bedrock_region = json.loads(key).get("bedrock_region", "")
-        elif self.provider == SupportedLiteLLMProvider.OpenRouter:
+        if self.provider == SupportedLiteLLMProvider.OpenRouter:
             self.api_key = json.loads(key).get("api_key", "")
             self.provider_order = json.loads(key).get("provider_order", "")
         elif self.provider == SupportedLiteLLMProvider.Azure_OpenAI:
@@ -1624,17 +1620,38 @@ class LiteLLMBase(ABC):
         if self.provider in FACTORY_DEFAULT_BASE_URL:
             completion_args.update({"api_base": self.base_url})
         elif self.provider == SupportedLiteLLMProvider.Bedrock:
+            import boto3
+
             completion_args.pop("api_key", None)
             completion_args.pop("api_base", None)
-            bedrock_credentials = { "aws_region_name": self.bedrock_region }
-            if self.bedrock_ak and self.bedrock_sk:
-                bedrock_credentials["aws_access_key_id"] = self.bedrock_ak
-                bedrock_credentials["aws_secret_access_key"] = self.bedrock_sk
+
+            bedrock_key = json.loads(self.api_key)
+            mode = bedrock_key.get("auth_mode")
+            if not mode:
+                logging.error("Bedrock auth_mode is not provided in the key")
+                raise ValueError("Bedrock auth_mode must be provided in the key")
+
+            bedrock_region = bedrock_key.get("bedrock_region")
+            bedrock_credentials = {"bedrock_region": bedrock_region}
+
+            if mode == "access_key_secret":
+                bedrock_credentials["aws_access_key_id"] = bedrock_key.get("bedrock_ak")
+                bedrock_credentials["aws_secret_access_key"] = bedrock_key.get("bedrock_sk")
+            elif mode == "iam_role":
+                aws_role_arn = bedrock_key.get("aws_role_arn")
+                sts_client = boto3.client("sts", region_name=bedrock_region)
+                resp = sts_client.assume_role(RoleArn=aws_role_arn, RoleSessionName="BedrockSession")
+                creds = resp["Credentials"]
+                bedrock_credentials["aws_access_key_id"] = creds["AccessKeyId"]
+                bedrock_credentials["aws_secret_access_key"] = creds["SecretAccessKey"]
+                bedrock_credentials["aws_session_token"] = creds["SessionToken"]
+
             completion_args.update(
                 {
                     "bedrock_credentials": bedrock_credentials,
                 }
             )
+
         elif self.provider == SupportedLiteLLMProvider.OpenRouter:
             if self.provider_order:
 

rag/llm/embedding_model.py

@@ -463,20 +463,44 @@ class BedrockEmbed(Base):
 
     def __init__(self, key, model_name, **kwargs):
         import boto3
+        # `key` protocol (backend stores as JSON string in `api_key`):
+        # - Must decode into a dict.
+        # - Required: `auth_mode`, `bedrock_region`.
+        # - Supported auth modes:
+        #   - "access_key_secret": requires `bedrock_ak` + `bedrock_sk`.
+        #   - "iam_role": requires `aws_role_arn` and assumes role via STS.
+        #   - else: treated as "assume_role" (default AWS credential chain).
+        key = json.loads(key)
+        mode = key.get("auth_mode")
+        if not mode:
+            logging.error("Bedrock auth_mode is not provided in the key")
+            raise ValueError("Bedrock auth_mode must be provided in the key")
+
+        self.bedrock_region = key.get("bedrock_region")
 
