Merge branch 'develop' into feature/ui_upgrade_zy

api/app/aioRedis.py

@@ -1,6 +1,8 @@
 import asyncio
 import json
 import logging
+import os
+import threading
 from typing import Dict, Any, Optional
 
 import redis.asyncio as redis
@@ -21,6 +23,50 @@ pool = ConnectionPool.from_url(
 )
 aio_redis = redis.StrictRedis(connection_pool=pool)
 
+_REDIS_URL = f"redis://{settings.REDIS_HOST}:{settings.REDIS_PORT}"
+
+# Thread-local storage for connection pools.
+# Each thread (and each forked process) gets its own pool to avoid
+# "Future attached to a different loop" errors in Celery --pool=threads
+# and stale connections after fork in --pool=prefork.
+_thread_local = threading.local()
+
+
+def get_thread_safe_redis() -> redis.StrictRedis:
+    """Return a Redis client whose connection pool is bound to the current
+    thread, process **and** event loop.
+
+    The pool is recreated when:
+    - The PID changes (fork, Celery --pool=prefork)
+    - The thread has no pool yet (Celery --pool=threads)
+    - The previously-cached event loop has been closed (Celery tasks call
+      ``_shutdown_loop_gracefully`` which closes the loop after each run)
+    """
+    current_pid = os.getpid()
+    cached_loop = getattr(_thread_local, "loop", None)
+    loop_stale = cached_loop is not None and cached_loop.is_closed()
+
+    if not hasattr(_thread_local, "pool") \
+            or getattr(_thread_local, "pid", None) != current_pid \
+            or loop_stale:
+        _thread_local.pid = current_pid
+        # Python 3.10+: get_event_loop() raises RuntimeError in threads
+        # where no loop has been set yet (e.g. Celery --pool=threads).
+        try:
+            _thread_local.loop = asyncio.get_event_loop()
+        except RuntimeError:
+            _thread_local.loop = None
+        _thread_local.pool = ConnectionPool.from_url(
+            _REDIS_URL,
+            db=settings.REDIS_DB,
+            password=settings.REDIS_PASSWORD,
+            decode_responses=True,
+            max_connections=5,
+            health_check_interval=30,
+        )
+
+    return redis.StrictRedis(connection_pool=_thread_local.pool)
+
 
 async def get_redis_connection():
     """获取Redis连接"""
@@ -44,10 +90,8 @@ async def aio_redis_set(key: str, val: str | dict, expire: int = None):
             val = json.dumps(val, ensure_ascii=False)
 
         if expire is not None:
-            # 设置带过期时间的键值
             await aio_redis.set(key, val, ex=expire)
         else:
-            # 设置永久键值
             await aio_redis.set(key, val)
     except Exception as e:
         logger.error(f"Redis set错误: {str(e)}")

api/app/cache/memory/activity_stats_cache.py

@@ -10,7 +10,7 @@ import logging
 from typing import Optional, Dict, Any
 from datetime import datetime
 
-from app.aioRedis import aio_redis
+from app.aioRedis import get_thread_safe_redis
 
 logger = logging.getLogger(__name__)
 
@@ -68,7 +68,7 @@ class ActivityStatsCache:
                 "cached": True,
             }
             value = json.dumps(payload, ensure_ascii=False)
-            await aio_redis.set(key, value, ex=expire)
+            await get_thread_safe_redis().set(key, value, ex=expire)
             logger.info(f"设置活动统计缓存成功: {key}, 过期时间: {expire}秒")
             return True
         except Exception as e:
@@ -90,7 +90,7 @@ class ActivityStatsCache:
         """
         try:
             key = cls._get_key(workspace_id)
-            value = await aio_redis.get(key)
+            value = await get_thread_safe_redis().get(key)
             if value:
                 payload = json.loads(value)
                 logger.info(f"命中活动统计缓存: {key}")
@@ -116,7 +116,7 @@ class ActivityStatsCache:
         """
         try:
             key = cls._get_key(workspace_id)
-            result = await aio_redis.delete(key)
+            result = await get_thread_safe_redis().delete(key)
             logger.info(f"删除活动统计缓存: {key}, 结果: {result}")
             return result > 0
         except Exception as e:

api/app/cache/memory/interest_memory.py

@@ -9,7 +9,7 @@ import logging
 from typing import Optional, List, Dict, Any
 from datetime import datetime
 
-from app.aioRedis import aio_redis
+from app.aioRedis import get_thread_safe_redis
 
 logger = logging.getLogger(__name__)
 
@@ -62,7 +62,7 @@ class InterestMemoryCache:
                 "cached": True,
             }
             value = json.dumps(payload, ensure_ascii=False)
-            await aio_redis.set(key, value, ex=expire)
+            await get_thread_safe_redis().set(key, value, ex=expire)
             logger.info(f"设置兴趣分布缓存成功: {key}, 过期时间: {expire}秒")
             return True
         except Exception as e:
@@ -86,7 +86,7 @@ class InterestMemoryCache:
         """
         try:
             key = cls._get_key(end_user_id, language)
-            value = await aio_redis.get(key)
+            value = await get_thread_safe_redis().get(key)
             if value:
                 payload = json.loads(value)
                 logger.info(f"命中兴趣分布缓存: {key}")
@@ -114,7 +114,7 @@ class InterestMemoryCache:
         """
         try:
             key = cls._get_key(end_user_id, language)
-            result = await aio_redis.delete(key)
+            result = await get_thread_safe_redis().delete(key)
             logger.info(f"删除兴趣分布缓存: {key}, 结果: {result}")
             return result > 0
         except Exception as e:

api/app/celery_app.py

@@ -1,5 +1,6 @@
 import os
 import platform
+import re
 from datetime import timedelta
 from urllib.parse import quote
 
@@ -11,21 +12,24 @@ from app.core.logging_config import get_logger
 
 logger = get_logger(__name__)
 
+
+def _mask_url(url: str) -> str:
+    """隐藏 URL 中的密码部分，适用于 redis:// 和 amqp:// 等协议"""
+    return re.sub(r'(://[^:]*:)[^@]+(@)', r'\1***\2', url)
+
 # macOS fork() safety - must be set before any Celery initialization
 if platform.system() == 'Darwin':
     os.environ.setdefault('OBJC_DISABLE_INITIALIZE_FORK_SAFETY', 'YES')
 
 # 创建 Celery 应用实例
-# broker: 任务队列（使用 Redis DB，由 CELERY_BROKER_DB 指定）
-# backend: 结果存储（使用 Redis DB，由 CELERY_BACKEND_DB 指定）
+# broker: 优先使用环境变量 CELERY_BROKER_URL（支持 amqp:// 等任意协议），
+#         未配置则回退到 Redis 方案
+# backend: 结果存储（使用 Redis）
 # NOTE: 不要在 .env 中设置 BROKER_URL / RESULT_BACKEND / CELERY_BROKER / CELERY_BACKEND，
 #       这些名称会被 Celery CLI 的 Click 框架劫持，详见 docs/celery-env-bug-report.md
 
-# Build canonical broker/backend URLs and force them into os.environ so that
-# Celery's Settings.broker_url property (which checks CELERY_BROKER_URL first)
-# cannot be overridden by stray env vars.
-# See: https://github.com/celery/celery/issues/4284
-_broker_url = f"redis://:{quote(settings.REDIS_PASSWORD)}@{settings.REDIS_HOST}:{settings.REDIS_PORT}/{settings.REDIS_DB_CELERY_BROKER}"
+_broker_url = os.getenv("CELERY_BROKER_URL") or \
+    f"redis://:{quote(settings.REDIS_PASSWORD)}@{settings.REDIS_HOST}:{settings.REDIS_PORT}/{settings.REDIS_DB_CELERY_BROKER}"
 _backend_url = f"redis://:{quote(settings.REDIS_PASSWORD)}@{settings.REDIS_HOST}:{settings.REDIS_PORT}/{settings.REDIS_DB_CELERY_BACKEND}"
 os.environ["CELERY_BROKER_URL"] = _broker_url
 os.environ["CELERY_RESULT_BACKEND"] = _backend_url
@@ -45,8 +49,8 @@ celery_app = Celery(
 logger.info(
     "Celery app initialized",
     extra={
-        "broker": _broker_url.replace(quote(settings.REDIS_PASSWORD), "***"),
-        "backend": _backend_url.replace(quote(settings.REDIS_PASSWORD), "***"),
+        "broker": _mask_url(_broker_url),
+        "backend": _mask_url(_backend_url),
     },
 )
 # Default queue for unrouted tasks
@@ -104,6 +108,9 @@ celery_app.conf.update(
         'app.core.memory.agent.long_term_storage.time': {'queue': 'memory_tasks'},
         'app.core.memory.agent.long_term_storage.aggregate': {'queue': 'memory_tasks'},
 
+        # Clustering tasks → memory_tasks queue (使用相同的 worker，避免 macOS fork 问题)
+        'app.tasks.run_incremental_clustering': {'queue': 'memory_tasks'},
+
         # Document tasks → document_tasks queue (prefork worker)
         'app.core.rag.tasks.parse_document': {'queue': 'document_tasks'},
         'app.core.rag.tasks.build_graphrag_for_kb': {'queue': 'document_tasks'},

api/app/controllers/__init__.py

@@ -14,7 +14,6 @@ from . import (
     document_controller,
     emotion_config_controller,
     emotion_controller,
-    end_user_controller,
     file_controller,
     file_storage_controller,
     home_page_controller,
@@ -99,6 +98,5 @@ manager_router.include_router(file_storage_controller.router)
 manager_router.include_router(ontology_controller.router)
 manager_router.include_router(skill_controller.router)
 manager_router.include_router(i18n_controller.router)
-manager_router.include_router(end_user_controller.router)
 
 __all__ = ["manager_router"]

api/app/controllers/app_controller.py

@@ -57,7 +57,6 @@ def list_apps(
         page: int = 1,
         pagesize: int = 10,
         ids: Optional[str] = None,
-        api_key: Optional[str] = None,
         db: Session = Depends(get_db),
         current_user=Depends(get_current_user),
 ):
@@ -66,7 +65,7 @@ def list_apps(
     - 默认包含本工作空间的应用和分享给本工作空间的应用
     - 设置 include_shared=false 可以只查看本工作空间的应用
     - 当提供 ids 参数时，按逗号分割获取指定应用，不分页
-    - 当提供 api_key 参数时，查找该 API Key 关联的应用
+    - search 参数支持：应用名称模糊搜索、API Key 精确搜索
     """
     from sqlalchemy import select as sa_select
     from app.models.api_key_model import ApiKey
@@ -74,23 +73,34 @@ def list_apps(
     workspace_id = current_user.current_workspace_id
     service = app_service.AppService(db)
 
