Merge branch 'release/v0.2.8' into develop

api/app/controllers/file_storage_controller.py

@@ -91,7 +91,7 @@ async def upload_file(
 
     if file_size > settings.MAX_FILE_SIZE:
         raise HTTPException(
-            status_code=status.HTTP_400_BAD_REQUEST,
+            status_code=status.HTTP_413_CONTENT_TOO_LARGE,
             detail=f"The file size exceeds the {settings.MAX_FILE_SIZE} byte limit"
         )
 
@@ -172,7 +172,6 @@ async def upload_file_with_share_token(
     
     # Get share and release info from share_token
     service = ReleaseShareService(db)
-    share_info = service.get_shared_release_info(share_token=share_data.share_token)
     
     # Get share object to access app_id
     share = service.repo.get_by_share_token(share_data.share_token)
@@ -499,6 +498,51 @@ async def get_file_url(
         )
 
 
+@router.get("/files/{file_id}/public-url", response_model=ApiResponse)
+async def get_permanent_file_url(
+    file_id: uuid.UUID,
+    db: Session = Depends(get_db),
+    storage_service: FileStorageService = Depends(get_file_storage_service),
+):
+    """
+    获取文件的永久公开 URL（无过期时间）。
+
+    - 本地存储：返回 API 永久访问地址（基于 FILE_LOCAL_SERVER_URL 配置）
+    - 远程存储（OSS/S3）：返回 bucket 公读地址（需 bucket 已配置公共读权限）
+    """
+    file_metadata = db.query(FileMetadata).filter(FileMetadata.id == file_id).first()
+    if not file_metadata:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="The file does not exist")
+
+    if file_metadata.status != "completed":
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST,
+                            detail=f"File upload not completed, status: {file_metadata.status}")
+
+    file_key = file_metadata.file_key
+    storage = storage_service.storage
+
+    try:
+        if isinstance(storage, LocalStorage):
+            url = f"{settings.FILE_LOCAL_SERVER_URL}/storage/permanent/{file_id}"
+        else:
+            url = await storage.get_permanent_url(file_key)
+            if not url:
+                raise HTTPException(status_code=status.HTTP_501_NOT_IMPLEMENTED,
+                                    detail="Permanent URL not supported for current storage backend")
+
+        api_logger.info(f"Generated permanent URL: file_id={file_id}")
+        return success(
+            data={"url": url, "expires_in": None, "permanent": True, "file_name": file_metadata.file_name},
+            msg="Permanent file URL generated successfully"
+        )
+    except HTTPException:
+        raise
+    except Exception as e:
+        api_logger.error(f"Failed to generate permanent URL: {e}")
+        raise HTTPException(status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+                            detail=f"Failed to generate permanent URL: {str(e)}")
+
+
 @router.get("/public/{file_id}", response_model=Any)
 async def public_download_file(
     request: Request,

api/app/controllers/memory_dashboard_controller.py

@@ -195,10 +195,9 @@ async def get_workspace_end_users(
         api_logger.warning(f"Redis 缓存写入失败: {str(e)}")
 
     # 触发社区聚类补全任务（异步，不阻塞接口响应）
-    # 对有 ExtractedEntity 但无 Community 节点的存量用户自动补跑全量聚类
     try:
         from app.tasks import init_community_clustering_for_users
-        init_community_clustering_for_users.delay(end_user_ids=end_user_ids)
+        init_community_clustering_for_users.delay(end_user_ids=end_user_ids, workspace_id=str(workspace_id))
         api_logger.info(f"已触发社区聚类补全任务，候选用户数: {len(end_user_ids)}")
     except Exception as e:
         api_logger.warning(f"触发社区聚类补全任务失败（不影响主流程）: {str(e)}")

api/app/controllers/memory_working_controller.py

@@ -33,35 +33,47 @@ def get_memory_count(
 @router.get("/{end_user_id}/conversations", response_model=ApiResponse)
 def get_conversations(
         end_user_id: uuid.UUID,
+        page: int = 1,
+        pagesize: int = 20,
         current_user: User = Depends(get_current_user),
         db: Session = Depends(get_db)
 ):
     """
-    Retrieve all conversations for the current user in a specific group.
+    Retrieve conversations for the current user in a specific group with pagination.
 
     Args:
         end_user_id (UUID): The group identifier.
+        page (int): Page number (1-based). Defaults to 1.
+        pagesize (int): Number of items per page. Defaults to 20.
         current_user (User, optional): The authenticated user.
         db (Session, optional): SQLAlchemy session.
 
     Returns:
-        ApiResponse: Contains a list of conversation IDs.
-
-    Notes:
-        - Initializes the ConversationService with the current DB session.
-        - Returns only conversation IDs for lightweight response.
-        - Logs can be added to trace requests in production.
+        ApiResponse: Contains a paginated list of conversations.
     """
+    page = max(1, page)
+    page_size = max(1, min(pagesize, 100))  # Limit page size between 1 and 100
     conversation_service = ConversationService(db)
-    conversations = conversation_service.get_user_conversations(
-        end_user_id
+    conversations, total = conversation_service.get_user_conversations(
+        end_user_id,
+        page=page,
+        page_size=page_size
     )
-    return success(data=[
-        {
-            "id": conversation.id,
-            "title": conversation.title
-        } for conversation in conversations
-    ], msg="get conversations success")
+    return success(data={
+        "items": [
+            {
+                "id": conversation.id,
+                "title": conversation.title
+            } for conversation in conversations
+        ],
+        "total": total,
+        "page": {
+            "page": page,
+            "pagesize": page_size,
+            "total": total,
+            "hasnext": (page * page_size) < total
+        },
+    }, msg="get conversations success")
 
 
 @router.get("/{end_user_id}/messages", response_model=ApiResponse)

api/app/controllers/tool_controller.py

@@ -76,6 +76,8 @@ async def get_tool_methods(
         if methods is None:
             raise HTTPException(status_code=404, detail="工具不存在")
         return success(data=methods, msg="获取工具方法成功")
+    except HTTPException:
+        raise
     except Exception as e:
         raise HTTPException(status_code=500, detail=str(e))
 
@@ -121,6 +123,8 @@ async def create_tool(
         raise HTTPException(status_code=400, detail=e.message)
     except ValueError as e:
         raise HTTPException(status_code=400, detail=str(e))
+    except HTTPException:
+        raise
     except Exception as e:
         raise HTTPException(status_code=500, detail=str(e))
 
@@ -149,6 +153,8 @@ async def update_tool(
         return success(msg="工具更新成功")
     except ValueError as e:
         raise HTTPException(status_code=400, detail=str(e))
+    except HTTPException:
+        raise
     except Exception as e:
         raise HTTPException(status_code=500, detail=str(e))
 
@@ -191,6 +197,8 @@ async def set_tool_active(
         return success(msg=f"工具已{action}")
     except ValueError as e:
         raise HTTPException(status_code=400, detail=str(e))
+    except HTTPException:
+        raise
     except Exception as e:
         raise HTTPException(status_code=500, detail=str(e))
 
@@ -223,6 +231,8 @@ async def execute_tool(
             },
             msg="工具执行完成"
         )
+    except HTTPException:
+        raise
     except Exception as e:
         raise HTTPException(status_code=500, detail=str(e))
 

api/app/core/config.py

@@ -97,6 +97,7 @@ class Settings:
 
     # File Upload
     MAX_FILE_SIZE: int = int(os.getenv("MAX_FILE_SIZE", "52428800"))
+    MAX_FILE_COUNT: int = int(os.getenv("MAX_FILE_COUNT", "20"))
     FILE_PATH: str = os.getenv("FILE_PATH", "/files")
     FILE_URL_EXPIRES: int = int(os.getenv("FILE_URL_EXPIRES", "3600"))
 

api/app/core/memory/storage_services/clustering_engine/label_propagation.py

@@ -71,11 +71,13 @@ class LabelPropagationEngine:
         connector: Neo4jConnector,
         llm_model_id: Optional[str] = None,
         embedding_model_id: Optional[str] = None,
+        embedding_model_id: Optional[str] = None,
     ):
         self.connector = connector
         self.repo = CommunityRepository(connector)
         self.llm_model_id = llm_model_id
         self.embedding_model_id = embedding_model_id
+        self.embedding_model_id = embedding_model_id
 
     # ──────────────────────────────────────────────────────────────────────────
     # 公开接口

api/app/core/storage/base.py

@@ -121,3 +121,18 @@ class StorageBackend(ABC):
             URL for accessing the file.
         """
         pass
+
+    async def get_permanent_url(self, file_key: str) -> Optional[str]:
+        """
+        Get a permanent public URL for the file (no expiration).
+
+        Returns None by default; remote storage backends should override this
+        if the bucket is configured for public read access.
+
+        Args:
+            file_key: Unique identifier for the file in the storage system.
+
+        Returns:
+            A permanent public URL, or None if not supported.
+        """
+        return None

api/app/core/storage/oss.py

@@ -261,3 +261,13 @@ class OSSStorage(StorageBackend):
             logger.error(f"Failed to generate presigned URL for {file_key}: {e}")
             # Return a basic URL format as fallback
             return f"https://{self.bucket_name}.{self.endpoint.replace('https://', '').replace('http://', '')}/{file_key}"
+
+    async def get_permanent_url(self, file_key: str) -> str:
+        """
+        Get a permanent public URL for the file (requires bucket public read).
+
+        Returns:
+            A permanent URL in the format: https://{bucket}.{endpoint}/{file_key}
+        """
+        host = self.endpoint.replace("https://", "").replace("http://", "")
+        return f"https://{self.bucket_name}.{host}/{file_key}"

api/app/core/storage/s3.py

@@ -378,3 +378,12 @@ class S3Storage(StorageBackend):
             logger.error(f"Failed to generate presigned URL for {file_key}: {e}")
             # Return a basic URL format as fallback
             return f"https://{self.bucket_name}.s3.{self.region}.amazonaws.com/{file_key}"
+
+    async def get_permanent_url(self, file_key: str) -> str:
+        """
+        Get a permanent public URL for the file (requires bucket public read).
+
+        Returns:
+            A permanent URL in the format: https://{bucket}.s3.{region}.amazonaws.com/{file_key}
+        """
+        return f"https://{self.bucket_name}.s3.{self.region}.amazonaws.com/{file_key}"

api/app/core/workflow/nodes/knowledge/node.py

@@ -5,7 +5,7 @@ from typing import Any
 from app.core.error_codes import BizCode
 from app.core.exceptions import BusinessException
 from app.core.models import RedBearRerank, RedBearModelConfig
-from app.core.rag.vdb.elasticsearch.elasticsearch_vector import ElasticSearchVectorFactory
+from app.core.rag.vdb.elasticsearch.elasticsearch_vector import ElasticSearchVectorFactory, ElasticSearchVector
 from app.core.workflow.engine.state_manager import WorkflowState
 from app.core.workflow.engine.variable_pool import VariablePool
 from app.core.workflow.nodes.base_node import BaseNode
@@ -24,6 +24,7 @@ class KnowledgeRetrievalNode(BaseNode):
     def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
         super().__init__(node_config, workflow_config)
         self.typed_config: KnowledgeRetrievalNodeConfig | None = None
+        self.vector_service: ElasticSearchVector | None = None
 
     def _output_types(self) -> dict[str, VariableType]:
         return {
@@ -163,6 +164,50 @@ class KnowledgeRetrievalNode(BaseNode):
         )
         return reranker
 