-        self.bedrock_ak = json.loads(key).get("bedrock_ak", "")
-        self.bedrock_sk = json.loads(key).get("bedrock_sk", "")
-        self.bedrock_region = json.loads(key).get("bedrock_region", "")
         self.model_name = model_name
         self.is_amazon = self.model_name.split(".")[0] == "amazon"
         self.is_cohere = self.model_name.split(".")[0] == "cohere"
-
-        if self.bedrock_ak == "" or self.bedrock_sk == "":
-            # Try to create a client using the default credentials if ak/sk are not provided.
-            # Must provide a region.
-            self.client = boto3.client("bedrock-runtime", region_name=self.bedrock_region)
-        else:
+        
+        if mode == "access_key_secret":
+            self.bedrock_ak = key.get("bedrock_ak")
+            self.bedrock_sk = key.get("bedrock_sk")
             self.client = boto3.client(service_name="bedrock-runtime", region_name=self.bedrock_region, aws_access_key_id=self.bedrock_ak, aws_secret_access_key=self.bedrock_sk)
+        elif mode == "iam_role":
+            self.aws_role_arn = key.get("aws_role_arn")
+            sts_client = boto3.client("sts", region_name=self.bedrock_region)
+            resp = sts_client.assume_role(RoleArn=self.aws_role_arn, RoleSessionName="BedrockSession")
+            creds = resp["Credentials"]
+
+            self.client = boto3.client(
+                service_name="bedrock-runtime",
+                aws_access_key_id=creds["AccessKeyId"],
+                aws_secret_access_key=creds["SecretAccessKey"],
+                aws_session_token=creds["SessionToken"],
+            )
+        else: # assume_role
+            self.client = boto3.client("bedrock-runtime", region_name=self.bedrock_region)
+
 
     def encode(self, texts: list):
         texts = [truncate(t, 8196) for t in texts]

web/src/locales/en.ts

@@ -787,6 +787,15 @@ This auto-tagging feature enhances retrieval by adding another layer of domain-s
       deleteModel: 'Delete model',
       bedrockCredentialsHint:
         'Tip: Leave Access Key / Secret Key blank to use AWS IAM authentication.',
+      awsAuthModeAccessKeySecret: 'Access Key',
+      awsAuthModeIamRole: 'IAM Role',
+      awsAuthModeAssumeRole: 'Assume Role',
+      awsAccessKeyId: 'AWS Access Key ID',
+      awsSecretAccessKey: 'AWS Secret Access Key',
+      awsRoleArn: 'AWS Role ARN',
+      awsRoleArnMessage: 'Please enter AWS Role ARN',
+      awsAssumeRoleTip:
+        'If you select this mode, the Amazon EC2 instance will assume its existing role to access AWS services. No additional credentials are required.',
       modelEmptyTip:
         'No models available. <br>Please add models from the panel on the right.',
       sourceEmptyTip: 'No data sources added yet. Select one below to connect.',
@@ -1101,6 +1110,9 @@ Example: Virtual Hosted Style`,
       mcp: 'MCP',
       mineru: {
         modelNameRequired: 'Model name is required',
+        apiServerRequired: 'MinerU API Server Configuration is required',
+        serverUrlBackendLimit:
+          'MinerU Server URL Address is only available for the HTTP client backend',
         apiserver: 'MinerU API Server Configuration',
         outputDir: 'MinerU Output Directory Path',
         backend: 'MinerU Processing Backend Type',
@@ -1112,6 +1124,11 @@ Example: Virtual Hosted Style`,
           vlmTransformers: 'Vision Language Model with Transformers',
           vlmVllmEngine: 'Vision Language Model with vLLM Engine',
           vlmHttpClient: 'Vision Language Model via HTTP Client',
+          vlmMlxEngine: 'Vision Language Model with MLX Engine',
+          vlmVllmAsyncEngine:
+            'Vision Language Model with vLLM Async Engine (Experimental)',
+          vlmLmdeployEngine:
+            'Vision Language Model with LMDeploy Engine (Experimental)',
         },
       },
     },

web/src/locales/zh-traditional.ts

@@ -546,6 +546,15 @@ export default {
       profileDescription: '在此更新您的照片和個人詳細信息。',
       bedrockCredentialsHint:
         '提示：Access Key / Secret Key 可留空，以啟用 AWS IAM 自動驗證。',
+      awsAuthModeAccessKeySecret: 'Access Key 和 Secret',
+      awsAuthModeIamRole: 'IAM Role',
+      awsAuthModeAssumeRole: 'Assume Role',
+      awsAccessKeyId: 'AWS Access Key ID',
+      awsSecretAccessKey: 'AWS Secret Access Key',
+      awsRoleArn: 'AWS Role ARN',
+      awsRoleArnMessage: '請輸入 AWS Role ARN',
+      awsAssumeRoleTip:
+        '選擇此模式後，EC2 執行個體將使用其既有的 IAM Role 存取 AWS 服務，無需額外憑證。',
       maxTokens: '最大token數',
       maxTokensMessage: '最大token數是必填項',
       maxTokensTip:

web/src/locales/zh.ts

@@ -876,6 +876,15 @@ General：实体和关系提取提示来自 GitHub - microsoft/graphrag：基于
       bedrockAKMessage: '请输入 ACCESS KEY',
       addBedrockSK: 'SECRET KEY',
       bedrockSKMessage: '请输入 SECRET KEY',
+      awsAuthModeAccessKeySecret: 'Access Key 和 Secret',
+      awsAuthModeIamRole: 'IAM Role',
+      awsAuthModeAssumeRole: 'Assume Role',
+      awsAccessKeyId: 'AWS Access Key ID',
+      awsSecretAccessKey: 'AWS Secret Access Key',
+      awsRoleArn: 'AWS Role ARN',
+      awsRoleArnMessage: '请输入 AWS Role ARN',
+      awsAssumeRoleTip:
+        '选择此模式后，EC2 实例将使用其已有的 IAM Role 访问 AWS 服务，无需额外的凭证。',
       bedrockRegion: 'AWS Region',
       bedrockRegionMessage: '请选择！',
       'us-east-1': '美国东部 (弗吉尼亚北部)',
@@ -950,6 +959,8 @@ General：实体和关系提取提示来自 GitHub - microsoft/graphrag：基于
       mcp: 'MCP',
       mineru: {
         modelNameRequired: '模型名称为必填项',
+        apiServerRequired: 'MinerU API服务器配置为必填项',
+        serverUrlBackendLimit: '仅在backend 为vlm-http-client 时可填写',
         apiserver: 'MinerU API服务器配置',
         outputDir: 'MinerU输出目录路径',
         backend: 'MinerU处理后端类型',
@@ -961,6 +972,9 @@ General：实体和关系提取提示来自 GitHub - microsoft/graphrag：基于
           vlmTransformers: '基于Transformers的视觉语言模型',
           vlmVllmEngine: '基于vLLM引擎的视觉语言模型',
           vlmHttpClient: '通过HTTP客户端连接的视觉语言模型',
+          vlmMlxEngine: '基于MLX引擎的视觉语言模型',
+          vlmVllmAsyncEngine: '基于vLLM异步引擎的视觉语言模型（实验性）',
+          vlmLmdeployEngine: '基于LMDeploy引擎的视觉语言模型（实验性）',
         },
       },
     },

web/src/pages/user-setting/setting-model/hooks.tsx

@@ -472,10 +472,13 @@ export const useSubmitMinerU = () => {
 
   const onMineruOk = useCallback(
     async (payload: MinerUFormValues) => {
-      const cfg = {
+      const cfg: any = {
         ...payload,
         mineru_delete_output: payload.mineru_delete_output ?? true ? '1' : '0',
       };
+      if (payload.mineru_backend !== 'vlm-http-client') {
+        delete cfg.mineru_server_url;
+      }
       const req: IAddLlmRequestBody = {
         llm_factory: LLMFactory.MinerU,
         llm_name: payload.llm_name,

web/src/pages/user-setting/setting-model/modal/bedrock-modal/index.tsx

@@ -1,15 +1,25 @@
 import { useTranslate } from '@/hooks/common-hooks';
 import { IModalProps } from '@/interfaces/common';
 import { IAddLlmRequestBody } from '@/interfaces/request/llm';
-import { Form, Input, InputNumber, Modal, Select, Typography } from 'antd';
-import { useMemo } from 'react';
+import {
+  Form,
+  Input,
+  InputNumber,
+  Modal,
+  Segmented,
+  Select,
+  Typography,
+} from 'antd';
+import { useMemo, useState } from 'react';
 import { LLMHeader } from '../../components/llm-header';
 import { BedrockRegionList } from '../../constant';
 
 type FieldType = IAddLlmRequestBody & {
+  auth_mode?: 'access_key_secret' | 'iam_role' | 'assume_role';
   bedrock_ak: string;
   bedrock_sk: string;
   bedrock_region: string;
+  aws_role_arn?: string;
 };
 
 const { Option } = Select;
@@ -23,6 +33,8 @@ const BedrockModal = ({
   llmFactory,
 }: IModalProps<IAddLlmRequestBody> & { llmFactory: string }) => {
   const [form] = Form.useForm<FieldType>();
+  const [authMode, setAuthMode] =
+    useState<FieldType['auth_mode']>('access_key_secret');
 
   const { t } = useTranslate('setting');
   const options = useMemo(
@@ -33,13 +45,32 @@ const BedrockModal = ({
   const handleOk = async () => {
     const values = await form.validateFields();
 