-    # 通过 API Key 搜索：精确匹配，将 resource_id 注入 ids 走统一分页流程
-    if api_key:
-        matched_id = db.execute(
-            sa_select(ApiKey.resource_id).where(
-                ApiKey.workspace_id == workspace_id,
-                ApiKey.api_key == api_key,
-                ApiKey.resource_id.isnot(None),
-            )
-        ).scalar_one_or_none()
-        ids = str(matched_id) if matched_id else ""
+    # 通过 search 参数搜索：支持应用名称模糊搜索和 API Key 精确搜索
+    if search:
+        search = search.strip()
+        # 尝试作为 API Key 精确匹配（API Key 通常较长）
+        if len(search) >= 10:
+            matched_id = db.execute(
+                sa_select(ApiKey.resource_id).where(
+                    ApiKey.workspace_id == workspace_id,
+                    ApiKey.api_key == search,
+                    ApiKey.resource_id.isnot(None),
+                )
+            ).scalar_one_or_none()
+            if matched_id:
+                # 找到 API Key，直接返回关联的应用
+                ids = str(matched_id)
 
-    # 当 ids 存在且不为 None 时，根据 ids 获取应用
+    # 当 ids 存在时，根据 ids 获取应用（不分页）
     if ids is not None:
         app_ids = [app_id.strip() for app_id in ids.split(',') if app_id.strip()]
-        items_orm = app_service.get_apps_by_ids(db, app_ids, workspace_id)
-        items = [service._convert_to_schema(app, workspace_id) for app in items_orm]
-        return success(data=items)
+        if app_ids:
+            items_orm = app_service.get_apps_by_ids(db, app_ids, workspace_id)
+            items = [service._convert_to_schema(app, workspace_id) for app in items_orm]
+            # 返回标准分页格式
+            meta = PageMeta(page=1, pagesize=len(items), total=len(items), hasnext=False)
+            return success(data=PageData(page=meta, items=items))
+        # ids 为空时，返回空列表
+        meta = PageMeta(page=1, pagesize=0, total=0, hasnext=False)
+        return success(data=PageData(page=meta, items=[]))
 
     # 正常分页查询
     items_orm, total = app_service.list_apps(

api/app/controllers/app_log_controller.py

@@ -3,17 +3,16 @@ import uuid
 from typing import Optional
 
 from fastapi import APIRouter, Depends, Query
-from sqlalchemy import select, desc, func
 from sqlalchemy.orm import Session
 
 from app.core.logging_config import get_business_logger
 from app.core.response_utils import success
 from app.db import get_db
 from app.dependencies import get_current_user, cur_workspace_access_guard
-from app.models.conversation_model import Conversation, Message
-from app.schemas.app_log_schema import AppLogConversation, AppLogConversationDetail, AppLogMessage
+from app.schemas.app_log_schema import AppLogConversation, AppLogConversationDetail
 from app.schemas.response_schema import PageData, PageMeta
 from app.services.app_service import AppService
+from app.services.app_log_service import AppLogService
 
 router = APIRouter(prefix="/apps", tags=["App Logs"])
 logger = get_business_logger()
@@ -25,52 +24,35 @@ def list_app_logs(
         app_id: uuid.UUID,
         page: int = Query(1, ge=1),
         pagesize: int = Query(20, ge=1, le=100),
-        user_id: Optional[str] = None,
         is_draft: Optional[bool] = None,
         db: Session = Depends(get_db),
         current_user=Depends(get_current_user),
 ):
     """查看应用下所有会话记录（分页）
 
-    - 支持按 user_id 筛选
     - 支持按 is_draft 筛选（草稿会话 / 发布会话）
     - 按最新更新时间倒序排列
+    - 所有人（包括共享者和被共享者）都只能查看自己的会话记录
     """
     workspace_id = current_user.current_workspace_id
 
     # 验证应用访问权限
-    service = AppService(db)
-    service.get_app(app_id, workspace_id)
+    app_service = AppService(db)
+    app_service.get_app(app_id, workspace_id)
 
-    stmt = select(Conversation).where(
-        Conversation.app_id == app_id,
-        Conversation.workspace_id == workspace_id,
-        Conversation.is_active.is_(True),
+    # 使用 Service 层查询
+    log_service = AppLogService(db)
+    conversations, total = log_service.list_conversations(
+        app_id=app_id,
+        workspace_id=workspace_id,
+        page=page,
+        pagesize=pagesize,
+        is_draft=is_draft
     )
 
-    if user_id:
-        stmt = stmt.where(Conversation.user_id == user_id)
-
-    if is_draft is not None:
-        stmt = stmt.where(Conversation.is_draft == is_draft)
-
-    total = int(db.execute(
-        select(func.count()).select_from(stmt.subquery())
-    ).scalar_one())
-
-    stmt = stmt.order_by(desc(Conversation.updated_at))
-    stmt = stmt.offset((page - 1) * pagesize).limit(pagesize)
-
-    conversations = list(db.scalars(stmt).all())
-
     items = [AppLogConversation.model_validate(c) for c in conversations]
     meta = PageMeta(page=page, pagesize=pagesize, total=total, hasnext=(page * pagesize) < total)
 
-    logger.info(
-        "查询应用日志会话列表",
-        extra={"app_id": str(app_id), "total": total, "page": page}
-    )
-
     return success(data=PageData(page=meta, items=items))
 
 
@@ -86,44 +68,22 @@ def get_app_log_detail(
 
     - 返回会话基本信息 + 所有消息（按时间正序）
     - 消息 meta_data 包含模型名、token 用量等信息
+    - 所有人（包括共享者和被共享者）都只能查看自己的会话详情
     """
     workspace_id = current_user.current_workspace_id
 
     # 验证应用访问权限
-    service = AppService(db)
-    service.get_app(app_id, workspace_id)
+    app_service = AppService(db)
+    app_service.get_app(app_id, workspace_id)
 
-    # 查询会话（确保属于该应用和工作空间）
-    conversation = db.scalars(
-        select(Conversation).where(
-            Conversation.id == conversation_id,
-            Conversation.app_id == app_id,
-            Conversation.workspace_id == workspace_id,
-            Conversation.is_active.is_(True),
-        )
-    ).first()
-
-    if not conversation:
-        from app.core.exceptions import ResourceNotFoundException
-        raise ResourceNotFoundException("会话", str(conversation_id))
-
-    # 查询消息（按时间正序）
-    messages = list(db.scalars(
-        select(Message)
-        .where(Message.conversation_id == conversation_id)
-        .order_by(Message.created_at)
-    ).all())
-
-    detail = AppLogConversationDetail.model_validate(conversation)
-    detail.messages = [AppLogMessage.model_validate(m) for m in messages]
-
-    logger.info(
-        "查询应用日志会话详情",
-        extra={
-            "app_id": str(app_id),
-            "conversation_id": str(conversation_id),
-            "message_count": len(messages)
-        }
+    # 使用 Service 层查询
+    log_service = AppLogService(db)
+    conversation = log_service.get_conversation_detail(
+        app_id=app_id,
+        conversation_id=conversation_id,
+        workspace_id=workspace_id
     )
 
+    detail = AppLogConversationDetail.model_validate(conversation)
+
     return success(data=detail)

api/app/controllers/end_user_controller.py

@@ -1,48 +0,0 @@
-"""End User 管理接口 - 无需认证"""
-
-from app.core.logging_config import get_business_logger
-from app.core.response_utils import success
-from app.db import get_db
-from app.repositories.end_user_repository import EndUserRepository
-from app.schemas.memory_api_schema import (
-    CreateEndUserRequest,
-    CreateEndUserResponse,
-)
-from fastapi import APIRouter, Depends
-from sqlalchemy.orm import Session
-
-router = APIRouter(prefix="/end_users", tags=["End Users"])
-logger = get_business_logger()
-
-
-@router.post("")
-async def create_end_user(
-    data: CreateEndUserRequest,
-    db: Session = Depends(get_db),
-):
-    """
-    Create an end user.
-    
-    Creates a new end user for the given workspace.
-    If an end user with the same other_id already exists in the workspace,
-    returns the existing one.
-    """
-    logger.info(f"Create end user request - other_id: {data.other_id}, workspace_id: {data.workspace_id}")
-
-    end_user_repo = EndUserRepository(db)
-    end_user = end_user_repo.get_or_create_end_user(
-        app_id=None,
-        workspace_id=data.workspace_id,
-        other_id=data.other_id,
-    )
-
-    logger.info(f"End user ready: {end_user.id}")
-
-    result = {
-        "id": str(end_user.id),
-        "other_id": end_user.other_id or "",
-        "other_name": end_user.other_name or "",
-        "workspace_id": str(end_user.workspace_id),
-    }
-
-    return success(data=CreateEndUserResponse(**result).model_dump(), msg="End user created successfully")

api/app/controllers/file_storage_controller.py

@@ -574,8 +574,12 @@ async def get_file_url(
             # For local storage, generate signed URL with expiration
             url = generate_signed_url(str(file_id), expires)
         else:
-            # For remote storage (OSS/S3), get presigned URL
-            url = await storage_service.get_file_url(file_key, expires=expires)
+            # For remote storage (OSS/S3), get presigned URL with forced download
+            url = await storage_service.get_file_url(
+                file_key,
+                expires=expires,
+                file_name=file_metadata.file_name,
+            )
             url = _match_scheme(request, url)
 
         api_logger.info(f"Generated file URL: file_id={file_id}")
@@ -786,7 +790,7 @@ async def permanent_download_file(
         # For remote storage, redirect to presigned URL with long expiration
         try:
             # Use a very long expiration (7 days max for most cloud providers)
-            presigned_url = await storage_service.get_file_url(file_key, expires=604800)
+            presigned_url = await storage_service.get_file_url(file_key, expires=604800, file_name=file_metadata.file_name)
             presigned_url = _match_scheme(request, presigned_url)
             return RedirectResponse(url=presigned_url, status_code=status.HTTP_302_FOUND)
         except Exception as e:

api/app/controllers/mcp_market_config_controller.py

@@ -91,9 +91,11 @@ async def get_mcp_servers(
 
     try:
         cookies = api.get_cookies(token)
+        headers=api.builder_headers(api.headers)
+        headers['Authorization'] = f'Bearer {token}'
         r = api.session.put(
             url=api.mcp_base_url,
-            headers=api.builder_headers(api.headers),
+            headers=headers,
             json=body,
             cookies=cookies)
         raise_for_http_status(r)
@@ -173,6 +175,7 @@ async def get_operational_mcp_servers(
 
     url = f'{api.mcp_base_url}/operational'
     headers = api.builder_headers(api.headers)
+    headers['Authorization'] = f'Bearer {token}'
 
     try:
         cookies = api.get_cookies(access_token=token, cookies_required=True)
@@ -260,7 +263,9 @@ async def create_mcp_market_config(
             api.login(create_data.token)
             body = {'filter': {}, 'page_number': 1, 'page_size': 1, 'search': None}
             cookies = api.get_cookies(create_data.token)
-            r = api.session.put(url=api.mcp_base_url, headers=api.builder_headers(api.headers), json=body, cookies=cookies)
+            headers = api.builder_headers(api.headers)
+            headers['Authorization'] = f'Bearer {create_data.token}'
+            r = api.session.put(url=api.mcp_base_url, headers=headers, json=body, cookies=cookies)
             raise_for_http_status(r)
         except Exception as e:
             api_logger.warning(f"Token validation failed for ModelScope MCP market: {str(e)}")
@@ -290,9 +295,11 @@ async def create_mcp_market_config(
                 'search': ""
             }
             cookies = api.get_cookies(token)
+            headers = api.builder_headers(api.headers)
+            headers['Authorization'] = f'Bearer {token}'
             r = api.session.put(
                 url=api.mcp_base_url,
-                headers=api.builder_headers(api.headers),
+                headers=headers,
                 json=body,
                 cookies=cookies)
             raise_for_http_status(r)
@@ -393,7 +400,9 @@ async def update_mcp_market_config(
             api.login(update_data.token)
             body = {'filter': {}, 'page_number': 1, 'page_size': 1, 'search': None}
             cookies = api.get_cookies(update_data.token)
-            r = api.session.put(url=api.mcp_base_url, headers=api.builder_headers(api.headers), json=body, cookies=cookies)
+            headers = api.builder_headers(api.headers)
+            headers['Authorization'] = f'Bearer {update_data.token}'
+            r = api.session.put(url=api.mcp_base_url, headers=headers, json=body, cookies=cookies)
             raise_for_http_status(r)
         except Exception as e:
             api_logger.warning(f"Token validation failed for ModelScope MCP market: {str(e)}")

api/app/controllers/memory_agent_controller.py

@@ -118,142 +118,142 @@ async def download_log(
             return fail(BizCode.INTERNAL_ERROR, "启动日志流式传输失败", str(e))
 