+    def knowledge_retrieval(self, db, query, rs, db_knowledge, kb_config):
+        if db_knowledge.type == knowledge_model.KnowledgeType.FOLDER:
+            children = knowledge_repository.get_knowledges_by_parent_id(db=db, parent_id=db_knowledge.id)
+            for child in children:
+                if not (child and child.chunk_num > 0 and child.status == 1):
+                    continue
+                kb_config.kb_id = child.id
+                self.knowledge_retrieval(db, query, rs, child, kb_config)
+            return
+        self.vector_service = ElasticSearchVectorFactory().init_vector(knowledge=db_knowledge)
+        indices = f"Vector_index_{kb_config.kb_id}_Node".lower()
+        match kb_config.retrieve_type:
+            case RetrieveType.PARTICIPLE:
+                rs.extend(self.vector_service.search_by_full_text(query=query, top_k=kb_config.top_k,
+                                                                  indices=indices,
+                                                                  score_threshold=kb_config.similarity_threshold))
+            case RetrieveType.SEMANTIC:
+                rs.extend(self.vector_service.search_by_vector(query=query, top_k=kb_config.top_k,
+                                                               indices=indices,
+                                                               score_threshold=kb_config.vector_similarity_weight))
+            case RetrieveType.HYBRID:
+                rs1 = self.vector_service.search_by_vector(query=query, top_k=kb_config.top_k,
+                                                           indices=indices,
+                                                           score_threshold=kb_config.vector_similarity_weight)
+                rs2 = self.vector_service.search_by_full_text(query=query, top_k=kb_config.top_k,
+                                                              indices=indices,
+                                                              score_threshold=kb_config.similarity_threshold)
+
+                # Deduplicate hybrid retrieval results
+                unique_rs = self._deduplicate_docs(rs1, rs2)
+                if not unique_rs:
+                    return
+                if self.typed_config.reranker_id:
+                    self.vector_service.reranker = self.get_reranker_model()
+                    rs.extend(self.vector_service.rerank(query=query, docs=unique_rs, top_k=kb_config.top_k))
+                else:
+                    rs.extend(sorted(
+                        unique_rs,
+                        key=lambda d: d.metadata.get("score", 0),
+                        reverse=True
+                    )[:kb_config.top_k])
+            case _:
+                raise RuntimeError("Unknown retrieval type")
+
     async def execute(self, state: WorkflowState, variable_pool: VariablePool) -> Any:
         """
         Execute the knowledge retrieval workflow node.
@@ -191,56 +236,19 @@ class KnowledgeRetrievalNode(BaseNode):
         query = self._render_template(self.typed_config.query, variable_pool)
         with get_db_read() as db:
             knowledge_bases = self.typed_config.knowledge_bases
-            existing_ids = self._get_existing_kb_ids(db, [kb.kb_id for kb in knowledge_bases])
-
-            if not existing_ids:
-                raise RuntimeError("Knowledge base retrieval failed: the knowledge base does not exist.")
 
             rs = []
             for kb_config in knowledge_bases:
                 db_knowledge = knowledge_repository.get_knowledge_by_id(db=db, knowledge_id=kb_config.kb_id)
                 if not db_knowledge:
                     raise RuntimeError("The knowledge base does not exist or access is denied.")
+                self.knowledge_retrieval(db, query, rs, db_knowledge, kb_config)
 
-                vector_service = ElasticSearchVectorFactory().init_vector(knowledge=db_knowledge)
-                indices = f"Vector_index_{kb_config.kb_id}_Node".lower()
-                match kb_config.retrieve_type:
-                    case RetrieveType.PARTICIPLE:
-                        rs.extend(vector_service.search_by_full_text(query=query, top_k=kb_config.top_k,
-                                                                     indices=indices,
-                                                                     score_threshold=kb_config.similarity_threshold))
-                    case RetrieveType.SEMANTIC:
-                        rs.extend(vector_service.search_by_vector(query=query, top_k=kb_config.top_k,
-                                                                  indices=indices,
-                                                                  score_threshold=kb_config.vector_similarity_weight))
-                    case RetrieveType.HYBRID:
-                        rs1 = vector_service.search_by_vector(query=query, top_k=kb_config.top_k,
-                                                              indices=indices,
-                                                              score_threshold=kb_config.vector_similarity_weight)
-                        rs2 = vector_service.search_by_full_text(query=query, top_k=kb_config.top_k,
-                                                                 indices=indices,
-                                                                 score_threshold=kb_config.similarity_threshold)
-
-                        # Deduplicate hy    brid retrieval results
-                        unique_rs = self._deduplicate_docs(rs1, rs2)
-                        if not unique_rs:
-                            continue
-                        if self.typed_config.reranker_id:
-                            vector_service.reranker = self.get_reranker_model()
-                            rs.extend(vector_service.rerank(query=query, docs=unique_rs, top_k=kb_config.top_k))
-                        else:
-                            rs.extend(sorted(
-                                unique_rs,
-                                key=lambda d: d.metadata.get("score", 0),
-                                reverse=True
-                            )[:kb_config.top_k])
-                    case _:
-                        raise RuntimeError("Unknown retrieval type")
             if not rs:
                 return []
             if self.typed_config.reranker_id:
-                vector_service.reranker = self.get_reranker_model()
-                final_rs = vector_service.rerank(query=query, docs=rs, top_k=self.typed_config.reranker_top_k)
+                self.vector_service.reranker = self.get_reranker_model()
+                final_rs = self.vector_service.rerank(query=query, docs=rs, top_k=self.typed_config.reranker_top_k)
             else:
                 final_rs = sorted(
                     rs,

api/app/main.py

@@ -506,10 +506,13 @@ async def http_exception_handler(request: Request, exc: HTTPException):
         404: "errors.common.not_found",
         405: "errors.common.method_not_allowed",
         409: "errors.common.conflict",
+        413: "errors.common.payload_too_large",
         422: "errors.common.validation_failed",
         429: "errors.common.too_many_requests",
         500: "errors.common.internal_error",
+        502: "errors.common.bad_gateway",
         503: "errors.common.service_unavailable",
+        504: "errors.common.gateway_timeout",
     }
     
     # 如果有对应的翻译键，使用翻译
@@ -534,7 +537,7 @@ async def http_exception_handler(request: Request, exc: HTTPException):
     
     return JSONResponse(
         status_code=exc.status_code,
-        content=fail(code=exc.status_code, msg=translated_message, error=translated_message)
+        content=fail(code=exc.status_code, msg=translated_message, error=exc.detail)
     )
 
 

api/app/repositories/conversation_repository.py

@@ -90,27 +90,27 @@ class ConversationRepository:
             self,
             user_id: uuid.UUID,
             workspace_id: uuid.UUID = None,
-            limit: int = 10,
-            is_activate: bool = True
-    ) -> list[Conversation]:
+            is_activate: bool = True,
+            page: int = 1,
+            page_size: int = 20
+    ) -> tuple[list[Conversation], int]:
         """
-        Retrieve recent conversations for a specific user.
+        Retrieve recent conversations for a specific user with pagination.
 
         This method queries conversations associated with the given user ID,
         optionally scoped to a specific workspace. Results are ordered by the
-        most recently updated conversations and limited to a fixed number.
+        most recently updated conversations.
 
         Args:
             user_id (uuid.UUID): Unique identifier of the user.
             workspace_id (uuid.UUID, optional): Workspace scope for the query.
                 If provided, only conversations under this workspace will be returned.
-            limit (int): Maximum number of conversations to return.
-                Defaults to 10.
-            is_activate (bool): Convsersation State limit
+            is_activate (bool): Conversation State limit.
+            page (int): Page number (1-based). Defaults to 1.
+            page_size (int): Number of items per page. Defaults to 20.
 
         Returns:
-            list[Conversation]: A list of conversation entities ordered by
-            last updated time (descending).
+            tuple[list[Conversation], int]: A list of conversation entities and total count.
         """
         logger.info(f"Fetching conversation by user_id: {user_id}")
 
@@ -122,18 +122,25 @@ class ConversationRepository:
         if workspace_id:
             stmt = stmt.where(Conversation.workspace_id == workspace_id)
 
-        stmt = stmt.order_by(desc(Conversation.updated_at))
-        stmt = stmt.limit(limit)
+        # Calculate total count
+        total = int(self.db.execute(
+            select(func.count()).select_from(stmt.subquery())
+        ).scalar_one())
 
-        convsersations = list(self.db.scalars(stmt).all())
+        # Apply ordering and pagination
+        stmt = stmt.order_by(desc(Conversation.updated_at))
+        stmt = stmt.offset((page - 1) * page_size).limit(page_size)
+
+        conversations = list(self.db.scalars(stmt).all())
         logger.info(
             "Conversation fetched successfully",
             extra={
                 "user_id": str(user_id),
                 "workspace_id": str(workspace_id),
+                "total": total,
             }
         )
-        return convsersations
+        return conversations, total
 
     def list_conversations(
             self,

api/app/repositories/neo4j/community_repository.py

@@ -24,6 +24,10 @@ from app.repositories.neo4j.cypher_queries import (
     CHECK_USER_HAS_COMMUNITIES,
     UPDATE_COMMUNITY_MEMBER_COUNT,
     UPDATE_COMMUNITY_METADATA,
+    GET_INCOMPLETE_COMMUNITIES,
+    GET_INCOMPLETE_COMMUNITIES_WITH_EMBEDDING,
+    CHECK_COMMUNITY_IS_COMPLETE,
+    CHECK_COMMUNITY_IS_COMPLETE_WITH_EMBEDDING,
     BATCH_UPDATE_COMMUNITY_METADATA,
 )
 