+    // Only submit fields related to the active auth mode.
+    const cleanedValues: Record<string, any> = { ...values };
+
+    const fieldsByMode: Record<string, string[]> = {
+      access_key_secret: ['bedrock_ak', 'bedrock_sk'],
+      iam_role: ['aws_role_arn'],
+      assume_role: [],
+    };
+
+    cleanedValues.auth_mode = authMode;
+
+    Object.keys(fieldsByMode).forEach((mode) => {
+      if (mode !== authMode) {
+        fieldsByMode[mode].forEach((field) => {
+          delete cleanedValues[field];
+        });
+      }
+    });
+
     const data = {
-      ...values,
+      ...cleanedValues,
       llm_factory: llmFactory,
       max_tokens: values.max_tokens,
     };
 
-    onOk?.(data);
+    onOk?.(data as unknown as IAddLlmRequestBody);
   };
 
   return (
@@ -47,9 +78,6 @@ const BedrockModal = ({
       title={
         <div>
           <LLMHeader name={llmFactory} />
-          <Text type="secondary" style={{ display: 'block', marginTop: 4 }}>
-            {t('bedrockCredentialsHint')}
-          </Text>
         </div>
       }
       open={visible}
@@ -82,20 +110,75 @@ const BedrockModal = ({
         >
           <Input placeholder={t('bedrockModelNameMessage')} />
         </Form.Item>
-        <Form.Item<FieldType>
-          label={t('addBedrockEngineAK')}
-          name="bedrock_ak"
-          rules={[{ message: t('bedrockAKMessage') }]}
-        >
-          <Input placeholder={t('bedrockAKMessage')} />
-        </Form.Item>
-        <Form.Item<FieldType>
-          label={t('addBedrockSK')}
-          name="bedrock_sk"
-          rules={[{ message: t('bedrockSKMessage') }]}
-        >
-          <Input placeholder={t('bedrockSKMessage')} />
+
+        {/* AWS Credential Mode Switch (AK/SK section only) */}
+        <Form.Item>
+          <Segmented
+            block
+            value={authMode}
+            onChange={(v) => {
+              const next = v as FieldType['auth_mode'];
+              setAuthMode(next);
+              // Clear non-active fields so they won't be validated/submitted by accident.
+              if (next !== 'access_key_secret') {
+                form.setFieldsValue({ bedrock_ak: '', bedrock_sk: '' } as any);
+              }
+              if (next !== 'iam_role') {
+                form.setFieldsValue({ aws_role_arn: '' } as any);
+              }
+              if (next !== 'assume_role') {
+                form.setFieldsValue({ role_arn: '' } as any);
+              }
+            }}
+            options={[
+              {
+                label: t('awsAuthModeAccessKeySecret'),
+                value: 'access_key_secret',
+              },
+              { label: t('awsAuthModeIamRole'), value: 'iam_role' },
+              { label: t('awsAuthModeAssumeRole'), value: 'assume_role' },
+            ]}
+          />
         </Form.Item>
+
+        {authMode === 'access_key_secret' && (
+          <>
+            <Form.Item<FieldType>
+              label={t('awsAccessKeyId')}
+              name="bedrock_ak"
+              rules={[{ required: true, message: t('bedrockAKMessage') }]}
+            >
+              <Input placeholder={t('bedrockAKMessage')} />
+            </Form.Item>
+            <Form.Item<FieldType>
+              label={t('awsSecretAccessKey')}
+              name="bedrock_sk"
+              rules={[{ required: true, message: t('bedrockSKMessage') }]}
+            >
+              <Input placeholder={t('bedrockSKMessage')} />
+            </Form.Item>
+          </>
+        )}
+
+        {authMode === 'iam_role' && (
+          <Form.Item<FieldType>
+            label={t('awsRoleArn')}
+            name="aws_role_arn"
+            rules={[{ required: true, message: t('awsRoleArnMessage') }]}
+          >
+            <Input placeholder={t('awsRoleArnMessage')} />
+          </Form.Item>
+        )}
+
+        {authMode === 'assume_role' && (
+          <Form.Item
+            style={{ marginTop: -8, marginBottom: 16 }}
+            // keep layout consistent with other modes
+          >
+            <Text type="secondary">{t('awsAssumeRoleTip')}</Text>
+          </Form.Item>
+        )}
+
         <Form.Item<FieldType>
           label={t('bedrockRegion')}
           name="bedrock_region"