 
-@router.post("/writer_service", response_model=ApiResponse)
-@cur_workspace_access_guard()
-async def write_server(
-        user_input: Write_UserInput,
-        language_type: str = Header(default=None, alias="X-Language-Type"),
-        db: Session = Depends(get_db),
-        current_user: User = Depends(get_current_user)
-):
-    """
-    Write service endpoint - processes write operations synchronously
-    
-    Args:
-        user_input: Write request containing message and end_user_id
-        language_type: 语言类型 ("zh" 中文, "en" 英文)，通过 X-Language-Type Header 传递
-    
-    Returns:
-        Response with write operation status
-    """
-    # 使用集中化的语言校验
-    language = get_language_from_header(language_type)
-
-    config_id = user_input.config_id
-    workspace_id = current_user.current_workspace_id
-    api_logger.info(f"Write service: workspace_id={workspace_id}, config_id={config_id}, language_type={language}")
-
-    # 获取 storage_type，如果为 None 则使用默认值
-    storage_type = workspace_service.get_workspace_storage_type(
-        db=db,
-        workspace_id=workspace_id,
-        user=current_user
-    )
-    if storage_type is None: storage_type = 'neo4j'
-    user_rag_memory_id = ''
-
-    # 如果 storage_type 是 rag，必须确保有有效的 user_rag_memory_id
-    if storage_type == 'rag':
-        if workspace_id:
-            knowledge = knowledge_repository.get_knowledge_by_name(
-                db=db,
-                name="USER_RAG_MERORY",
-                workspace_id=workspace_id
-            )
-            if knowledge:
-                user_rag_memory_id = str(knowledge.id)
-            else:
-                api_logger.warning(
-                    f"未找到名为 'USER_RAG_MERORY' 的知识库，workspace_id: {workspace_id}，将使用 neo4j 存储")
-                storage_type = 'neo4j'
-        else:
-            api_logger.warning("workspace_id 为空，无法使用 rag 存储，将使用 neo4j 存储")
-            storage_type = 'neo4j'
-
-    api_logger.info(
-        f"Write service requested for group {user_input.end_user_id}, storage_type: {storage_type}, user_rag_memory_id: {user_rag_memory_id}")
-    try:
-        messages_list = memory_agent_service.get_messages_list(user_input)
-        result = await memory_agent_service.write_memory(
-            user_input.end_user_id,
-            messages_list,
-            config_id,
-            db,
-            storage_type,
-            user_rag_memory_id,
-            language
-        )
-
-        return success(data=result, msg="写入成功")
-    except BaseException as e:
-        # Handle ExceptionGroup from TaskGroup (Python 3.11+) or BaseExceptionGroup
-        if hasattr(e, 'exceptions'):
-            error_messages = [f"{type(sub_e).__name__}: {str(sub_e)}" for sub_e in e.exceptions]
-            detailed_error = "; ".join(error_messages)
-            api_logger.error(f"Write operation error (TaskGroup): {detailed_error}", exc_info=True)
-            return fail(BizCode.INTERNAL_ERROR, "写入失败", detailed_error)
-        api_logger.error(f"Write operation error: {str(e)}", exc_info=True)
-        return fail(BizCode.INTERNAL_ERROR, "写入失败", str(e))
-
-
-@router.post("/writer_service_async", response_model=ApiResponse)
-@cur_workspace_access_guard()
-async def write_server_async(
-        user_input: Write_UserInput,
-        language_type: str = Header(default=None, alias="X-Language-Type"),
-        db: Session = Depends(get_db),
-        current_user: User = Depends(get_current_user)
-):
-    """
-    Async write service endpoint - enqueues write processing to Celery
-    
-    Args:
-        user_input: Write request containing message and end_user_id
-        language_type: 语言类型 ("zh" 中文, "en" 英文)，通过 X-Language-Type Header 传递
-    
-    Returns:
-        Task ID for tracking async operation
-        Use GET /memory/write_result/{task_id} to check task status and get result
-    """
-    # 使用集中化的语言校验
-    language = get_language_from_header(language_type)
-
-    config_id = user_input.config_id
-    workspace_id = current_user.current_workspace_id
-    api_logger.info(
-        f"Async write service: workspace_id={workspace_id}, config_id={config_id}, language_type={language}")
-
-    # 获取 storage_type，如果为 None 则使用默认值
-    storage_type = workspace_service.get_workspace_storage_type(
-        db=db,
-        workspace_id=workspace_id,
-        user=current_user
-    )
-    if storage_type is None: storage_type = 'neo4j'
-    user_rag_memory_id = ''
-    if workspace_id:
-
-        knowledge = knowledge_repository.get_knowledge_by_name(
-            db=db,
-            name="USER_RAG_MERORY",
-            workspace_id=workspace_id
-        )
-        if knowledge: user_rag_memory_id = str(knowledge.id)
-    api_logger.info(f"Async write: storage_type={storage_type}, user_rag_memory_id={user_rag_memory_id}")
-    try:
-        # 获取标准化的消息列表
-        messages_list = memory_agent_service.get_messages_list(user_input)
-
-        task = celery_app.send_task(
-            "app.core.memory.agent.write_message",
-            args=[user_input.end_user_id, messages_list, config_id, storage_type, user_rag_memory_id, language]
-        )
-        api_logger.info(f"Write task queued: {task.id}")
-
-        return success(data={"task_id": task.id}, msg="写入任务已提交")
-    except Exception as e:
-        api_logger.error(f"Async write operation failed: {str(e)}")
-        return fail(BizCode.INTERNAL_ERROR, "写入失败", str(e))
+# @router.post("/writer_service", response_model=ApiResponse)
+# @cur_workspace_access_guard()
+# async def write_server(
+#         user_input: Write_UserInput,
+#         language_type: str = Header(default=None, alias="X-Language-Type"),
+#         db: Session = Depends(get_db),
+#         current_user: User = Depends(get_current_user)
+# ):
+#     """
+#     Write service endpoint - processes write operations synchronously
+#
+#     Args:
+#         user_input: Write request containing message and end_user_id
+#         language_type: 语言类型 ("zh" 中文, "en" 英文)，通过 X-Language-Type Header 传递
+#
+#     Returns:
+#         Response with write operation status
+#     """
+#     # 使用集中化的语言校验
+#     language = get_language_from_header(language_type)
+#
+#     config_id = user_input.config_id
+#     workspace_id = current_user.current_workspace_id
+#     api_logger.info(f"Write service: workspace_id={workspace_id}, config_id={config_id}, language_type={language}")
+#
+#     # 获取 storage_type，如果为 None 则使用默认值
+#     storage_type = workspace_service.get_workspace_storage_type(
+#         db=db,
+#         workspace_id=workspace_id,
+#         user=current_user
+#     )
+#     if storage_type is None: storage_type = 'neo4j'
+#     user_rag_memory_id = ''
+#
+#     # 如果 storage_type 是 rag，必须确保有有效的 user_rag_memory_id
+#     if storage_type == 'rag':
+#         if workspace_id:
+#             knowledge = knowledge_repository.get_knowledge_by_name(
+#                 db=db,
+#                 name="USER_RAG_MERORY",
+#                 workspace_id=workspace_id
+#             )
+#             if knowledge:
+#                 user_rag_memory_id = str(knowledge.id)
+#             else:
+#                 api_logger.warning(
+#                     f"未找到名为 'USER_RAG_MERORY' 的知识库，workspace_id: {workspace_id}，将使用 neo4j 存储")
+#                 storage_type = 'neo4j'
+#         else:
+#             api_logger.warning("workspace_id 为空，无法使用 rag 存储，将使用 neo4j 存储")
+#             storage_type = 'neo4j'
+#
+#     api_logger.info(
+#         f"Write service requested for group {user_input.end_user_id}, storage_type: {storage_type}, user_rag_memory_id: {user_rag_memory_id}")
+#     try:
+#         messages_list = memory_agent_service.get_messages_list(user_input)
+#         result = await memory_agent_service.write_memory(
+#             user_input.end_user_id,
+#             messages_list,
+#             config_id,
+#             db,
+#             storage_type,
+#             user_rag_memory_id,
+#             language
+#         )
+#
+#         return success(data=result, msg="写入成功")
+#     except BaseException as e:
+#         # Handle ExceptionGroup from TaskGroup (Python 3.11+) or BaseExceptionGroup
+#         if hasattr(e, 'exceptions'):
+#             error_messages = [f"{type(sub_e).__name__}: {str(sub_e)}" for sub_e in e.exceptions]
+#             detailed_error = "; ".join(error_messages)
+#             api_logger.error(f"Write operation error (TaskGroup): {detailed_error}", exc_info=True)
+#             return fail(BizCode.INTERNAL_ERROR, "写入失败", detailed_error)
+#         api_logger.error(f"Write operation error: {str(e)}", exc_info=True)
+#         return fail(BizCode.INTERNAL_ERROR, "写入失败", str(e))
+#
+#
+# @router.post("/writer_service_async", response_model=ApiResponse)
+# @cur_workspace_access_guard()
+# async def write_server_async(
+#         user_input: Write_UserInput,
+#         language_type: str = Header(default=None, alias="X-Language-Type"),
+#         db: Session = Depends(get_db),
+#         current_user: User = Depends(get_current_user)
+# ):
+#     """
+#     Async write service endpoint - enqueues write processing to Celery
+#
+#     Args:
+#         user_input: Write request containing message and end_user_id
+#         language_type: 语言类型 ("zh" 中文, "en" 英文)，通过 X-Language-Type Header 传递
+#
+#     Returns:
+#         Task ID for tracking async operation
+#         Use GET /memory/write_result/{task_id} to check task status and get result
+#     """
+#     # 使用集中化的语言校验
+#     language = get_language_from_header(language_type)
+#
+#     config_id = user_input.config_id
+#     workspace_id = current_user.current_workspace_id
+#     api_logger.info(
+#         f"Async write service: workspace_id={workspace_id}, config_id={config_id}, language_type={language}")
+#
+#     # 获取 storage_type，如果为 None 则使用默认值
+#     storage_type = workspace_service.get_workspace_storage_type(
+#         db=db,
+#         workspace_id=workspace_id,
+#         user=current_user
+#     )
+#     if storage_type is None: storage_type = 'neo4j'
+#     user_rag_memory_id = ''
+#     if workspace_id:
+#
+#         knowledge = knowledge_repository.get_knowledge_by_name(
+#             db=db,
+#             name="USER_RAG_MERORY",
+#             workspace_id=workspace_id
+#         )
+#         if knowledge: user_rag_memory_id = str(knowledge.id)
+#     api_logger.info(f"Async write: storage_type={storage_type}, user_rag_memory_id={user_rag_memory_id}")
+#     try:
+#         # 获取标准化的消息列表
+#         messages_list = memory_agent_service.get_messages_list(user_input)
+#
+#         task = celery_app.send_task(
+#             "app.core.memory.agent.write_message",
+#             args=[user_input.end_user_id, messages_list, config_id, storage_type, user_rag_memory_id, language]
+#         )
+#         api_logger.info(f"Write task queued: {task.id}")
+#
+#         return success(data={"task_id": task.id}, msg="写入任务已提交")
+#     except Exception as e:
+#         api_logger.error(f"Async write operation failed: {str(e)}")
+#         return fail(BizCode.INTERNAL_ERROR, "写入失败", str(e))
 