@@ -249,6 +253,31 @@ class CommunityRepository:
             logger.error(f"refresh_member_count failed: {e}")
             return 0
 
+    async def get_incomplete_communities(self, end_user_id: str, check_embedding: bool = False) -> List[str]:
+        """查询该用户下属性不完整的 Community 节点 ID 列表。
+
+        Args:
+            end_user_id: 用户 ID
+            check_embedding: 为 True 时额外检查 summary_embedding 是否缺失（仅当用户有 embedding 模型配置时传 True）
+        """
+        try:
+            query = GET_INCOMPLETE_COMMUNITIES_WITH_EMBEDDING if check_embedding else GET_INCOMPLETE_COMMUNITIES
+            result = await self.connector.execute_query(query, end_user_id=end_user_id)
+            return [row["community_id"] for row in result]
+        except Exception as e:
+            logger.error(f"get_incomplete_communities failed: {e}")
+            return []
+
+    async def is_community_complete(self, community_id: str, end_user_id: str, check_embedding: bool = False) -> bool:
+        """检查单个社区节点的属性是否完整。"""
+        try:
+            query = CHECK_COMMUNITY_IS_COMPLETE_WITH_EMBEDDING if check_embedding else CHECK_COMMUNITY_IS_COMPLETE
+            result = await self.connector.execute_query(query, community_id=community_id, end_user_id=end_user_id)
+            return result[0]["is_complete"] if result else False
+        except Exception as e:
+            logger.error(f"is_community_complete failed: {e}")
+            return False
+
     async def update_community_metadata(
         self,
         community_id: str,
@@ -258,7 +287,7 @@ class CommunityRepository:
         core_entities: List[str],
         summary_embedding: Optional[List[float]] = None,
     ) -> bool:
-        """更新社区的名称、摘要、核心实体列表和摘要向量。"""
+        """更新社区的名称、摘要、核心实体列表及 summary_embedding。"""
         try:
             result = await self.connector.execute_query(
                 UPDATE_COMMUNITY_METADATA,

api/app/repositories/neo4j/cypher_queries.py

@@ -1326,3 +1326,38 @@ RETURN s.statement AS statement,
 ORDER BY COALESCE(s.activation_value, 0) DESC
 LIMIT $limit
 """
+
+CHECK_COMMUNITY_IS_COMPLETE = """
+MATCH (c:Community {community_id: $community_id, end_user_id: $end_user_id})
+RETURN (
+    c.name IS NOT NULL AND c.name <> '' AND
+    c.summary IS NOT NULL AND c.summary <> '' AND
+    c.core_entities IS NOT NULL
+) AS is_complete
+"""
+
+CHECK_COMMUNITY_IS_COMPLETE_WITH_EMBEDDING = """
+MATCH (c:Community {community_id: $community_id, end_user_id: $end_user_id})
+RETURN (
+    c.name IS NOT NULL AND c.name <> '' AND
+    c.summary IS NOT NULL AND c.summary <> '' AND
+    c.core_entities IS NOT NULL AND
+    c.summary_embedding IS NOT NULL
+) AS is_complete
+"""
+
+GET_INCOMPLETE_COMMUNITIES = """
+MATCH (c:Community {end_user_id: $end_user_id})
+WHERE c.name IS NULL OR c.summary IS NULL OR c.core_entities IS NULL
+   OR c.name = '' OR c.summary = ''
+RETURN c.community_id AS community_id
+"""
+
+GET_INCOMPLETE_COMMUNITIES_WITH_EMBEDDING = """
+MATCH (c:Community {end_user_id: $end_user_id})
+WHERE c.name IS NULL OR c.name = ''
+   OR c.summary IS NULL OR c.summary = ''
+   OR c.core_entities IS NULL
+   OR (c.summary_embedding IS NULL AND c.summary IS NOT NULL AND c.summary <> '(empty)')
+RETURN c.community_id AS community_id
+"""

api/app/repositories/workflow_repository.py

@@ -43,6 +43,7 @@ class WorkflowConfigRepository:
         edges: list[dict[str, Any]],
         variables: list[dict[str, Any]] | None = None,
         execution_config: dict[str, Any] | None = None,
+        features: dict[str, Any] | None = None,
         triggers: list[dict[str, Any]] | None = None
     ) -> WorkflowConfig:
         """创建或更新工作流配置
@@ -53,6 +54,7 @@ class WorkflowConfigRepository:
             edges: 边列表
             variables: 变量列表
             execution_config: 执行配置
+            features: 功能特性
             triggers: 触发器列表
         
         Returns:
@@ -82,6 +84,7 @@ class WorkflowConfigRepository:
                 edges=edges,
                 variables=variables or [],
                 execution_config=execution_config or {},
+                features=features or {},
                 triggers=triggers or []
             )
             self.db.add(config)

api/app/schemas/app_schema.py

@@ -149,18 +149,26 @@ class FileUploadConfig(BaseModel):
     )
     # 通用文件：PDF/DOCX/XLSX/TXT/CSV/JSON，最大 100MB
     document_enabled: bool = Field(default=False)
-    document_max_size_mb: int = Field(default=100)
+    document_max_size_mb: int = Field(default=50)
     document_allowed_extensions: List[str] = Field(
-        default=["pdf", "docx", "xlsx", "txt", "csv", "json", "md"]
+        default=["pdf", "docx", "doc", "xlsx", "xls", "txt", "csv", "json", "md"]
     )
     # 视频文件：MP4/MOV/AVI/WebM，最大 500MB
     video_enabled: bool = Field(default=False)
-    video_max_size_mb: int = Field(default=500)
+    video_max_size_mb: int = Field(default=50)
     video_allowed_extensions: List[str] = Field(
-        default=["mp4", "mov"]
+        default=["mp4"]
     )
     # 最大文件数量
-    max_file_count: int = Field(default=5, ge=1, le=20)
+    max_file_count: int = Field(default=5, ge=1)
+
+    @field_validator("max_file_count")
+    @classmethod
+    def validate_max_file_count(cls, v: int) -> int:
+        from app.core.config import settings
+        if v > settings.MAX_FILE_COUNT:
+            raise ValueError(f"max_file_count 不能超过 {settings.MAX_FILE_COUNT}")
+        return v
 
 
 class OpeningStatementConfig(BaseModel):

api/app/services/app_chat_service.py

@@ -118,28 +118,54 @@ class AppChatService:
 
         )
 
+        model_info = ModelInfo(
+            model_name=api_key_obj.model_name,
+            provider=api_key_obj.provider,
+            api_key=api_key_obj.api_key,
+            api_base=api_key_obj.api_base,
+            capability=api_key_obj.capability,
+            is_omni=api_key_obj.is_omni,
+            model_type=ModelType.LLM
+        )
+
         # 加载历史消息
         messages = self.conversation_service.get_messages(
             conversation_id=conversation_id,
             limit=10
         )
-        history = [
-            {"role": msg.role, "content": msg.content}
-            for msg in messages
-        ]
+        history = []
+        for msg in messages:
+            content = [{"type": "text", "text": msg.content}]
+
+            # 处理 meta_data 中的 files
+            if msg.meta_data and msg.meta_data.get("files"):
+                files = msg.meta_data.get("files", [])
+                # 使用 MultimodalService 处理文件
+                multimodal_service = MultimodalService(self.db, api_config=model_info)
+
+                # 将 files 转换为 FileInput 格式
+                file_inputs = []
+                for file in files:
+                    from app.schemas.app_schema import FileInput, TransferMethod
+                    file_input = FileInput(
+                        type=file.get("type"),
+                        transfer_method=TransferMethod.REMOTE_URL,
+                        url=file.get("url")
+                    )
+                    file_inputs.append(file_input)
+
+                history_processed_files = await multimodal_service.history_process_files(files=file_inputs)
+
+                content.extend(history_processed_files)
+
+            history.append({
+                "role": msg.role,
+                "content": content
+            })
 
         # 处理多模态文件
         processed_files = None
         if files:
-            model_info = ModelInfo(
-                model_name=api_key_obj.model_name,
-                provider=api_key_obj.provider,
-                api_key=api_key_obj.api_key,
-                api_base=api_key_obj.api_base,
-                capability=api_key_obj.capability,
-                is_omni=api_key_obj.is_omni,
-                model_type=ModelType.LLM
-            )
             multimodal_service = MultimodalService(self.db, model_info)
             processed_files = await multimodal_service.process_files(user_id, files)
             logger.info(f"处理了 {len(processed_files)} 个文件")
@@ -313,31 +339,54 @@ class AppChatService:
                 streaming=True
             )
 
+            model_info = ModelInfo(
+                model_name=api_key_obj.model_name,
+                provider=api_key_obj.provider,
+                api_key=api_key_obj.api_key,
+                api_base=api_key_obj.api_base,
+                capability=api_key_obj.capability,
+                is_omni=api_key_obj.is_omni,
+                model_type=ModelType.LLM
+            )
+
             # 加载历史消息
+            messages = self.conversation_service.get_messages(
+                conversation_id=conversation_id,
+                limit=10
+            )
             history = []
-            memory_config = {"enabled": True, 'max_history': 10}
-            if memory_config.get("enabled"):
-                messages = self.conversation_service.get_messages(
-                    conversation_id=conversation_id,
-                    limit=memory_config.get("max_history", 10)
-                )
-                history = [
-                    {"role": msg.role, "content": msg.content}
-                    for msg in messages
-                ]
+            for msg in messages:
+                content = [{"type": "text", "text": msg.content}]
+
+                # 处理 meta_data 中的 files
+                if msg.meta_data and msg.meta_data.get("files"):
+                    history_files = msg.meta_data.get("files", [])
+                    # 使用 MultimodalService 处理文件
+                    multimodal_service = MultimodalService(self.db, api_config=model_info)
+
+                    # 将 files 转换为 FileInput 格式
+                    file_inputs = []
+                    for file in history_files:
+                        from app.schemas.app_schema import FileInput, TransferMethod
+                        file_input = FileInput(
+                            type=file.get("type"),
+                            transfer_method=TransferMethod.REMOTE_URL,
+                            url=file.get("url")
+                        )
+                        file_inputs.append(file_input)
+
+                    history_processed_files = await multimodal_service.history_process_files(files=file_inputs)
+
+                    content.extend(history_processed_files)
+
+                history.append({
+                    "role": msg.role,
+                    "content": content
+                })
 