web/src/pages/user-setting/setting-model/modal/mineru-modal/index.tsx

@@ -16,7 +16,7 @@ import { IModalProps } from '@/interfaces/common';
 import { buildOptions } from '@/utils/form';
 import { zodResolver } from '@hookform/resolvers/zod';
 import { t } from 'i18next';
-import { useForm } from 'react-hook-form';
+import { useForm, useWatch } from 'react-hook-form';
 import { useTranslation } from 'react-i18next';
 import { z } from 'zod';
 import { LLMHeader } from '../../components/llm-header';
@@ -25,15 +25,18 @@ const FormSchema = z.object({
   llm_name: z.string().min(1, {
     message: t('setting.mineru.modelNameRequired'),
   }),
-  mineru_apiserver: z.string().optional(),
+  mineru_apiserver: z.string().url(),
   mineru_output_dir: z.string().optional(),
   mineru_backend: z.enum([
     'pipeline',
     'vlm-transformers',
     'vlm-vllm-engine',
     'vlm-http-client',
+    'vlm-mlx-engine',
+    'vlm-vllm-async-engine',
+    'vlm-lmdeploy-engine',
   ]),
-  mineru_server_url: z.string().optional(),
+  mineru_server_url: z.string().url().optional(),
   mineru_delete_output: z.boolean(),
 });
 
@@ -52,6 +55,9 @@ const MinerUModal = ({
     'vlm-transformers',
     'vlm-vllm-engine',
     'vlm-http-client',
+    'vlm-mlx-engine',
+    'vlm-vllm-async-engine',
+    'vlm-lmdeploy-engine',
   ]);
 
   const form = useForm<MinerUFormValues>({
@@ -62,6 +68,11 @@ const MinerUModal = ({
     },
   });
 
+  const backend = useWatch({
+    control: form.control,
+    name: 'mineru_backend',
+  });
+
   const handleOk = async (values: MinerUFormValues) => {
     const ret = await onOk?.(values as any);
     if (ret) {
@@ -93,6 +104,7 @@ const MinerUModal = ({
             <RAGFlowFormItem
               name="mineru_apiserver"
               label={t('setting.mineru.apiserver')}
+              required
             >
               <Input placeholder="http://host.docker.internal:9987" />
             </RAGFlowFormItem>
@@ -109,18 +121,25 @@ const MinerUModal = ({
               {(field) => (
                 <RAGFlowSelect
                   value={field.value}
-                  onChange={field.onChange}
+                  onChange={(value) => {
+                    field.onChange(value);
+                    if (value !== 'vlm-http-client') {
+                      form.setValue('mineru_server_url', undefined);
+                    }
+                  }}
                   options={backendOptions}
                   placeholder={t('setting.mineru.selectBackend')}
                 />
               )}
             </RAGFlowFormItem>
-            <RAGFlowFormItem
-              name="mineru_server_url"
-              label={t('setting.mineru.serverUrl')}
-            >
-              <Input placeholder="http://your-vllm-server:30000" />
-            </RAGFlowFormItem>
+            {backend === 'vlm-http-client' && (
+              <RAGFlowFormItem
+                name="mineru_server_url"
+                label={t('setting.mineru.serverUrl')}
+              >
+                <Input placeholder="http://your-vllm-server:30000" />
+              </RAGFlowFormItem>
+            )}
             <RAGFlowFormItem
               name="mineru_delete_output"
               label={t('setting.mineru.deleteOutput')}