 
 @router.post("/read_service", response_model=ApiResponse)

api/app/controllers/memory_dashboard_controller.py

@@ -1,3 +1,5 @@
+import asyncio
+import uuid
 from fastapi import APIRouter, Depends, HTTPException, status, Query
 from pydantic import BaseModel, Field
 from sqlalchemy.orm import Session
@@ -47,64 +49,64 @@ def get_workspace_total_end_users(
 
 @router.get("/end_users", response_model=ApiResponse)
 async def get_workspace_end_users(
+    workspace_id: Optional[uuid.UUID] = Query(None, description="工作空间ID（可选，默认当前用户工作空间）"),
+    keyword: Optional[str] = Query(None, description="搜索关键词（同时模糊匹配 other_name 和 id）"),
+    page: int = Query(1, ge=1, description="页码，从1开始"),
+    pagesize: int = Query(10, ge=1, description="每页数量"),
     db: Session = Depends(get_db),
     current_user: User = Depends(get_current_user),
 ):
     """
-    获取工作空间的宿主列表（高性能优化版本 v2）
-    
-    优化策略：
-    1. 批量查询 end_users（一次查询而非循环）
-    2. 并发查询所有用户的记忆数量（Neo4j）
-    3. RAG 模式使用批量查询（一次 SQL）
-    4. 只返回必要字段减少数据传输
-    5. 添加短期缓存减少重复查询
-    6. 并发执行配置查询和记忆数量查询
-    
-    返回格式：
-    {
-        "end_user": {"id": "uuid", "other_name": "名称"},
-        "memory_num": {"total": 数量},
-        "memory_config": {"memory_config_id": "id", "memory_config_name": "名称"}
-    }
+    获取工作空间的宿主列表（分页查询，支持模糊搜索）
+
+    返回工作空间下的宿主列表，支持分页查询和模糊搜索。
+    通过 keyword 参数同时模糊匹配 other_name 和 id 字段。
+
+    Args:
+        workspace_id: 工作空间ID（可选，默认当前用户工作空间）
+        keyword: 搜索关键词（可选，同时模糊匹配 other_name 和 id）
+        page: 页码（从1开始，默认1）
+        pagesize: 每页数量（默认10）
+        db: 数据库会话
+        current_user: 当前用户
+
+    Returns:
+        ApiResponse: 包含宿主列表和分页信息
     """
-    import asyncio
-    import json
-    from app.aioRedis import aio_redis_get, aio_redis_set
-    
-    workspace_id = current_user.current_workspace_id
-    
-    # 尝试从缓存获取（30秒缓存）
-    cache_key = f"end_users:workspace:{workspace_id}"
-    try:
-        cached_data = await aio_redis_get(cache_key)
-        if cached_data:
-            api_logger.info(f"从缓存获取宿主列表: workspace_id={workspace_id}")
-            return success(data=json.loads(cached_data), msg="宿主列表获取成功")
-    except Exception as e:
-        api_logger.warning(f"Redis 缓存读取失败: {str(e)}")
-    
+    # 如果未提供 workspace_id，使用当前用户的工作空间
+    if workspace_id is None:
+        workspace_id = current_user.current_workspace_id
     # 获取当前空间类型
     current_workspace_type = memory_dashboard_service.get_current_workspace_type(db, workspace_id, current_user)
-    api_logger.info(f"用户 {current_user.username} 请求获取工作空间 {workspace_id} 的宿主列表")
-    
-    # 获取 end_users（已优化为批量查询）
-    end_users = memory_dashboard_service.get_workspace_end_users(
+    api_logger.info(f"用户 {current_user.username} 请求获取工作空间 {workspace_id} 的宿主列表, 类型: {current_workspace_type}")
+
+    # 获取分页的 end_users
+    end_users_result = memory_dashboard_service.get_workspace_end_users_paginated(
         db=db,
         workspace_id=workspace_id,
-        current_user=current_user
+        current_user=current_user,
+        page=page,
+        pagesize=pagesize,
+        keyword=keyword
     )
+
+    end_users = end_users_result.get("items", [])
+    total = end_users_result.get("total", 0)
+
     if not end_users:
-        api_logger.info("工作空间下没有宿主")
-        # 缓存空结果，避免重复查询
-        try:
-            await aio_redis_set(cache_key, json.dumps([]), expire=30)
-        except Exception as e:
-            api_logger.warning(f"Redis 缓存写入失败: {str(e)}")
-        return success(data=[], msg="宿主列表获取成功")
-    
+        api_logger.info(f"工作空间下没有宿主或当前页无数据: total={total}, page={page}")
+        return success(data={
+            "items": [],
+            "page": {
+                "page": page,
+                "pagesize": pagesize,
+                "total": total,
+                "hasnext": (page * pagesize) < total
+            }
+        }, msg="宿主列表获取成功")
+
     end_user_ids = [str(user.id) for user in end_users]
-    
+
     # 并发执行两个独立的查询任务
     async def get_memory_configs():
         """获取记忆配置（在线程池中执行同步查询）"""
@@ -116,7 +118,7 @@ async def get_workspace_end_users(
         except Exception as e:
             api_logger.error(f"批量获取记忆配置失败: {str(e)}")
             return {}
-    
+
     async def get_memory_nums():
         """获取记忆数量"""
         if current_workspace_type == "rag":
@@ -130,26 +132,18 @@ async def get_workspace_end_users(
             except Exception as e:
                 api_logger.error(f"批量获取 RAG chunk 数量失败: {str(e)}")
                 return {uid: {"total": 0} for uid in end_user_ids}
-        
+
         elif current_workspace_type == "neo4j":
-            # Neo4j 模式：并发查询（带并发限制）
-            # 使用信号量限制并发数，避免大量用户时压垮 Neo4j
-            MAX_CONCURRENT_QUERIES = 10
-            semaphore = asyncio.Semaphore(MAX_CONCURRENT_QUERIES)
-            
-            async def get_neo4j_memory_num(end_user_id: str):
-                async with semaphore:
-                    try:
-                        return await memory_storage_service.search_all(end_user_id)
-                    except Exception as e:
-                        api_logger.error(f"获取用户 {end_user_id} Neo4j 记忆数量失败: {str(e)}")
-                        return {"total": 0}
-            
-            memory_nums_list = await asyncio.gather(*[get_neo4j_memory_num(uid) for uid in end_user_ids])
-            return {end_user_ids[i]: memory_nums_list[i] for i in range(len(end_user_ids))}
-        
+            # Neo4j 模式：批量查询（简化版本，只返回total）
+            try:
+                batch_result = await memory_storage_service.search_all_batch(end_user_ids)
+                return {uid: {"total": count} for uid, count in batch_result.items()}
+            except Exception as e:
+                api_logger.error(f"批量获取 Neo4j 记忆数量失败: {str(e)}")
+                return {uid: {"total": 0} for uid in end_user_ids}
+
         return {uid: {"total": 0} for uid in end_user_ids}
-    
+
     # 触发按需初始化：为 implicit_emotions_storage 中没有记录的用户异步生成数据
     try:
         from app.celery_app import celery_app as _celery_app
@@ -170,13 +164,13 @@ async def get_workspace_end_users(
         get_memory_configs(),
         get_memory_nums()
     )
-    
-    # 构建结果（优化：使用列表推导式）
-    result = []
+
+    # 构建结果列表
+    items = []
     for end_user in end_users:
         user_id = str(end_user.id)
         config_info = memory_configs_map.get(user_id, {})
-        result.append({
+        items.append({
             'end_user': {
                 'id': user_id,
                 'other_name': end_user.other_name
@@ -187,12 +181,6 @@ async def get_workspace_end_users(
                 "memory_config_name": config_info.get("memory_config_name")
             }
         })
-    
-    # 写入缓存（30秒过期）
-    try:
-        await aio_redis_set(cache_key, json.dumps(result), expire=30)
-    except Exception as e:
-        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）
     try:
@@ -202,7 +190,18 @@ async def get_workspace_end_users(
     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_logger.info(f"成功获取 {len(end_users)} 个宿主记录")
+    # 构建分页响应
+    result = {
+        "items": items,
+        "page": {
+            "page": page,
+            "pagesize": pagesize,
+            "total": total,
+            "hasnext": (page * pagesize) < total
+        }
+    }
+
+    api_logger.info(f"成功获取 {len(end_users)} 个宿主记录，总计 {total} 条")
     return success(data=result, msg="宿主列表获取成功")
 
 
@@ -663,9 +662,12 @@ async def dashboard_data(
                 rag_data["total_memory"] = total_chunk
                 
                 # total_app: 统计当前空间下的所有app数量
-                from app.repositories import app_repository
-                apps_orm = app_repository.get_apps_by_workspace_id(db, workspace_id)
-                rag_data["total_app"] = len(apps_orm)
+                # 包含自有app + 被分享给本工作空间的app
+                from app.services import app_service as _app_svc
+                _, total_app = _app_svc.AppService(db).list_apps(
+                    workspace_id=workspace_id, include_shared=True, pagesize=1
+                )
+                rag_data["total_app"] = total_app
                 
                 # total_knowledge: 使用 total_kb（总知识库数）
                 total_kb = memory_dashboard_service.get_rag_total_kb(db, current_user)
@@ -687,7 +689,7 @@ async def dashboard_data(
                     api_logger.warning(f"获取RAG模式API调用统计失败，使用默认值: {str(e)}")
                     rag_data["total_api_call"] = 0
                 