             # 处理多模态文件
             processed_files = None
             if files:
-                model_info = ModelInfo(
-                    model_name=api_key_obj.model_name,
-                    provider=api_key_obj.provider,
-                    api_key=api_key_obj.api_key,
-                    api_base=api_key_obj.api_base,
-                    capability=api_key_obj.capability,
-                    is_omni=api_key_obj.is_omni,
-                    model_type=ModelType.LLM
-                )
                 multimodal_service = MultimodalService(self.db, model_info)
                 processed_files = await multimodal_service.process_files(user_id, files)
                 logger.info(f"处理了 {len(processed_files)} 个文件")
@@ -347,8 +396,14 @@ class AppChatService:
             total_tokens = 0
 
             text_queue: asyncio.Queue = asyncio.Queue()
+            api_key_config = {
+                "model_name": api_key_obj.model_name,
+                "api_key": api_key_obj.api_key,
+                "api_base": api_key_obj.api_base,
+                "provider": api_key_obj.provider,
+            }
             stream_audio_url, tts_task = await self.agent_service._generate_tts_streaming(
-                features_config, api_key_obj,
+                features_config, api_key_config,
                 text_queue=text_queue,
                 tenant_id=tenant_id, workspace_id=workspace_id
             )

api/app/services/app_dsl_service.py

@@ -16,6 +16,7 @@ from app.models.app_release_model import AppRelease
 from app.models.knowledge_model import Knowledge
 from app.models.models_model import ModelConfig
 from app.models.tool_model import ToolConfig as ToolConfigModel
+from app.models.skill_model import Skill
 from app.models.workflow_model import WorkflowConfig
 from app.services.workflow_service import WorkflowService
 from app.core.workflow.adapters.memory_bear.memory_bear_adapter import MemoryBearAdapter
@@ -84,7 +85,9 @@ class AppDslService:
             if "knowledge_retrieval" in cfg:
                 enriched["knowledge_retrieval"] = self._enrich_knowledge_retrieval(cfg["knowledge_retrieval"])
             if "tools" in cfg:
-                enriched["tools"] = self._enrich_tools(cfg["tools"])
+                enriched["tools"] = self._enrich_tools(cfg.get("tools"))
+            if "skills" in cfg:
+                enriched["skills"] = self._enrich_skills(cfg.get("skills"))
             return enriched
         if app_type == AppType.MULTI_AGENT:
             enriched = {**cfg}
@@ -108,6 +111,7 @@ class AppDslService:
                 "variables": config.variables if config else [],
                 "edges": config.edges if config else [],
                 "nodes": config.nodes if config else [],
+                "features": config.features if config else {},
                 "execution_config": config.execution_config if config else {},
                 "triggers": config.triggers if config else [],
             } if config else {}
@@ -123,7 +127,8 @@ class AppDslService:
                 "memory": config.memory if config else None,
                 "variables": config.variables if config else [],
                 "tools": self._enrich_tools(config.tools) if config else [],
-                "skills": config.skills if config else {},
+                "skills": self._enrich_skills(config.skills) if config else {},
+                "features": config.features if config else {}
             } if config else {}
             dsl = {**meta, "app": app_meta, "agent_config": config_data}
 
@@ -185,6 +190,22 @@ class AppDslService:
     def _enrich_tools(self, tools: list) -> list:
         return [{**t, "_ref": self._tool_ref(t.get("tool_id"))} for t in (tools or [])]
 
+    def _skill_ref(self, skill_id) -> Optional[dict]:
+        if not skill_id:
+            return None
+        s = self.db.query(Skill).filter(Skill.id == skill_id).first()
+        return {"id": str(skill_id), "name": s.name} if s else {"id": str(skill_id)}
+
+    def _enrich_skills(self, skills: Optional[dict]) -> Optional[dict]:
+        if not skills:
+            return skills
+        skill_ids = skills.get("skill_ids", [])
+        enriched_ids = [
+            {"id": sid, "_ref": self._skill_ref(sid)}
+            for sid in (skill_ids or [])
+        ]
+        return {**skills, "skill_ids": enriched_ids}
+
     def _agent_ref(self, agent_id) -> Optional[dict]:
         if not agent_id:
             return None
@@ -249,7 +270,8 @@ class AppDslService:
                 memory=self._resolve_memory(cfg.get("memory"), workspace_id, warnings),
                 variables=cfg.get("variables", []),
                 tools=self._resolve_tools(cfg.get("tools", []), tenant_id, warnings),
-                skills=cfg.get("skills", {}),
+                skills=self._resolve_skills(cfg.get("skills", {}), tenant_id, warnings),
+                features=cfg.get("features", {}),
                 is_active=True,
                 created_at=now,
                 updated_at=now,
@@ -290,6 +312,7 @@ class AppDslService:
                 edges=[e.model_dump() for e in result.edges],
                 variables=[v.model_dump() for v in result.variables],
                 execution_config=wf.get("execution_config", {}),
+                features=wf.get("features", {}),
                 triggers=wf.get("triggers", []),
                 validate=False,
             )
@@ -444,6 +467,46 @@ class AppDslService:
             return {**memory, "memory_config_id": None, "enabled": False}
         return memory
 
+    def _resolve_skills(self, skills: Optional[dict], tenant_id: uuid.UUID, warnings: list) -> dict:
+        if not skills:
+            return skills or {}
+        resolved_ids = []
+        for entry in (skills.get("skill_ids") or []):
+            # entry 可能是 {"id": "...", "_ref": {...}} 或直接是字符串
+            if isinstance(entry, dict):
+                ref = entry.get("_ref") or ({"name": None, "id": entry.get("id")} if entry.get("id") else None)
+                skill_id = self._resolve_skill(ref, tenant_id, warnings)
+            else:
+                skill_id = self._resolve_skill({"id": str(entry)}, tenant_id, warnings)
+            if skill_id:
+                resolved_ids.append(str(skill_id))
+        return {**{k: v for k, v in skills.items() if k != "skill_ids"}, "skill_ids": resolved_ids}
+
+    def _resolve_skill(self, ref: Optional[dict], tenant_id: uuid.UUID, warnings: list) -> Optional[str]:
+        if not ref:
+            return None
+        # 先按 id 匹配
+        if ref.get("id"):
+            try:
+                s = self.db.query(Skill).filter(
+                    Skill.id == uuid.UUID(str(ref["id"])),
+                    Skill.tenant_id == tenant_id
+                ).first()
+                if s:
+                    return str(s.id)
+            except Exception:
+                pass
+        # 再按名称匹配
+        if ref.get("name"):
+            s = self.db.query(Skill).filter(
+                Skill.name == ref["name"],
+                Skill.tenant_id == tenant_id
+            ).first()
+            if s:
+                return str(s.id)
+        warnings.append(f"未找到技能: {ref}")
+        return None
+
     def _resolve_tools(self, tools: list, tenant_id: uuid.UUID, warnings: list) -> list:
         result = []
         for t in (tools or []):

api/app/services/app_service.py

@@ -833,8 +833,6 @@ class AppService:
 
             # 跨工作空间时，获取目标工作空间的 tenant_id 用于判断模型配置是否可用
             target_tenant_id = None
-            available_model_ids: set = set()
-            available_kb_ids: set = set()
             if is_cross_workspace:
                 target_ws = self.db.get(Workspace, target_workspace_id)
                 if not target_ws:
@@ -849,28 +847,29 @@ class AppService:
 
                 if source_config:
                     if is_cross_workspace:
-                        # Batch-collect and preload all referenced resources
-                        model_ids, kb_ids = self._collect_resource_ids_from_config(
-                            source_config.default_model_config_id,
-                            source_config.knowledge_retrieval,
-                            source_config.tools
+                        # 跨工作空间：model/tools/skills 属于 tenant 级别直接保留，
+                        # knowledge_bases 属于 workspace 级别需过滤，memory_config 需清空
+                        _, kb_ids = self._collect_resource_ids_from_config(
+                            None, source_config.knowledge_retrieval
                         )
-                        available_model_ids, available_kb_ids = self._preload_cross_workspace_resources(
-                            target_tenant_id, target_workspace_id, model_ids, kb_ids
-                        )
-                        new_model_config_id = self._is_model_available(
-                            source_config.default_model_config_id, available_model_ids
+                        _, available_kb_ids = self._preload_cross_workspace_resources(
+                            target_tenant_id, target_workspace_id, set(), kb_ids
                         )
+                        new_model_config_id = source_config.default_model_config_id
                         new_knowledge_retrieval = self._clean_knowledge_retrieval(
                             source_config.knowledge_retrieval, available_kb_ids
                         )
-                        new_tools = self._clean_tools(
-                            source_config.tools, available_kb_ids
+                        new_tools = copy.deepcopy(source_config.tools) if source_config.tools else []
+                        new_memory = self._clean_memory_cross_workspace(
+                            source_config.memory, target_workspace_id
                         )
+                        new_skills = copy.deepcopy(source_config.skills) if source_config.skills else {}
                     else:
                         new_model_config_id = source_config.default_model_config_id
                         new_knowledge_retrieval = copy.deepcopy(source_config.knowledge_retrieval) if source_config.knowledge_retrieval else None
                         new_tools = copy.deepcopy(source_config.tools) if source_config.tools else []
+                        new_memory = copy.deepcopy(source_config.memory) if source_config.memory else None
+                        new_skills = copy.deepcopy(source_config.skills) if source_config.skills else {}
 
                     new_config = AgentConfig(
                         id=uuid.uuid4(),
@@ -879,9 +878,11 @@ class AppService:
                         default_model_config_id=new_model_config_id,
                         model_parameters=copy.deepcopy(source_config.model_parameters) if source_config.model_parameters else None,
                         knowledge_retrieval=new_knowledge_retrieval,
-                        memory=copy.deepcopy(source_config.memory) if source_config.memory else None,
+                        memory=new_memory,
                         variables=copy.deepcopy(source_config.variables) if source_config.variables else [],
                         tools=new_tools,
+                        skills=new_skills,
+                        features=copy.deepcopy(source_config.features) if source_config.features else {},
                         is_active=True,
                         created_at=now,
                         updated_at=now,
@@ -894,28 +895,14 @@ class AppService:
                 ).first()
 
                 if source_config:
-                    if is_cross_workspace:
-                        model_ids, kb_ids = self._collect_resource_ids_from_workflow_nodes(
-                            source_config.nodes
-                        )
-                        available_model_ids, available_kb_ids = self._preload_cross_workspace_resources(
-                            target_tenant_id, target_workspace_id, model_ids, kb_ids
-                        )
-                        new_nodes = self._clean_workflow_nodes_for_cross_workspace(
-                            source_config.nodes or [],
-                            available_model_ids,
-                            available_kb_ids
-                        )
-                    else:
-                        new_nodes = copy.deepcopy(source_config.nodes) if source_config.nodes else []
-
                     new_config = WorkflowConfig(
                         id=uuid.uuid4(),
                         app_id=new_app.id,
-                        nodes=new_nodes,
+                        nodes=copy.deepcopy(source_config.nodes) if source_config.nodes else [],
                         edges=copy.deepcopy(source_config.edges) if source_config.edges else [],
                         variables=copy.deepcopy(source_config.variables) if source_config.variables else [],
                         execution_config=copy.deepcopy(source_config.execution_config) if source_config.execution_config else {},
+                        features=copy.deepcopy(source_config.features) if source_config.features else {},
                         triggers=copy.deepcopy(source_config.triggers) if source_config.triggers else [],
                         is_active=True,
                         created_at=now,
@@ -929,24 +916,15 @@ class AppService:
                 ).first()
 
                 if source_config:
-                    if is_cross_workspace:
-                        model_ids = {source_config.default_model_config_id} if source_config.default_model_config_id else set()
-                        available_model_ids, _ = self._preload_cross_workspace_resources(
-                            target_tenant_id, target_workspace_id, model_ids, set()
-                        )
-                        new_model_config_id = self._is_model_available(
-                            source_config.default_model_config_id, available_model_ids
-                        )
-                    else:
-                        new_model_config_id = source_config.default_model_config_id
-
+                    # multi_agent 的 model_config_id/sub_agents/routing_rules 均属于 tenant 级别直接保留
+                    # 跨空间时 master_agent_id（AppRelease）属于源空间，需清空
                     new_config = MultiAgentConfig(
                         id=uuid.uuid4(),
                         app_id=new_app.id,
                         master_agent_id=source_config.master_agent_id if not is_cross_workspace else None,
                         master_agent_name=source_config.master_agent_name,
-                        default_model_config_id=new_model_config_id,
-                        model_parameters=source_config.model_parameters,
+                        default_model_config_id=source_config.default_model_config_id,
+                        model_parameters=copy.deepcopy(source_config.model_parameters) if source_config.model_parameters else None,
                         orchestration_mode=source_config.orchestration_mode,
                         sub_agents=copy.deepcopy(source_config.sub_agents) if source_config.sub_agents else [],
                         routing_rules=copy.deepcopy(source_config.routing_rules) if source_config.routing_rules else None,
@@ -1037,8 +1015,7 @@ class AppService:
     @staticmethod
     def _collect_resource_ids_from_config(
             model_config_id: Optional[uuid.UUID],
-            knowledge_retrieval: Optional[dict],
-            tools: Optional[list]
+            knowledge_retrieval: Optional[dict]
     ) -> tuple:
         """Extract all model config IDs and knowledge base IDs from an app config."""
         model_ids: set = set()
@@ -1048,62 +1025,12 @@ class AppService:
             model_ids.add(model_config_id)
 
         if knowledge_retrieval and isinstance(knowledge_retrieval, dict):
-            if "kb_ids" in knowledge_retrieval:
-                for kid in knowledge_retrieval.get("kb_ids", []):
-                    if kid:
-                        kb_ids.add(str(kid))
-            if knowledge_retrieval.get("knowledge_id"):
-                kb_ids.add(str(knowledge_retrieval["knowledge_id"]))
-
-        if tools:
-            for tool in tools:
-                if isinstance(tool, dict):
-                    kid = tool.get("knowledge_id") or tool.get("kb_id")
-                    if kid:
-                        kb_ids.add(str(kid))
+            if "knowledge_bases" in knowledge_retrieval:
+                for kid in knowledge_retrieval.get("knowledge_bases", []):
+                    kb_ids.add(str(kid.get("kb_id")))
 
         return model_ids, kb_ids
 
-    @staticmethod
-    def _collect_resource_ids_from_workflow_nodes(nodes: list) -> tuple:
-        """Extract all model config IDs and knowledge base IDs from workflow nodes."""
-        model_ids: set = set()
-        kb_ids: set = set()
-
-        for node in (nodes or []):
-            if not isinstance(node, dict):
-                continue
-            data = node.get("data", {})
-            if not isinstance(data, dict):
-                continue
-            for key in ("model_config_id", "default_model_config_id"):
-                val = data.get(key)
-                if val:
-                    try:
-                        model_ids.add(uuid.UUID(str(val)))
-                    except (ValueError, AttributeError):
-                        pass
-            kr = data.get("knowledge_retrieval")
-            if isinstance(kr, dict):
-                for kid in kr.get("kb_ids", []):
-                    if kid:
-                        kb_ids.add(str(kid))
-                if kr.get("knowledge_id"):
-                    kb_ids.add(str(kr["knowledge_id"]))
-            if data.get("knowledge_id"):
-                kb_ids.add(str(data["knowledge_id"]))
-            for kid in data.get("kb_ids", []):
-                if kid:
-                    kb_ids.add(str(kid))
-
-        return model_ids, kb_ids
-
-    @staticmethod
-    def _is_model_available(model_config_id: Optional[uuid.UUID], available_model_ids: set) -> Optional[uuid.UUID]:
-        if not model_config_id:
-            return None
-        return model_config_id if model_config_id in available_model_ids else None
-
     @staticmethod
     def _is_kb_available(kb_id: Optional[str], available_kb_ids: set) -> Optional[str]:
         if not kb_id:
@@ -1124,95 +1051,53 @@ class AppService:
 
         cleaned = copy.deepcopy(knowledge_retrieval)
 
-        if "kb_ids" in cleaned and isinstance(cleaned["kb_ids"], list):
-            cleaned["kb_ids"] = [
-                kid for kid in cleaned["kb_ids"]
-                if self._is_kb_available(kid, available_kb_ids)
+        if "knowledge_bases" in cleaned and isinstance(cleaned["knowledge_bases"], list):
+            cleaned["knowledge_bases"] = [
+                kb for kb in cleaned["knowledge_bases"]
+                if self._is_kb_available(kb.get("kb_id"), available_kb_ids)
             ]
 
-        if "knowledge_id" in cleaned:
-            cleaned["knowledge_id"] = self._is_kb_available(
-                cleaned.get("knowledge_id"), available_kb_ids
-            )
-
         return cleaned
 
-    def _clean_tools(
+    def _clean_memory_cross_workspace(
             self,
-            tools: Optional[list],
-            available_kb_ids: set
-    ) -> list:
-        """Clean tools config, keeping built-in tools and tools with available KBs."""
-        if not tools:
-            return []
+            memory: Optional[dict],
+            target_workspace_id: uuid.UUID
+    ) -> Optional[dict]:
+        """Clear memory_config_id/memory_content if it doesn't belong to target workspace."""
+        if not memory:
+            return None
 
-        cleaned = []
-        for tool in tools:
-            if not isinstance(tool, dict):
-                cleaned.append(tool)
-                continue
+        from app.models.memory_config_model import MemoryConfig
 
-            tool_type = tool.get("type", "")
-            if tool_type in ("builtin", "built_in", "system"):
-                cleaned.append(copy.deepcopy(tool))
-                continue
+        cleaned = copy.deepcopy(memory)
+        # 兼容旧字段 memory_content 和新字段 memory_config_id
+        mid = cleaned.get("memory_config_id") or cleaned.get("memory_content")
+        if mid:
+            try:
+                mid_uuid = uuid.UUID(str(mid))
+            except (ValueError, AttributeError):
+                exists = self.db.query(MemoryConfig).filter(
+                    MemoryConfig.config_id_old == int(mid),
+                    MemoryConfig.workspace_id == target_workspace_id
+                ).first()
+                if not exists:
+                    cleaned["memory_config_id"] = None
+                    cleaned.pop("memory_content", None)
+                    cleaned["enabled"] = False
+                return cleaned
 
-            kb_id = tool.get("knowledge_id") or tool.get("kb_id")
-            if kb_id:
-                if self._is_kb_available(kb_id, available_kb_ids):
-                    cleaned.append(copy.deepcopy(tool))
-                continue
+            exists = self.db.query(
+                self.db.query(MemoryConfig).filter(
+                    MemoryConfig.config_id == mid_uuid,
+                    MemoryConfig.workspace_id == target_workspace_id
+                ).exists()
+            ).scalar()
+            if not exists:
+                cleaned["memory_config_id"] = None
+                cleaned.pop("memory_content", None)
+                cleaned["enabled"] = False
 
-            cleaned.append(copy.deepcopy(tool))
-
-        return cleaned
-
-    def _clean_workflow_nodes_for_cross_workspace(
-            self,
-            nodes: list,
-            available_model_ids: set,
-            available_kb_ids: set
-    ) -> list:
-        """Clean workflow nodes, using pre-loaded resource sets. Uses deepcopy to avoid mutating source."""
-        if not nodes:
-            return []
-
-        cleaned = []
-        for node in nodes:
-            if not isinstance(node, dict):
-                cleaned.append(node)
-                continue
-
-            node_copy = copy.deepcopy(node)
-            data = node_copy.get("data")
-            if not isinstance(data, dict):
-                cleaned.append(node_copy)
-                continue
-
-            for key in ("model_config_id", "default_model_config_id"):
-                if key in data and data[key]:
-                    try:
-                        mid = uuid.UUID(str(data[key]))
-                    except (ValueError, AttributeError):
-                        data[key] = None
-                        continue
-                    data[key] = str(mid) if mid in available_model_ids else None
-
-            if "knowledge_retrieval" in data and data["knowledge_retrieval"]:
-                data["knowledge_retrieval"] = self._clean_knowledge_retrieval(
-                    data["knowledge_retrieval"], available_kb_ids
-                )
-            if "knowledge_id" in data:
-                data["knowledge_id"] = self._is_kb_available(
-                    data.get("knowledge_id"), available_kb_ids
-                )
-            if "kb_ids" in data and isinstance(data["kb_ids"], list):
-                data["kb_ids"] = [
-                    kid for kid in data["kb_ids"]
-                    if self._is_kb_available(kid, available_kb_ids)
-                ]
-
-            cleaned.append(node_copy)
         return cleaned
 
     def list_apps(

api/app/services/conversation_service.py

@@ -21,6 +21,7 @@ from app.models.conversation_model import ConversationDetail
 from app.models.prompt_optimizer_model import RoleType
 from app.repositories.conversation_repository import ConversationRepository, MessageRepository
 from app.schemas.conversation_schema import ConversationOut
+from app.schemas.model_schema import ModelInfo
 from app.services import workspace_service
 from app.services.model_service import ModelConfigService
 
@@ -119,25 +120,27 @@ class ConversationService:
 
     def get_user_conversations(
             self,
-            user_id: uuid.UUID
-    ) -> list[Conversation]:
+            user_id: uuid.UUID,
+            page: int = 1,
+            page_size: int = 20
+    ) -> tuple[list[Conversation], int]:
         """
-        Retrieve recent conversations for a specific user
-
-        This method delegates persistence logic to the repository layer and
-        applies service-level defaults (e.g. recent conversation limit).
+        Retrieve recent conversations for a specific user with pagination.
 