-                api_logger.info(f"成功获取RAG相关数据: memory={total_chunk}, app={len(apps_orm)}, knowledge={total_kb}, api_calls={rag_data['total_api_call']}")
+                api_logger.info(f"成功获取RAG相关数据: memory={total_chunk}, app={total_app}, knowledge={total_kb}, api_calls={rag_data['total_api_call']}")
             except Exception as e:
                 api_logger.warning(f"获取RAG相关数据失败: {str(e)}")
             

api/app/controllers/memory_forget_controller.py

@@ -31,6 +31,7 @@ from app.schemas.memory_storage_schema import (
     ForgettingCurveRequest,
     ForgettingCurveResponse,
     ForgettingCurvePoint,
+    PendingNodesResponse,
 )
 from app.schemas.response_schema import ApiResponse
 from app.services.memory_forget_service import MemoryForgetService
@@ -308,6 +309,100 @@ async def get_forgetting_stats(
         return fail(BizCode.INTERNAL_ERROR, "获取遗忘引擎统计失败", str(e))
 
 
+@router.get("/pending-nodes", response_model=ApiResponse)
+async def get_pending_nodes(
+    end_user_id: str,
+    page: int = 1,
+    pagesize: int = 10,
+    current_user: User = Depends(get_current_user),
+    db: Session = Depends(get_db)
+):
+    """
+    获取待遗忘节点列表（独立分页接口）
+
+    查询满足遗忘条件的节点（激活值低于阈值且最后访问时间超过最小天数）。
+    此接口独立分页，与 /stats 接口分离。
+
+    Args:
+        end_user_id: 组ID（即 end_user_id，必填）
+        page: 页码（从1开始，默认1）
+        pagesize: 每页数量（默认10）
+        current_user: 当前用户
+        db: 数据库会话
+
+    Returns:
+        ApiResponse: 包含待遗忘节点列表和分页信息的响应
+
+    Examples:
+        - 第1页，每页10条：GET /memory/forget-memory/pending-nodes?end_user_id=xxx&page=1&pagesize=10
+        - 第2页，每页20条：GET /memory/forget-memory/pending-nodes?end_user_id=xxx&page=2&pagesize=20
+
+    Notes:
+        - page 从1开始，pagesize 必须大于0
+        - 返回格式：{"items": [...], "page": {"page": 1, "pagesize": 10, "total": 100, "hasnext": true}}
+    """
+    workspace_id = current_user.current_workspace_id
+    # 检查用户是否已选择工作空间
+    if workspace_id is None:
+        api_logger.warning(f"用户 {current_user.username} 尝试获取待遗忘节点但未选择工作空间")
+        return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
+
+    # 验证 end_user_id 必填
+    if not end_user_id:
+        api_logger.warning(f"用户 {current_user.username} 尝试获取待遗忘节点但未提供 end_user_id")
+        return fail(BizCode.INVALID_PARAMETER, "end_user_id 不能为空", "end_user_id is required")
+
+    # 通过 end_user_id 获取关联的 config_id
+    try:
+        from app.services.memory_agent_service import get_end_user_connected_config
+
+        connected_config = get_end_user_connected_config(end_user_id, db)
+        config_id = connected_config.get("memory_config_id")
+        config_id = resolve_config_id(config_id, db)
+
+        if config_id is None:
+            api_logger.warning(f"终端用户 {end_user_id} 未关联记忆配置")
+            return fail(BizCode.INVALID_PARAMETER, f"终端用户 {end_user_id} 未关联记忆配置", "memory_config_id is None")
+
+        api_logger.debug(f"通过 end_user_id={end_user_id} 获取到 config_id={config_id}")
+    except ValueError as e:
+        api_logger.warning(f"获取终端用户配置失败: {str(e)}")
+        return fail(BizCode.INVALID_PARAMETER, str(e), "ValueError")
+    except Exception as e:
+        api_logger.error(f"获取终端用户配置时发生错误: {str(e)}")
+        return fail(BizCode.INTERNAL_ERROR, "获取终端用户配置失败", str(e))
+
+    # 验证分页参数
+    if page < 1:
+        return fail(BizCode.INVALID_PARAMETER, "page 必须大于等于1", "page < 1")
+    if pagesize < 1:
+        return fail(BizCode.INVALID_PARAMETER, "pagesize 必须大于等于1", "pagesize < 1")
+
+    api_logger.info(
+        f"用户 {current_user.username} 在工作空间 {workspace_id} 请求获取待遗忘节点: "
+        f"end_user_id={end_user_id}, page={page}, pagesize={pagesize}"
+    )
+
+    try:
+        # 调用服务层获取待遗忘节点列表
+        result = await forget_service.get_pending_nodes(
+            db=db,
+            end_user_id=end_user_id,
+            config_id=config_id,
+            page=page,
+            pagesize=pagesize
+        )
+
+        # 构建响应
+        response_data = PendingNodesResponse(**result)
+
+        return success(data=response_data.model_dump(), msg="查询成功")
+
+    except Exception as e:
+        api_logger.error(f"获取待遗忘节点列表失败: {str(e)}")
+        return fail(BizCode.INTERNAL_ERROR, "获取待遗忘节点列表失败", str(e))
+
+
 @router.post("/forgetting_curve", response_model=ApiResponse)
 async def get_forgetting_curve(
     request: ForgettingCurveRequest,

api/app/controllers/memory_storage_controller.py

@@ -54,8 +54,8 @@ router = APIRouter(
 
 @router.get("/info", response_model=ApiResponse)
 async def get_storage_info(
-    storage_id: str,
-    current_user: User = Depends(get_current_user)
+        storage_id: str,
+        current_user: User = Depends(get_current_user)
 ):
     """
     Example wrapper endpoint - retrieves storage information
@@ -75,24 +75,19 @@ async def get_storage_info(
         return fail(BizCode.INTERNAL_ERROR, "存储信息获取失败", str(e))
 
 
-
-
-
-
-
-@router.post("/create_config", response_model=ApiResponse)   # 创建配置文件，其他参数默认
+@router.post("/create_config", response_model=ApiResponse)  # 创建配置文件，其他参数默认
 def create_config(
-    payload: ConfigParamsCreate,
-    current_user: User = Depends(get_current_user),
-    db: Session = Depends(get_db),
-    x_language_type: Optional[str] = Header(None, alias="X-Language-Type"),
+        payload: ConfigParamsCreate,
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
+        x_language_type: Optional[str] = Header(None, alias="X-Language-Type"),
 ) -> dict:
     workspace_id = current_user.current_workspace_id
     # 检查用户是否已选择工作空间
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试创建配置但未选择工作空间")
         return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
-    
+
     api_logger.info(f"用户 {current_user.username} 在工作空间 {workspace_id} 请求创建配置: {payload.config_name}")
     try:
         # 将 workspace_id 注入到 payload 中（保持为 UUID 类型）
@@ -107,9 +102,11 @@ def create_config(
             api_logger.warning(f"重复的配置名称 '{config_name}' 在工作空间 {workspace_id}")
             lang = get_language_from_header(x_language_type)
             if lang == "en":
-                msg = fail(BizCode.BAD_REQUEST, "Config name already exists", f"A config named \"{config_name}\" already exists in the current workspace. Please use a different name.")
+                msg = fail(BizCode.BAD_REQUEST, "Config name already exists",
+                           f"A config named \"{config_name}\" already exists in the current workspace. Please use a different name.")
             else:
-                msg = fail(BizCode.BAD_REQUEST, "配置名称已存在", f"当前工作空间下已存在名为「{config_name}」的记忆配置，请使用其他名称")
+                msg = fail(BizCode.BAD_REQUEST, "配置名称已存在",
+                           f"当前工作空间下已存在名为「{config_name}」的记忆配置，请使用其他名称")
             return JSONResponse(status_code=400, content=msg)
         api_logger.error(f"Create config failed: {err_str}")
         return fail(BizCode.INTERNAL_ERROR, "创建配置失败", err_str)
@@ -119,9 +116,11 @@ def create_config(
             api_logger.warning(f"重复的配置名称 '{payload.config_name}' 在工作空间 {workspace_id}")
             lang = get_language_from_header(x_language_type)
             if lang == "en":
-                msg = fail(BizCode.BAD_REQUEST, "Config name already exists", f"A config named \"{payload.config_name}\" already exists in the current workspace. Please use a different name.")
+                msg = fail(BizCode.BAD_REQUEST, "Config name already exists",
+                           f"A config named \"{payload.config_name}\" already exists in the current workspace. Please use a different name.")
             else:
-                msg = fail(BizCode.BAD_REQUEST, "配置名称已存在", f"当前工作空间下已存在名为「{payload.config_name}」的记忆配置，请使用其他名称")
+                msg = fail(BizCode.BAD_REQUEST, "配置名称已存在",
+                           f"当前工作空间下已存在名为「{payload.config_name}」的记忆配置，请使用其他名称")
             return JSONResponse(status_code=400, content=msg)
         api_logger.error(f"Create config failed: {str(e)}")
         return fail(BizCode.INTERNAL_ERROR, "创建配置失败", str(e))
@@ -129,10 +128,10 @@ def create_config(
 
 @router.delete("/delete_config", response_model=ApiResponse)  # 删除数据库中的内容（按配置名称）
 def delete_config(
-    config_id: UUID|int,
-    force: bool = Query(False, description="是否强制删除（即使有终端用户正在使用）"),
-    current_user: User = Depends(get_current_user),
-    db: Session = Depends(get_db),
+        config_id: UUID | int,
+        force: bool = Query(False, description="是否强制删除（即使有终端用户正在使用）"),
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
 ) -> dict:
     """删除记忆配置（带终端用户保护）
     
@@ -145,24 +144,24 @@ def delete_config(
         force: 设置为 true 可强制删除（即使有终端用户正在使用）
     """
     workspace_id = current_user.current_workspace_id
-    config_id=resolve_config_id(config_id, db)
+    config_id = resolve_config_id(config_id, db)
     # 检查用户是否已选择工作空间
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试删除配置但未选择工作空间")
         return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
-    
+
     api_logger.info(
         f"用户 {current_user.username} 在工作空间 {workspace_id} 请求删除配置: "
         f"config_id={config_id}, force={force}"
     )
-    
+
     try:
         # 使用带保护的删除服务
         from app.services.memory_config_service import MemoryConfigService
-        
+
         config_service = MemoryConfigService(db)
         result = config_service.delete_config(config_id=config_id, force=force)
-        
+
         if result["status"] == "error":
             api_logger.warning(
                 f"记忆配置删除被拒绝: config_id={config_id}, reason={result['message']}"
@@ -172,7 +171,7 @@ def delete_config(
                 msg=result["message"],
                 data={"config_id": str(config_id), "is_default": result.get("is_default", False)}
             )
-        
+
         if result["status"] == "warning":
             api_logger.warning(
                 f"记忆配置正在使用，无法删除: config_id={config_id}, "
@@ -186,7 +185,7 @@ def delete_config(
                     "force_required": result["force_required"]
                 }
             )
-        
+
         api_logger.info(
             f"记忆配置删除成功: config_id={config_id}, "
             f"affected_users={result['affected_users']}"
@@ -195,7 +194,7 @@ def delete_config(
             msg=result["message"],
             data={"affected_users": result["affected_users"]}
         )
-        
+
     except Exception as e:
         api_logger.error(f"Delete config failed: {str(e)}", exc_info=True)
         return fail(BizCode.INTERNAL_ERROR, "删除配置失败", str(e))
@@ -203,9 +202,9 @@ def delete_config(
 