         Args:
             user_id (uuid.UUID): Unique identifier of the user.
+            page (int): Page number (1-based). Defaults to 1.
+            page_size (int): Number of items per page. Defaults to 20.
 
         Returns:
-            list[Conversation]: A list of recent conversation entities.
+            tuple[list[Conversation], int]: A list of recent conversation entities and total count.
         """
-        conversations = self.conversation_repo.get_conversation_by_user_id(
+        conversations, total = self.conversation_repo.get_conversation_by_user_id(
             user_id,
-            limit=10
+            page=page,
+            page_size=page_size
         )
-        return conversations
+        return conversations, total
 
     def list_conversations(
             self,
@@ -267,10 +270,11 @@ class ConversationService:
 
         return messages
 
-    def get_conversation_history(
+    async def get_conversation_history(
             self,
             conversation_id: uuid.UUID,
-            max_history: Optional[int] = None
+            max_history: Optional[int] = None,
+            api_config: Optional[ModelInfo] = None
     ) -> List[dict]:
         """
         Retrieve historical conversation messages formatted as dictionaries.
@@ -278,6 +282,7 @@ class ConversationService:
         Args:
             conversation_id (uuid.UUID): Conversation UUID.
             max_history (Optional[int]): Maximum number of messages to retrieve.
+            api_config (Optional[ModelInfo]): Model API configuration for multimodal processing.
 
         Returns:
             List[dict]: List of message dictionaries with keys 'role' and 'content'.
@@ -288,13 +293,37 @@ class ConversationService:
         )
 
         # 转换为字典格式
-        history = [
-            {
+        history = []
+        for msg in messages:
+            content = [{"type": "text", "text": msg.content}]
+            
+            # 处理 meta_data 中的 files
+            if msg.meta_data and msg.meta_data.get("files"):
+                files = msg.meta_data.get("files", [])
+                if api_config:
+                    # 使用 MultimodalService 处理文件
+                    from app.services.multimodal_service import MultimodalService
+                    multimodal_service = MultimodalService(self.db, api_config=api_config)
+                    
+                    # 将 files 转换为 FileInput 格式
+                    file_inputs = []
+                    for file in files:
+                        from app.schemas.app_schema import FileInput, TransferMethod
+                        file_input = FileInput(
+                            type=file.get("type"),
+                            transfer_method=TransferMethod.REMOTE_URL,
+                            url=file.get("url")
+                        )
+                        file_inputs.append(file_input)
+
+                    processed_files = await multimodal_service.history_process_files(files=file_inputs)
+
+                    content.extend(processed_files)
+            
+            history.append({
                 "role": msg.role,
-                "content": msg.content
-            }
-            for msg in messages
-        ]
+                "content": content
+            })
 
         return history
 
@@ -522,9 +551,18 @@ class ConversationService:
             type=ModelType(model_type)
         )
 
-        conversation_messages = self.get_conversation_history(
+        conversation_messages = await self.get_conversation_history(
             conversation_id=conversation_id,
-            max_history=20
+            max_history=20,
+            api_config=ModelInfo(
+                model_name=model_name,
+                provider=provider,
+                api_key=api_key,
+                api_base=api_base,
+                capability=api_config.capability,
+                is_omni=api_config.is_omni,
+                model_type=model_type
+            )
         )
         if len(conversation_messages) == 0:
             return ConversationOut(

api/app/services/draft_run_service.py

@@ -579,9 +579,20 @@ class AgentRunService:
                 user_id=user_id
             )
 
+            model_info = ModelInfo(
+                model_name=api_key_config["model_name"],
+                provider=api_key_config["provider"],
+                api_key=api_key_config["api_key"],
+                api_base=api_key_config["api_base"],
+                capability=api_key_config["capability"],
+                is_omni=api_key_config["is_omni"],
+                model_type=model_config.type
+            )
+
             # 6. 加载历史消息
             history = await self._load_conversation_history(
                 conversation_id=conversation_id,
+                api_config=model_info,
                 max_history=10
             )
 
@@ -589,15 +600,6 @@ class AgentRunService:
             processed_files = None
             if files:
                 # 获取 provider 信息
-                model_info = ModelInfo(
-                    model_name=api_key_config["model_name"],
-                    provider=api_key_config["provider"],
-                    api_key=api_key_config["api_key"],
-                    api_base=api_key_config["api_base"],
-                    capability=api_key_config["capability"],
-                    is_omni=api_key_config["is_omni"],
-                    model_type=ModelType.LLM
-                )
                 provider = api_key_config.get("provider", "openai")
                 multimodal_service = MultimodalService(self.db, model_info)
                 processed_files = await multimodal_service.process_files(user_id, files)
@@ -815,9 +817,20 @@ class AgentRunService:
                 sub_agent=sub_agent
             )
 
+            model_info = ModelInfo(
+                model_name=api_key_config["model_name"],
+                provider=api_key_config["provider"],
+                api_key=api_key_config["api_key"],
+                api_base=api_key_config["api_base"],
+                capability=api_key_config["capability"],
+                is_omni=api_key_config["is_omni"],
+                model_type=model_config.type
+            )
+
             # 6. 加载历史消息
             history = await self._load_conversation_history(
                 conversation_id=conversation_id,
+                api_config=model_info,
                 max_history=memory_config.get("max_history", 10)
             )
 
@@ -825,15 +838,6 @@ class AgentRunService:
             processed_files = None
             if files:
                 # 获取 provider 信息
-                model_info = ModelInfo(
-                    model_name=api_key_config["model_name"],
-                    provider=api_key_config["provider"],
-                    api_key=api_key_config["api_key"],
-                    api_base=api_key_config["api_base"],
-                    capability=api_key_config["capability"],
-                    is_omni=api_key_config["is_omni"],
-                    model_type=ModelType.LLM
-                )
                 provider = api_key_config.get("provider", "openai")
                 multimodal_service = MultimodalService(self.db, model_info)
                 processed_files = await multimodal_service.process_files(user_id, files)
@@ -1115,6 +1119,7 @@ class AgentRunService:
     async def _load_conversation_history(
             self,
             conversation_id: str,
+            api_config: ModelInfo | None = None,
             max_history: int = 10
     ) -> List[Dict[str, str]]:
         """加载会话历史消息
@@ -1129,9 +1134,11 @@ class AgentRunService:
         try:
 
             conversation_service = ConversationService(self.db)
-            history = conversation_service.get_conversation_history(
+            # 获取 API 配置用于多模态处理
+            history = await conversation_service.get_conversation_history(
                 conversation_id=uuid.UUID(conversation_id),
-                max_history=max_history
+                max_history=max_history,
+                api_config=api_config
             )
 
             logger.debug(

api/app/services/memory_forget_service.py

@@ -619,7 +619,7 @@ class MemoryForgetService:
                     recent_trends.append({
                         'date': date_str,
                         'merged_count': record.merged_count,
-                        'average_activation': record.average_activation_value,
+                        'average_activation': round(record.average_activation_value, 2) if record.average_activation_value is not None else None,
                         'total_nodes': record.total_nodes,
                         'execution_time': int(record.execution_time.timestamp() * 1000)
                     })

api/app/services/multimodal_service.py

@@ -11,6 +11,8 @@
 import base64
 import io
 import uuid
+import zipfile
+import chardet
 from abc import ABC, abstractmethod
 from typing import List, Dict, Any, Optional
 
@@ -42,12 +44,10 @@ PDF_MIME = ['application/pdf']
 DOC_MIME = [
     'application/msword',
     'application/vnd.openxmlformats-officedocument.wordprocessingml.document',
-    'application/zip'
 ]
 XLSX_MIME = [
     'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet',
     'application/vnd.ms-excel',
-    'application/zip'
 ]
 CSV_MIME = ['text/csv', 'application/csv']
 JSON_MIME = ['application/json']
@@ -418,6 +418,71 @@ class MultimodalService:
         logger.info(f"成功处理 {len(result)}/{len(files)} 个文件，provider={self.provider}")
         return result
 
+    async def history_process_files(
+            self,
+            files: Optional[List[FileInput]],
+    ) -> List[Dict[str, Any]]:
+        """
+        处理文件列表，返回 LLM 可用的格式
+
+        Args:
+            files: 文件输入列表
+
+        Returns:
+            List[Dict]: LLM 可用的内容格式列表（根据 provider 返回不同格式）
+        """
+        if not files:
+            return []
+
+        # 获取对应的策略
+        # dashscope 的 omni 模型使用 OpenAI 兼容格式
+        if self.provider == "dashscope" and self.is_omni:
+            strategy_class = OpenAIFormatStrategy
+        else:
+            strategy_class = PROVIDER_STRATEGIES.get(self.provider)
+            if not strategy_class:
+                logger.warning(f"未找到 provider '{self.provider}' 的策略，使用默认策略")
+                strategy_class = DashScopeFormatStrategy
+
+        result = []
+        for idx, file in enumerate(files):
+            strategy = strategy_class(file)
+            if not file.url:
+                file.url = await self.get_file_url(file)
+            try:
+                if file.type == FileType.IMAGE and "vision" in self.capability:
+                    is_support, content = await self._process_image(file, strategy)
+                    result.append(content)
+                elif file.type == FileType.DOCUMENT:
+                    is_support, content = await self._process_document(file, strategy)
+                    result.append(content)
+                elif file.type == FileType.AUDIO and "audio" in self.capability:
+                    is_support, content = await self._process_audio(file, strategy)
+                    result.append(content)
+                elif file.type == FileType.VIDEO and "video" in self.capability:
+                    is_support, content = await self._process_video(file, strategy)
+                    result.append(content)
+                else:
+                    logger.warning(f"不支持的文件类型: {file.type}")
+            except Exception as e:
+                logger.error(
+                    f"处理文件失败",
+                    extra={
+                        "file_index": idx,
+                        "file_type": file.type,
+                        "error": str(e)
+                    },
+                    exc_info=True
+                )
+                # 继续处理其他文件，不中断整个流程
+                result.append({
+                    "type": "text",
+                    "text": f"[文件处理失败: {str(e)}]"
+                })
+
+        logger.info(f"成功处理 {len(result)}/{len(files)} 个文件，provider={self.provider}")
+        return result
+
     def write_perceptual_memory(
             self,
             end_user_id: str,
@@ -588,12 +653,12 @@ class MultimodalService:
                     file.set_content(file_content)
             file_mime_type = magic.from_buffer(file_content, mime=True)
             if file_mime_type in TEXT_MIME:
-                return file_content.decode("utf-8")
+                return self._decode_text_safe(file_content)
             elif file_mime_type in PDF_MIME:
                 return await self._extract_pdf_text(file_content)
-            elif file_mime_type in DOC_MIME and file.file_type.endswith(('docx', 'doc')):
+            elif self._is_word_file(file_content, file_mime_type):
                 return await self._extract_word_text(file_content)
-            elif file_mime_type in XLSX_MIME and file.file_type.endswith(("xlsx", "xls")):
+            elif self._is_excel_file(file_content, file_mime_type):
                 return await self._extract_xlsx_text(file_content)
             elif file_mime_type in CSV_MIME:
                 return await self._extract_csv_text(file_content)
@@ -622,52 +687,156 @@ class MultimodalService:
 