 @router.post("/update_config", response_model=ApiResponse)  # 更新配置文件中name和desc
 def update_config(
-    payload: ConfigUpdate,
-    current_user: User = Depends(get_current_user),
-    db: Session = Depends(get_db),
+        payload: ConfigUpdate,
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
 ) -> dict:
     workspace_id = current_user.current_workspace_id
     payload.config_id = resolve_config_id(payload.config_id, db)
@@ -213,12 +212,13 @@ def update_config(
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试更新配置但未选择工作空间")
         return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
-    
+
     # 校验至少有一个字段需要更新
     if payload.config_name is None and payload.config_desc is None and payload.scene_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试更新配置但未提供任何更新字段")
-        return fail(BizCode.INVALID_PARAMETER, "请至少提供一个需要更新的字段", "config_name, config_desc, scene_id 均为空")
-    
+        return fail(BizCode.INVALID_PARAMETER, "请至少提供一个需要更新的字段",
+                    "config_name, config_desc, scene_id 均为空")
+
     api_logger.info(f"用户 {current_user.username} 在工作空间 {workspace_id} 请求更新配置: {payload.config_id}")
     try:
         svc = DataConfigService(db)
@@ -231,9 +231,9 @@ def update_config(
 
 @router.post("/update_config_extracted", response_model=ApiResponse)  # 更新数据库中的部分内容 所有业务字段均可选
 def update_config_extracted(
-    payload: ConfigUpdateExtracted,
-    current_user: User = Depends(get_current_user),
-    db: Session = Depends(get_db),
+        payload: ConfigUpdateExtracted,
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
 ) -> dict:
     workspace_id = current_user.current_workspace_id
     payload.config_id = resolve_config_id(payload.config_id, db)
@@ -241,7 +241,7 @@ def update_config_extracted(
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试更新提取配置但未选择工作空间")
         return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
-    
+
     api_logger.info(f"用户 {current_user.username} 在工作空间 {workspace_id} 请求更新提取配置: {payload.config_id}")
     try:
         svc = DataConfigService(db)
@@ -256,11 +256,11 @@ def update_config_extracted(
 # 遗忘引擎配置接口已迁移到 memory_forget_controller.py
 # 使用新接口: /api/memory/forget/read_config 和 /api/memory/forget/update_config
 
-@router.get("/read_config_extracted", response_model=ApiResponse) # 通过查询参数读取某条配置（固定路径） 没有意义的话就删除
+@router.get("/read_config_extracted", response_model=ApiResponse)  # 通过查询参数读取某条配置（固定路径） 没有意义的话就删除
 def read_config_extracted(
-    config_id: UUID | int,
-    current_user: User = Depends(get_current_user),
-    db: Session = Depends(get_db),
+        config_id: UUID | int,
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
 ) -> dict:
     workspace_id = current_user.current_workspace_id
     config_id = resolve_config_id(config_id, db)
@@ -268,7 +268,7 @@ def read_config_extracted(
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试读取提取配置但未选择工作空间")
         return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
-    
+
     api_logger.info(f"用户 {current_user.username} 在工作空间 {workspace_id} 请求读取提取配置: {config_id}")
     try:
         svc = DataConfigService(db)
@@ -278,18 +278,19 @@ def read_config_extracted(
         api_logger.error(f"Read config extracted failed: {str(e)}")
         return fail(BizCode.INTERNAL_ERROR, "查询配置失败", str(e))
 
-@router.get("/read_all_config", response_model=ApiResponse) # 读取所有配置文件列表
+
+@router.get("/read_all_config", response_model=ApiResponse)  # 读取所有配置文件列表
 def read_all_config(
-    current_user: User = Depends(get_current_user),
-    db: Session = Depends(get_db),
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
 ) -> dict:
     workspace_id = current_user.current_workspace_id
-    
+
     # 检查用户是否已选择工作空间
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试查询配置但未选择工作空间")
         return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
-    
+
     api_logger.info(f"用户 {current_user.username} 在工作空间 {workspace_id} 请求读取所有配置")
     try:
         svc = DataConfigService(db)
@@ -303,14 +304,14 @@ def read_all_config(
 
 @router.post("/pilot_run", response_model=None)
 async def pilot_run(
-    payload: ConfigPilotRun,
-    language_type: str = Header(default=None, alias="X-Language-Type"),
-    current_user: User = Depends(get_current_user),
-    db: Session = Depends(get_db),
+        payload: ConfigPilotRun,
+        language_type: str = Header(default=None, alias="X-Language-Type"),
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
 ) -> StreamingResponse:
     # 使用集中化的语言校验
     language = get_language_from_header(language_type)
-    
+
     api_logger.info(
         f"Pilot run requested: config_id={payload.config_id}, "
         f"dialogue_text_length={len(payload.dialogue_text)}, "
@@ -333,9 +334,9 @@ async def pilot_run(
 
 @router.get("/search/kb_type_distribution", response_model=ApiResponse)
 async def get_kb_type_distribution(
-    end_user_id: Optional[str] = None,
-    current_user: User = Depends(get_current_user),
-    ) -> dict:
+        end_user_id: Optional[str] = None,
+        current_user: User = Depends(get_current_user),
+) -> dict:
     api_logger.info(f"KB type distribution requested for end_user_id: {end_user_id}")
     try:
         result = await kb_type_distribution(end_user_id)
@@ -344,12 +345,12 @@ async def get_kb_type_distribution(
         api_logger.error(f"KB type distribution failed: {str(e)}")
         return fail(BizCode.INTERNAL_ERROR, "知识库类型分布查询失败", str(e))
 
-    
+
 @router.get("/search/dialogue", response_model=ApiResponse)
 async def search_dialogues_num(
-    end_user_id: Optional[str] = None,
-    current_user: User = Depends(get_current_user),
-    ) -> dict:
+        end_user_id: Optional[str] = None,
+        current_user: User = Depends(get_current_user),
+) -> dict:
     api_logger.info(f"Search dialogue requested for end_user_id: {end_user_id}")
     try:
         result = await search_dialogue(end_user_id)
@@ -361,9 +362,9 @@ async def search_dialogues_num(
 
 @router.get("/search/chunk", response_model=ApiResponse)
 async def search_chunks_num(
-    end_user_id: Optional[str] = None,
-    current_user: User = Depends(get_current_user),
-    ) -> dict:
+        end_user_id: Optional[str] = None,
+        current_user: User = Depends(get_current_user),
+) -> dict:
     api_logger.info(f"Search chunk requested for end_user_id: {end_user_id}")
     try:
         result = await search_chunk(end_user_id)
@@ -375,9 +376,9 @@ async def search_chunks_num(
 
 @router.get("/search/statement", response_model=ApiResponse)
 async def search_statements_num(
-    end_user_id: Optional[str] = None,
-    current_user: User = Depends(get_current_user),
-    ) -> dict:
+        end_user_id: Optional[str] = None,
+        current_user: User = Depends(get_current_user),
+) -> dict:
     api_logger.info(f"Search statement requested for end_user_id: {end_user_id}")
     try:
         result = await search_statement(end_user_id)
@@ -389,9 +390,9 @@ async def search_statements_num(
 
 @router.get("/search/entity", response_model=ApiResponse)
 async def search_entities_num(
-    end_user_id: Optional[str] = None,
-    current_user: User = Depends(get_current_user),
-    ) -> dict:
+        end_user_id: Optional[str] = None,
+        current_user: User = Depends(get_current_user),
+) -> dict:
     api_logger.info(f"Search entity requested for end_user_id: {end_user_id}")
     try:
         result = await search_entity(end_user_id)
@@ -403,9 +404,9 @@ async def search_entities_num(
 
 @router.get("/search", response_model=ApiResponse)
 async def search_all_num(
-    end_user_id: Optional[str] = None,
-    current_user: User = Depends(get_current_user),
-    ) -> dict:
+        end_user_id: Optional[str] = None,
+        current_user: User = Depends(get_current_user),
+) -> dict:
     api_logger.info(f"Search all requested for end_user_id: {end_user_id}")
     try:
         result = await search_all(end_user_id)
@@ -417,9 +418,9 @@ async def search_all_num(
 
 @router.get("/search/detials", response_model=ApiResponse)
 async def search_entities_detials(
-    end_user_id: Optional[str] = None,
-    current_user: User = Depends(get_current_user),
-    ) -> dict:
+        end_user_id: Optional[str] = None,
+        current_user: User = Depends(get_current_user),
+) -> dict:
     api_logger.info(f"Search details requested for end_user_id: {end_user_id}")
     try:
         result = await search_detials(end_user_id)
@@ -431,9 +432,9 @@ async def search_entities_detials(
 
 @router.get("/search/edges", response_model=ApiResponse)
 async def search_entity_edges(
-    end_user_id: Optional[str] = None,
-    current_user: User = Depends(get_current_user),
-    ) -> dict:
+        end_user_id: Optional[str] = None,
+        current_user: User = Depends(get_current_user),
+) -> dict:
     api_logger.info(f"Search edges requested for end_user_id: {end_user_id}")
     try:
         result = await search_edges(end_user_id)
@@ -443,14 +444,12 @@ async def search_entity_edges(
         return fail(BizCode.INTERNAL_ERROR, "边查询失败", str(e))
 
 
-
-
 @router.get("/analytics/hot_memory_tags", response_model=ApiResponse)
 async def get_hot_memory_tags_api(
-    limit: int = 10,
-    db: Session = Depends(get_db),
-    current_user: User = Depends(get_current_user),
-    ) -> dict:
+        limit: int = 10,
+        db: Session = Depends(get_db),
+        current_user: User = Depends(get_current_user),
+) -> dict:
     """
     获取热门记忆标签（带Redis缓存）
     