     @staticmethod
     async def _extract_word_text(file_content: bytes) -> str:
-        """提取 Word 文档文本"""
+        """提取 Word 文档文本（支持 .docx 和旧版 .doc）"""
+        # 先尝试 docx（ZIP 格式）
+        if file_content[:2] == b'PK':
+            try:
+                word_file = io.BytesIO(file_content)
+                doc = Document(word_file)
+                return '\n'.join(p.text for p in doc.paragraphs)
+            except Exception as e:
+                logger.error(f"提取 docx 文本失败: {e}")
+                return f"[docx 提取失败: {str(e)}]"
+
+        # 旧版 .doc（OLE2 格式）
         try:
-            word_file = io.BytesIO(file_content)
-            doc = Document(word_file)
-            text_parts = [paragraph.text for paragraph in doc.paragraphs]
-            return '\n'.join(text_parts)
+            import olefile
+            ole = olefile.OleFileIO(io.BytesIO(file_content))
+            if not ole.exists('WordDocument'):
+                return "[doc 提取失败: 未找到 WordDocument 流]"
+            # 读取 WordDocument 流，提取可见 ASCII/Unicode 文本
+            stream = ole.openstream('WordDocument').read()
+            # Word Binary Format: 文本在流中以 UTF-16-LE 编码存储
+            # 简单提取：过滤出可打印字符段
+            try:
+                text = stream.decode('utf-16-le', errors='ignore')
+            except Exception:
+                text = stream.decode('latin-1', errors='ignore')
+            # 过滤控制字符，保留可打印内容
+            import re
+            text = re.sub(r'[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]', '', text)
+            text = re.sub(r' +', ' ', text).strip()
+            ole.close()
+            return text
         except Exception as e:
-            logger.error(f"提取 Word 文本失败: {e}")
-            return f"[Word 提取失败: {str(e)}]"
+            logger.error(f"提取 doc 文本失败: {e}")
+            return f"[doc 提取失败: {str(e)}]"
 
     @staticmethod
     async def _extract_xlsx_text(file_content: bytes) -> str:
-        """提取 Excel 文本"""
+        """提取 Excel 文本（支持 .xlsx 和旧版 .xls）"""
+        # xlsx（ZIP 格式）
+        if file_content[:2] == b'PK':
+            try:
+                wb = openpyxl.load_workbook(io.BytesIO(file_content), read_only=True, data_only=True)
+                parts = []
+                for sheet in wb.worksheets:
+                    parts.append(f"[Sheet: {sheet.title}]")
+                    for row in sheet.iter_rows(values_only=True):
+                        parts.append('\t'.join('' if v is None else str(v) for v in row))
+                return '\n'.join(parts)
+            except Exception as e:
+                logger.error(f"提取 xlsx 文本失败: {e}")
+                return f"[xlsx 提取失败: {str(e)}]"
+
+        # xls（OLE2/BIFF 格式）
         try:
-            wb = openpyxl.load_workbook(io.BytesIO(file_content), read_only=True, data_only=True)
+            import xlrd
+            wb = xlrd.open_workbook(file_contents=file_content)
             parts = []
-            for sheet in wb.worksheets:
-                parts.append(f"[Sheet: {sheet.title}]")
-                for row in sheet.iter_rows(values_only=True):
-                    parts.append('\t'.join('' if v is None else str(v) for v in row))
+            for sheet in wb.sheets():
+                parts.append(f"[Sheet: {sheet.name}]")
+                for row_idx in range(sheet.nrows):
+                    parts.append('\t'.join(str(sheet.cell_value(row_idx, col)) for col in range(sheet.ncols)))
             return '\n'.join(parts)
         except Exception as e:
-            logger.error(f"提取 Excel 文本失败: {e}")
-            return f"[Excel 提取失败: {str(e)}]"
+            logger.error(f"提取 xls 文本失败: {e}")
+            return f"[xls 提取失败: {str(e)}]"
 
-    @staticmethod
-    async def _extract_csv_text(file_content: bytes) -> str:
+    async def _extract_csv_text(self, file_content: bytes) -> str:
         """提取 CSV 文本"""
         try:
-            text = file_content.decode('utf-8-sig')
+            text = self._decode_text_safe(file_content)
             reader = csv.reader(io.StringIO(text))
             return '\n'.join('\t'.join(row) for row in reader)
         except Exception as e:
             logger.error(f"提取 CSV 文本失败: {e}")
             return f"[CSV 提取失败: {str(e)}]"
 
-    @staticmethod
-    async def _extract_json_text(file_content: bytes) -> str:
+    async def _extract_json_text(self, file_content: bytes) -> str:
         """提取 JSON 文本"""
         try:
-            data = json.loads(file_content.decode('utf-8'))
+            text = self._decode_text_safe(file_content)
+            data = json.loads(text)
             return json.dumps(data, ensure_ascii=False, indent=2)
         except Exception as e:
             logger.error(f"提取 JSON 文本失败: {e}")
             return f"[JSON 提取失败: {str(e)}]"
 
+    def _is_word_file(self, file_content: bytes, mime_type: str) -> bool:
+        """判断是不是 Word 文件（doc / docx），不依赖后缀"""
+        # 旧版 .doc
+        if mime_type == 'application/msword':
+            return True
+
+        # 新版 .docx（ZIP 内部包含 word/document.xml）
+        header = file_content[:4]
+        if header == b'PK\x03\x04':
+            try:
+                with zipfile.ZipFile(io.BytesIO(file_content)) as zf:
+                    return "word/document.xml" in zf.namelist()
+            except:
+                pass
+
+        return False
+
+    def _is_excel_file(self, file_content: bytes, mime_type: str) -> bool:
+        """判断是不是 Excel 文件（xls / xlsx），不依赖后缀"""
+        # 旧版 .xls
+        if mime_type == 'application/vnd.ms-excel':
+            return True
+
+        # 新版 .xlsx（ZIP 内部包含 xl/workbook.xml）
+        header = file_content[:4]
+        if header == b'PK\x03\x04':
+            try:
+                with zipfile.ZipFile(io.BytesIO(file_content)) as zf:
+                    return "xl/workbook.xml" in zf.namelist()
+            except:
+                pass
+
+        return False
+
+    @staticmethod
+    def _decode_text_safe(file_content: bytes) -> str:
+        """
+        【万能文本解码】
+        自动检测编码，支持 utf-8 / gbk / gb2312 / utf-8-sig / ascii 等
+        永远不报错，永远不乱码
+        """
+        if not file_content:
+            return ""
+
+        # 1. 自动检测文件编码
+        detect = chardet.detect(file_content)
+        encoding = detect.get("encoding") or "utf-8"
+        encoding = encoding.lower()
+
+        # 2. 兼容常见中文编码
+        compatible_encodings = ["utf-8", "gbk", "gb18030", "gb2312", "ascii", "latin-1"]
+
+        # 3. 按优先级尝试解码
+        for enc in [encoding] + compatible_encodings:
+            if not enc:
+                continue
+            try:
+                return file_content.decode(enc.strip())
+            except (UnicodeDecodeError, LookupError):
+                continue
+
+        # 终极兜底
+        return file_content.decode("utf-8", errors="replace")
+
 
 def get_multimodal_service(db: Session) -> MultimodalService:
     """获取多模态服务实例（依赖注入）"""

api/app/services/user_memory_service.py

@@ -1408,12 +1408,11 @@ async def analytics_memory_types(
     if end_user_id:
         try:
             conversation_repo = ConversationRepository(db)
-            conversations = conversation_repo.get_conversation_by_user_id(
+            conversations, total = conversation_repo.get_conversation_by_user_id(
                 user_id=uuid.UUID(end_user_id),
-                limit=100,  # 获取更多会话以准确统计
                 is_activate=True
             )
-            work_count = len(conversations)
+            work_count = total
             logger.debug(f"工作记忆数量（会话数）: {work_count} (end_user_id={end_user_id})")
         except Exception as e:
             logger.warning(f"获取会话数量失败，工作记忆数量设为0: {str(e)}")

api/app/services/workflow_service.py

@@ -55,6 +55,7 @@ class WorkflowService:
             edges: list[dict[str, Any]],
             variables: list[dict[str, Any]] | None = None,
             execution_config: dict[str, Any] | None = None,
+            features: dict[str, Any] | None = None,
             triggers: list[dict[str, Any]] | None = None,
             validate: bool = True
     ) -> WorkflowConfig:
@@ -66,6 +67,7 @@ class WorkflowService:
             edges: 边列表
             variables: 变量列表
             execution_config: 执行配置
+            features: 功能特性
             triggers: 触发器列表
             validate: 是否验证配置
 
@@ -81,6 +83,7 @@ class WorkflowService:
             "edges": edges,
             "variables": variables or [],
             "execution_config": execution_config or {},
+            "features": features or {},
             "triggers": triggers or []
         }
 
@@ -101,6 +104,7 @@ class WorkflowService:
             edges=edges,
             variables=variables,
             execution_config=execution_config,
+            features=features,
             triggers=triggers
         )
 

api/app/tasks.py

@@ -2675,13 +2675,15 @@ def write_perceptual_memory(
     time_limit=7200,   # 2小时硬超时
     soft_time_limit=6900,
 )
-def init_community_clustering_for_users(self, end_user_ids: List[str]) -> Dict[str, Any]:
+def init_community_clustering_for_users(self, end_user_ids: List[str], workspace_id: Optional[str] = None) -> Dict[str, Any]:
     """触发型任务：检查指定用户列表，对有 ExtractedEntity 但无 Community 节点的用户执行全量聚类。
 
     由 /dashboard/end_users 接口触发，已有社区节点的用户直接跳过。
+    任务完成且所有用户数据均完整时，写入 Redis 标记，避免下次重复投递。
 
     Args:
         end_user_ids: 需要检查的用户 ID 列表
+        workspace_id: 工作空间 ID，用于完成标记
 
     Returns:
         包含任务执行结果的字典
@@ -2707,6 +2709,7 @@ def init_community_clustering_for_users(self, end_user_ids: List[str]) -> Dict[s
 
             # 批量预取所有用户的配置（内置兜底：用户配置不可用时自动回退到工作空间默认配置）
             user_llm_map: Dict[str, Optional[str]] = {}
+            user_embedding_map: Dict[str, Optional[str]] = {}
             try:
                 with get_db_context() as db:
                     from app.services.memory_agent_service import get_end_users_connected_configs_batch
@@ -2718,21 +2721,54 @@ def init_community_clustering_for_users(self, end_user_ids: List[str]) -> Dict[s
                             try:
                                 cfg = MemoryConfigService(db).load_memory_config(config_id=config_id)
                                 user_llm_map[uid] = str(cfg.llm_model_id) if cfg.llm_model_id else None
+                                user_embedding_map[uid] = str(cfg.embedding_model_id) if cfg.embedding_model_id else None
                             except Exception as e:
-                                logger.warning(f"[CommunityCluster] 用户 {uid} 加载 LLM 配置失败，将使用 None: {e}")
+                                logger.warning(f"[CommunityCluster] 用户 {uid} 加载配置失败，将使用 None: {e}")
                                 user_llm_map[uid] = None
+                                user_embedding_map[uid] = None
                         else:
                             user_llm_map[uid] = None
+                            user_embedding_map[uid] = None
             except Exception as e:
-                logger.warning(f"[CommunityCluster] 批量获取 LLM 配置失败，所有用户将使用 None: {e}")
+                logger.warning(f"[CommunityCluster] 批量获取配置失败，所有用户将使用 None: {e}")
 
             for end_user_id in end_user_ids:
                 try:
-                    # 已有社区节点则跳过
+                    # 已有社区节点时，检查是否存在属性不完整的节点
                     has_communities = await repo.has_communities(end_user_id)
                     if has_communities:
-                        skipped += 1
-                        logger.debug(f"[CommunityCluster] 用户 {end_user_id} 已有社区节点，跳过")
+                        llm_model_id = user_llm_map.get(end_user_id)
+                        embedding_model_id = user_embedding_map.get(end_user_id)
+                        incomplete_ids = await repo.get_incomplete_communities(
+                            end_user_id, check_embedding=bool(embedding_model_id)
+                        )
+                        if not incomplete_ids:
+                            skipped += 1
+                            logger.debug(f"[CommunityCluster] 用户 {end_user_id} 社区节点均完整，跳过")
+                            continue
+
+                        # 对不完整的社区节点逐一补全元数据
+                        engine = LabelPropagationEngine(
+                            connector=connector,
+                            llm_model_id=llm_model_id,
+                            embedding_model_id=embedding_model_id,
+                        )
+                        logger.info(
+                            f"[CommunityCluster] 用户 {end_user_id} 发现 {len(incomplete_ids)} 个属性不完整的社区，开始补全"
+                        )
+                        patch_ok = 0
+                        patch_fail = 0
+                        for cid in incomplete_ids:
+                            try:
+                                await engine._generate_community_metadata(cid, end_user_id)
+                                patch_ok += 1
+                            except Exception as patch_err:
+                                patch_fail += 1
+                                logger.error(f"[CommunityCluster] 社区 {cid} 元数据补全失败: {patch_err}")
+                        logger.info(
+                            f"[CommunityCluster] 用户 {end_user_id} 社区补全完成: 成功={patch_ok}, 失败={patch_fail}"
+                        )
+                        initialized += 1
                         continue
 
                     # 检查是否有 ExtractedEntity 节点
@@ -2742,11 +2778,13 @@ def init_community_clustering_for_users(self, end_user_ids: List[str]) -> Dict[s
                         logger.debug(f"[CommunityCluster] 用户 {end_user_id} 无实体节点，跳过")
                         continue
 
-                    # 每个用户使用自己的 llm_model_id
+                    # 每个用户使用自己的 llm_model_id / embedding_model_id
                     llm_model_id = user_llm_map.get(end_user_id)
+                    embedding_model_id = user_embedding_map.get(end_user_id)
                     engine = LabelPropagationEngine(
                         connector=connector,
                         llm_model_id=llm_model_id,
+                        embedding_model_id=embedding_model_id,
                     )
 
                     logger.info(f"[CommunityCluster] 用户 {end_user_id} 有 {len(entities)} 个实体，开始全量聚类，llm_model_id={llm_model_id}")

api/app/version_info.json

@@ -1,4 +1,38 @@
 {
+  "v0.2.8": {
+    "introduction": {
+      "codeName": "景玉",
+      "releaseDate": "2026-3-20",
+      "upgradePosition": "🐻 MemoryBear v0.2.8 社区版全面升级应用共享、多模态交互与平台基础设施，引入语音交互、感知记忆和云端存储，打造更强大的开放 AI 记忆平台",
+      "coreUpgrades": [
+        "1. 应用共享与发布<br>* 应用共享（Agent、工作流、Agent 集群）：全类型应用共享至其他空间<br>* 分享应用默认开启记忆功能：发布分享后记忆默认开启，关闭时提醒<br>* 工作流记忆分享规则：按记忆配置自动控制分享页记忆开关<br>* 分享会话联网搜索修复：恢复分享应用的联网搜索能力",
+        "2. 多模态与交互 💬<br>* 语音输入：模型接口和应用支持语音输入<br>* 语音回复：应用支持语音回复模态<br>* 多模态感知记忆：记忆系统支持视觉、音频、图片和文件的感知记忆<br>* 对话框文件展示：试运行和体验分享中正确展示上传文件",
+        "3. 平台与基础设施 ⚙️<br>* i18n 国际化：全面多语言多地区支持<br>* 云端文件存储（OSS + S3）：支持阿里云 OSS 和 S3 云端上传<br>* Flower 容器监控：Celery 异步任务监控与管理",
+        "4. EndUser 身份迁移 🔐<br>* EndUser 从 app_id 迁移至 workspace_id：身份从应用级迁移至工作空间级",
+        "5. 情景记忆 🧠<br>* 情景记忆聚类算法：基于社区图谱的聚类算法，支持老用户图谱生成",
+        "6. 稳健性与缺陷修复 🔧<br>* MCP 服务删除后工具 404：修复删除 MCP 服务后接口报错<br>* 应用导出配置不一致：导出已保存配置而非画布状态<br>* 工作流节点 ID 重复：修复复制节点后 ID 冲突<br>* 条件分支连线错误：修复保存刷新后连线错乱<br>* 回复节点内容丢失：修复点击画布后内容消失<br>* 连接桩规则优化：禁止非法连接方向<br>* 知识库状态列宽度：锁定或自适应宽度<br>* 等待中文档预览：支持未完成解析文档预览<br>* 知识库关联修复：统一修复关联问题<br>* 多模态对话连续性：修复多模态内容后无法继续对话<br>* 时区统一：环境变量统一控制存储和任务时区<br>* 遗忘强度精度：修复小数显示过长",
+        "<br>",
+        "v0.2.8 社区版在应用共享和多模态交互方面实现重大升级，感知记忆扩展了平台的认知维度。后续将深化多智能体协作、情景记忆聚类，并持续优化平台稳定性与开放生态。",
+        "MemoryBear —— 让 AI 拥有记忆 🐻✨"
+      ]
+    },
+    "introduction_en": {
+      "codeName": "JingYu",
+      "releaseDate": "2026-3-20",
+      "upgradePosition": "🐻 MemoryBear v0.2.8 Community delivers multimodal interaction, perceptual memory, cloud storage, and workspace-level identity for a more capable open AI memory platform",
+      "coreUpgrades": [
+        "1. Application Sharing & Publishing<br>* Application Sharing (Agent, Workflow, Agent Cluster): Full sharing across all app types<br>* Memory Enabled by Default: Memory auto-enabled on shared apps with disable reminder<br>* Workflow Memory Sharing Rules: Auto-controlled based on memory configuration<br>* Shared Session Web Search Fix: Restored web search for shared apps",
+        "2. Multimodal & Interaction 💬<br>* Voice Input: Model interfaces and apps support voice input<br>* Voice Reply: Apps support voice reply modality<br>* Multimodal Perceptual Memory: Memory system supports visual, audio, image, and file perception<br>* File Display in Chat: Uploaded files display correctly in dry-run and sharing",
+        "3. Platform & Infrastructure ⚙️<br>* i18n Internationalization: Full multi-language multi-region support<br>* Cloud File Storage (OSS + S3): Alibaba Cloud OSS and S3 cloud uploads<br>* Flower Container Monitoring: Celery async task monitoring and management",
+        "4. EndUser Identity Migration 🔐<br>* EndUser Migration from app_id to workspace_id: Identity migrated to workspace level",
+        "5. Episodic Memory 🧠<br>* Episodic Memory Clustering: Community-graph-based clustering with legacy user support",
+        "6. Robustness & Bug Fixes 🔧<br>* MCP Service Deletion 404: Fixed tool endpoint error after MCP removal<br>* App Export Config Mismatch: Exports saved config instead of canvas state<br>* Workflow Duplicate Node ID: Fixed ID conflict on node duplication<br>* Conditional Branch Wiring: Fixed wiring reset after save/refresh<br>* Reply Node Content Loss: Fixed content disappearing on canvas click<br>* Port Connection Rules: Prohibited invalid connection directions<br>* Knowledge Base Status Width: Locked or adaptive column width<br>* Pending Document Preview: Preview support for unparsed documents<br>* Knowledge Base Association Fixes: Consolidated association fixes<br>* Multimodal Conversation Continuity: Fixed single-round limit after multimodal input<br>* Timezone Unification: Env-var controlled unified timezone<br>* Forgetting Strength Precision: Fixed excessive decimal display",
+        "<br>",
+        "v0.2.8 Community delivers major upgrades in application sharing and multimodal interaction, with perceptual memory expanding the platform's cognitive dimensions. Multi-agent collaboration, episodic clustering, and continued platform stability improvements are ahead.",
+        "MemoryBear — Give AI Memory 🐻✨"
+      ]
+    }
+  },
   "v0.2.7": {
     "introduction": {
       "codeName": "武陵",