@@ -461,18 +460,18 @@ async def get_hot_memory_tags_api(
     - 缓存未命中：~600-800ms（取决于LLM速度）
     """
     workspace_id = current_user.current_workspace_id
-    
+
     # 构建缓存键
     cache_key = f"hot_memory_tags:{workspace_id}:{limit}"
-    
+
     api_logger.info(f"Hot memory tags requested for workspace: {workspace_id}, limit: {limit}")
-    
+
     try:
         # 尝试从Redis缓存获取
         import json
 
         from app.aioRedis import aio_redis_get, aio_redis_set
-        
+
         cached_result = await aio_redis_get(cache_key)
         if cached_result:
             api_logger.info(f"Cache hit for key: {cache_key}")
@@ -481,11 +480,11 @@ async def get_hot_memory_tags_api(
                 return success(data=data, msg="查询成功（缓存）")
             except json.JSONDecodeError:
                 api_logger.warning(f"Failed to parse cached data, will refresh")
-        
+
         # 缓存未命中，执行查询
         api_logger.info(f"Cache miss for key: {cache_key}, executing query")
         result = await analytics_hot_memory_tags(db, current_user, limit)
-        
+
         # 写入缓存（过期时间：5分钟）
         # 注意：result是列表，需要转换为JSON字符串
         try:
@@ -495,9 +494,9 @@ async def get_hot_memory_tags_api(
         except Exception as cache_error:
             # 缓存写入失败不影响主流程
             api_logger.warning(f"Failed to cache result: {str(cache_error)}")
-        
+
         return success(data=result, msg="查询成功")
-        
+
     except Exception as e:
         api_logger.error(f"Hot memory tags failed: {str(e)}")
         return fail(BizCode.INTERNAL_ERROR, "热门标签查询失败", str(e))
@@ -505,8 +504,8 @@ async def get_hot_memory_tags_api(
 
 @router.delete("/analytics/hot_memory_tags/cache", response_model=ApiResponse)
 async def clear_hot_memory_tags_cache(
-    current_user: User = Depends(get_current_user),
-    ) -> dict:
+        current_user: User = Depends(get_current_user),
+) -> dict:
     """
     清除热门标签缓存
     
@@ -516,12 +515,12 @@ async def clear_hot_memory_tags_cache(
     - 数据更新后立即生效
     """
     workspace_id = current_user.current_workspace_id
-    
+
     api_logger.info(f"Clear hot memory tags cache requested for workspace: {workspace_id}")
-    
+
     try:
         from app.aioRedis import aio_redis_delete
-        
+
         # 清除所有limit的缓存（常见的limit值）
         cleared_count = 0
         for limit in [5, 10, 15, 20, 30, 50]:
@@ -530,12 +529,12 @@ async def clear_hot_memory_tags_cache(
             if result:
                 cleared_count += 1
                 api_logger.info(f"Cleared cache for key: {cache_key}")
-        
+
         return success(
-            data={"cleared_count": cleared_count}, 
+            data={"cleared_count": cleared_count},
             msg=f"成功清除 {cleared_count} 个缓存"
         )
-        
+
     except Exception as e:
         api_logger.error(f"Clear cache failed: {str(e)}")
         return fail(BizCode.INTERNAL_ERROR, "清除缓存失败", str(e))
@@ -543,7 +542,7 @@ async def clear_hot_memory_tags_cache(
 
 @router.get("/analytics/recent_activity_stats", response_model=ApiResponse)
 async def get_recent_activity_stats_api(
-    current_user: User = Depends(get_current_user),
+        current_user: User = Depends(get_current_user),
 ) -> dict:
     workspace_id = str(current_user.current_workspace_id) if current_user.current_workspace_id else None
     api_logger.info(f"Recent activity stats requested: workspace_id={workspace_id}")
@@ -553,4 +552,3 @@ async def get_recent_activity_stats_api(
     except Exception as e:
         api_logger.error(f"Recent activity stats failed: {str(e)}")
         return fail(BizCode.INTERNAL_ERROR, "最近活动统计失败", str(e))
-

api/app/controllers/model_controller.py

@@ -42,6 +42,7 @@ def get_model_strategies():
 @router.get("", response_model=ApiResponse)
 def get_model_list(
         type: Optional[list[str]] = Query(None, description="模型类型筛选（支持多个，如 ?type=LLM 或 ?type=LLM,EMBEDDING）"),
+        capability: Optional[list[str]] = Query(None, description="能力筛选（支持多个，如 ?capability=chat 或 ?capability=chat, embedding）"),
         provider: Optional[model_schema.ModelProvider] = Query(None, description="提供商筛选(基于API Key)"),
         is_active: Optional[bool] = Query(None, description="激活状态筛选"),
         is_public: Optional[bool] = Query(None, description="公开状态筛选"),
@@ -74,10 +75,21 @@ def get_model_list(
             unique_flat_type = list(dict.fromkeys(flat_type))
             type_list = [ModelType(t.lower()) for t in unique_flat_type]
 
+        capability_list = []
+        if capability is not None:
+            flat_capability = []
+            for item in capability:
+                split_items = [c.strip() for c in item.split(', ') if c.strip()]
+                flat_capability.extend(split_items)
+
+            unique_flat_capability = list(dict.fromkeys(flat_capability))
+            capability_list = unique_flat_capability
+
         api_logger.error(f"获取模型type_list: {type_list}")
         query = model_schema.ModelConfigQuery(
             type=type_list,
             provider=provider,
+            capability=capability_list,
             is_active=is_active,
             is_public=is_public,
             search=search,

api/app/controllers/public_share_controller.py

@@ -27,6 +27,7 @@ from app.services.conversation_service import ConversationService
 from app.services.release_share_service import ReleaseShareService
 from app.services.shared_chat_service import SharedChatService
 from app.services.workflow_service import WorkflowService
+from app.models.file_metadata_model import FileMetadata
 from app.utils.app_config_utils import workflow_config_4_app_release, \
     agent_config_4_app_release, multi_agent_config_4_app_release
 
@@ -259,8 +260,41 @@ def get_conversation(
     conv_service = ConversationService(db)
     messages = conv_service.get_messages(conversation_id)
 
-    # 构建响应
-    conv_dict = conversation_schema.Conversation.model_validate(conversation).model_dump()
+    file_ids = []
+    message_file_id_map = {}
+
+    # 第一次遍历：解析 audio_url，收集所有有效的 file_id
+    for idx, m in enumerate(messages):
+        if m.role == "assistant" and m.meta_data:
+            audio_url = m.meta_data.get("audio_url")
+            if not audio_url:
+                continue
+            try:
+                file_id = uuid.UUID(audio_url.rstrip("/").split("/")[-1])
+            except (ValueError, IndexError):
+                # audio_url 无法解析为 UUID，标记为 unknown
+                m.meta_data["audio_status"] = "unknown"
+                continue
+
+            file_ids.append(file_id)
+            message_file_id_map[idx] = file_id
+
+    # 批量查询所有相关的 FileMetadata
+    file_status_map = {}
+    if file_ids:
+        file_metas = (
+            db.query(FileMetadata)
+            .filter(FileMetadata.id.in_(set(file_ids)))
+            .all()
+        )
+        file_status_map = {fm.id: fm.status for fm in file_metas}
+
+    # 第二次遍历：将查询结果映射回消息
+    for idx, file_id in message_file_id_map.items():
+        m = messages[idx]
+        m.meta_data["audio_status"] = file_status_map.get(file_id, "unknown")
+
+    conv_dict = conversation_schema.Conversation.model_validate(conversation).model_dump(mode="json")
     conv_dict["messages"] = [
         conversation_schema.Message.model_validate(m) for m in messages
     ]
@@ -320,6 +354,16 @@ async def chat(
             other_id=other_id,
             original_user_id=user_id
         )
+
+        # Only extract and set memory_config_id when the end user doesn't have one yet
+        if not new_end_user.memory_config_id:
+            from app.services.memory_config_service import MemoryConfigService
+            memory_config_service = MemoryConfigService(db)
+            memory_config_id, _ = memory_config_service.extract_memory_config_id(release.type, release.config or {})
+            if memory_config_id:
+                new_end_user.memory_config_id = memory_config_id
+                db.commit()
+                db.refresh(new_end_user)
         end_user_id = str(new_end_user.id)
 
         # appid = share.app_id
@@ -410,30 +454,6 @@ async def chat(
         agent_config = agent_config_4_app_release(release)
 
         if payload.stream:
-            # async def event_generator():
-            #     async for event in service.chat_stream(
-            #         share_token=share_token,
-            #         message=payload.message,
-            #         conversation_id=conversation.id,  # 使用已创建的会话 ID
-            #         user_id=str(new_end_user.id),  # 转换为字符串
-            #         variables=payload.variables,
-            #         password=password,
-            #         web_search=payload.web_search,
-            #         memory=payload.memory,
-            #         storage_type=storage_type,
-            #         user_rag_memory_id=user_rag_memory_id
-            #     ):
-            #         yield event
-
-            # return StreamingResponse(
-            #     event_generator(),
-            #     media_type="text/event-stream",
-            #     headers={
-            #         "Cache-Control": "no-cache",
-            #         "Connection": "keep-alive",
-            #         "X-Accel-Buffering": "no"
-            #     }
-            # )
             async def event_generator():
                 async for event in app_chat_service.agnet_chat_stream(
                         message=payload.message,
@@ -459,20 +479,6 @@ async def chat(
                     "X-Accel-Buffering": "no"
                 }
             )
-        # 非流式返回
-        # result = await service.chat(
-        #     share_token=share_token,
-        #     message=payload.message,
-        #     conversation_id=conversation.id,  # 使用已创建的会话 ID
-        #     user_id=str(new_end_user.id),  # 转换为字符串
-        #     variables=payload.variables,
-        #     password=password,
-        #     web_search=payload.web_search,
-        #     memory=payload.memory,
-        #     storage_type=storage_type,
-        #     user_rag_memory_id=user_rag_memory_id
-        # )
-        # return success(data=conversation_schema.ChatResponse(**result))
         result = await app_chat_service.agnet_chat(
             message=payload.message,
             conversation_id=conversation.id,  # 使用已创建的会话 ID
@@ -531,48 +537,6 @@ async def chat(
         )
 
         return success(data=conversation_schema.ChatResponse(**result).model_dump(mode="json"))
-        # 多 Agent 流式返回
-        # if payload.stream:
-        #     async def event_generator():
-        #         async for event in service.multi_agent_chat_stream(
-        #             share_token=share_token,
-        #             message=payload.message,
-        #             conversation_id=conversation.id,  # 使用已创建的会话 ID
-        #             user_id=str(new_end_user.id),  # 转换为字符串
-        #             variables=payload.variables,
-        #             password=password,
-        #             web_search=payload.web_search,
-        #             memory=payload.memory,
-        #                 storage_type=storage_type,
-        #                 user_rag_memory_id=user_rag_memory_id
-        #         ):
-        #             yield event
-
-        #     return StreamingResponse(
-        #         event_generator(),
-        #         media_type="text/event-stream",
-        #         headers={
-        #             "Cache-Control": "no-cache",
-        #             "Connection": "keep-alive",
-        #             "X-Accel-Buffering": "no"
-        #         }
-        #     )
-
-        # # 多 Agent 非流式返回
-        # result = await service.multi_agent_chat(
-        #     share_token=share_token,
-        #     message=payload.message,
-        #     conversation_id=conversation.id,  # 使用已创建的会话 ID
-        #     user_id=str(new_end_user.id),  # 转换为字符串
-        #     variables=payload.variables,
-        #     password=password,
-        #     web_search=payload.web_search,
-        #     memory=payload.memory,
-        #     storage_type=storage_type,
-        #     user_rag_memory_id=user_rag_memory_id
-        # )
-
-        # return success(data=conversation_schema.ChatResponse(**result))
     elif app_type == AppType.WORKFLOW:
         config = workflow_config_4_app_release(release)
         if not config.id:
@@ -669,6 +633,7 @@ async def config_query(
         content = {
             "app_type": release.app.type,
             "variables": release.config.get("variables"),
+            "memory": release.config.get("memory", {}).get("enabled"),
             "features": release.config.get("features")
         }
     elif release.app.type == AppType.MULTI_AGENT:

api/app/controllers/service/app_api_controller.py

@@ -91,7 +91,7 @@ async def chat(
 
     app = app_service.get_app(api_key_auth.resource_id, api_key_auth.workspace_id)
     other_id = payload.user_id
-    workspace_id = app.workspace_id
+    workspace_id = api_key_auth.workspace_id
     end_user_repo = EndUserRepository(db)
     new_end_user = end_user_repo.get_or_create_end_user(
         app_id=app.id,

api/app/controllers/service/memory_api_controller.py

@@ -6,6 +6,8 @@ from app.core.response_utils import success
 from app.db import get_db
 from app.schemas.api_key_schema import ApiKeyAuth
 from app.schemas.memory_api_schema import (
+    CreateEndUserRequest,
+    CreateEndUserResponse,
     ListConfigsResponse,
     MemoryReadRequest,
     MemoryReadResponse,
@@ -113,3 +115,31 @@ async def list_memory_configs(
 
     logger.info(f"Listed {result['total']} configs for workspace: {api_key_auth.workspace_id}")
     return success(data=ListConfigsResponse(**result).model_dump(), msg="Configs listed successfully")
+
+
+@router.post("/end_users")
+@require_api_key(scopes=["memory"])
+async def create_end_user(
+    request: Request,
+    api_key_auth: ApiKeyAuth = None,
+    db: Session = Depends(get_db),
+):
+    """
+    Create an end user.
+    
+    Creates a new end user for the authorized workspace.
+    If an end user with the same other_id already exists, returns the existing one.
+    """
+    body = await request.json()
+    payload = CreateEndUserRequest(**body)
+    logger.info(f"Create end user request - other_id: {payload.other_id}, workspace_id: {api_key_auth.workspace_id}")
+
+    memory_api_service = MemoryAPIService(db)
+
+    result = memory_api_service.create_end_user(
+        workspace_id=api_key_auth.workspace_id,
+        other_id=payload.other_id,
+    )
+
+    logger.info(f"End user ready: {result['id']}")
+    return success(data=CreateEndUserResponse(**result).model_dump(), msg="End user created successfully")

api/app/controllers/user_controller.py

@@ -111,6 +111,18 @@ def get_current_user_info(
                 break
     
     api_logger.info(f"当前用户信息获取成功: {result.username}, 角色: {result_schema.role}, 工作空间: {result_schema.current_workspace_name}")
+
+    # 设置权限：如果用户来自 SSO Source，则使用该 Source 的 permissions；否则返回 "all" 表示拥有所有权限
+    if current_user.external_source:
+        from premium.sso.models import SSOSource
+        source = db.query(SSOSource).filter(SSOSource.source_code == current_user.external_source).first()
+        if source and source.permissions:
+            result_schema.permissions = source.permissions
+        else:
+            result_schema.permissions = []
+    else:
+        result_schema.permissions = ["all"]
+
     return success(data=result_schema, msg=t("users.info.get_success"))
 
 
@@ -135,7 +147,6 @@ def get_tenant_superusers(
     return success(data=superusers_schema, msg=t("users.list.superusers_success"))
 
 
-
 @router.get("/{user_id}", response_model=ApiResponse)
 def get_user_info_by_id(
     user_id: uuid.UUID,

api/app/controllers/user_memory_controllers.py

@@ -5,7 +5,7 @@
 from typing import Optional
 import datetime
 from sqlalchemy.orm import Session
-from fastapi import APIRouter, Depends,Header
+from fastapi import APIRouter, Depends, Header
 
 from app.db import get_db
 from app.core.language_utils import get_language_from_header
@@ -19,13 +19,15 @@ from app.services.user_memory_service import (
     analytics_graph_data,
     analytics_community_graph_data,
 )
-from app.services.memory_entity_relationship_service import MemoryEntityService,MemoryEmotion,MemoryInteraction
+from app.services.memory_entity_relationship_service import MemoryEntityService, MemoryEmotion, MemoryInteraction
 from app.schemas.response_schema import ApiResponse
 from app.schemas.memory_storage_schema import GenerateCacheRequest
 from app.repositories.workspace_repository import WorkspaceRepository
-from app.schemas.end_user_schema import (
-    EndUserProfileResponse,
-    EndUserProfileUpdate,
+from app.repositories.end_user_repository import EndUserRepository
+from app.schemas.end_user_info_schema import (
+    EndUserInfoResponse,
+    EndUserInfoCreate,
+    EndUserInfoUpdate,
 )
 from app.models.end_user_model import EndUser
 from app.dependencies import get_current_user
@@ -45,9 +47,9 @@ router = APIRouter(
 
 @router.get("/analytics/memory_insight/report", response_model=ApiResponse)
 async def get_memory_insight_report_api(
-    end_user_id: str,
-    current_user: User = Depends(get_current_user),
-    db: Session = Depends(get_db),
+        end_user_id: str,
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
 ) -> dict:
     """
     获取缓存的记忆洞察报告
@@ -73,10 +75,10 @@ async def get_memory_insight_report_api(
 
 @router.get("/analytics/user_summary", response_model=ApiResponse)
 async def get_user_summary_api(
-    end_user_id: str,
-    language_type: str = Header(default=None, alias="X-Language-Type"),
-    current_user: User = Depends(get_current_user),
-    db: Session = Depends(get_db),
+        end_user_id: str,
+        language_type: str = Header(default=None, alias="X-Language-Type"),
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
 ) -> dict:
     """
     获取缓存的用户摘要
@@ -90,7 +92,7 @@ async def get_user_summary_api(
     """
     # 使用集中化的语言校验
     language = get_language_from_header(language_type)
-    
+
     workspace_id = current_user.current_workspace_id
     workspace_repo = WorkspaceRepository(db)
     workspace_models = workspace_repo.get_workspace_models_configs(workspace_id)
@@ -102,7 +104,7 @@ async def get_user_summary_api(
     api_logger.info(f"用户摘要查询请求: end_user_id={end_user_id}, user={current_user.username}")
     try:
         # 调用服务层获取缓存数据
-        result = await user_memory_service.get_cached_user_summary(db, end_user_id,model_id,language)
+        result = await user_memory_service.get_cached_user_summary(db, end_user_id, model_id, language)
 
         if result["is_cached"]:
             api_logger.info(f"成功返回缓存的用户摘要: end_user_id={end_user_id}")
@@ -117,10 +119,10 @@ async def get_user_summary_api(
 
 @router.post("/analytics/generate_cache", response_model=ApiResponse)
 async def generate_cache_api(
-    request: GenerateCacheRequest,
-    language_type: str = Header(default=None, alias="X-Language-Type"),
-    current_user: User = Depends(get_current_user),
-    db: Session = Depends(get_db),
+        request: GenerateCacheRequest,
+        language_type: str = Header(default=None, alias="X-Language-Type"),
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
 ) -> dict:
     """
     手动触发缓存生成
@@ -134,7 +136,7 @@ async def generate_cache_api(
     """
     # 使用集中化的语言校验
     language = get_language_from_header(language_type)
-    
+
     workspace_id = current_user.current_workspace_id
 
     # 检查用户是否已选择工作空间
@@ -155,10 +157,12 @@ async def generate_cache_api(
             api_logger.info(f"开始为单个用户生成缓存: end_user_id={end_user_id}")
 
             # 生成记忆洞察
-            insight_result = await user_memory_service.generate_and_cache_insight(db, end_user_id, workspace_id, language=language)
+            insight_result = await user_memory_service.generate_and_cache_insight(db, end_user_id, workspace_id,
+                                                                                  language=language)
 
             # 生成用户摘要
-            summary_result = await user_memory_service.generate_and_cache_summary(db, end_user_id, workspace_id, language=language)
+            summary_result = await user_memory_service.generate_and_cache_summary(db, end_user_id, workspace_id,
+                                                                                  language=language)
 
             # 构建响应
             result = {
@@ -209,9 +213,9 @@ async def generate_cache_api(
 
 @router.get("/analytics/node_statistics", response_model=ApiResponse)
 async def get_node_statistics_api(
-    end_user_id: str,
-    current_user: User = Depends(get_current_user),
-    db: Session = Depends(get_db),
+        end_user_id: str,
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
 ) -> dict:
     workspace_id = current_user.current_workspace_id
 
@@ -220,7 +224,8 @@ async def get_node_statistics_api(
         api_logger.warning(f"用户 {current_user.username} 尝试查询节点统计但未选择工作空间")
         return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
 
-    api_logger.info(f"记忆类型统计请求: end_user_id={end_user_id}, user={current_user.username}, workspace={workspace_id}")
+    api_logger.info(
+        f"记忆类型统计请求: end_user_id={end_user_id}, user={current_user.username}, workspace={workspace_id}")
 
     try:
         # 调用新的记忆类型统计函数
@@ -228,21 +233,23 @@ async def get_node_statistics_api(
 
         # 计算总数用于日志
         total_count = sum(item["count"] for item in result)
-        api_logger.info(f"成功获取记忆类型统计: end_user_id={end_user_id}, 总记忆数={total_count}, 类型数={len(result)}")
+        api_logger.info(
+            f"成功获取记忆类型统计: end_user_id={end_user_id}, 总记忆数={total_count}, 类型数={len(result)}")
         return success(data=result, msg="查询成功")
     except Exception as e:
         api_logger.error(f"记忆类型查询失败: end_user_id={end_user_id}, error={str(e)}")
         return fail(BizCode.INTERNAL_ERROR, "记忆类型查询失败", str(e))
 
+
 @router.get("/analytics/graph_data", response_model=ApiResponse)
 async def get_graph_data_api(
-    end_user_id: str,
-    node_types: Optional[str] = None,
-    limit: int = 100,
-    depth: int = 1,
-    center_node_id: Optional[str] = None,
-    current_user: User = Depends(get_current_user),
-    db: Session = Depends(get_db),
+        end_user_id: str,
+        node_types: Optional[str] = None,
+        limit: int = 100,
+        depth: int = 1,
+        center_node_id: Optional[str] = None,
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
 ) -> dict:
     workspace_id = current_user.current_workspace_id
 
@@ -298,9 +305,9 @@ async def get_graph_data_api(
 
 @router.get("/analytics/community_graph", response_model=ApiResponse)
 async def get_community_graph_data_api(
-    end_user_id: str,
-    current_user: User = Depends(get_current_user),
-    db: Session = Depends(get_db),
+        end_user_id: str,
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
 ) -> dict:
     workspace_id = current_user.current_workspace_id
 
@@ -331,111 +338,130 @@ async def get_community_graph_data_api(
         api_logger.error(f"社区图谱查询失败: end_user_id={end_user_id}, error={str(e)}")
         return fail(BizCode.INTERNAL_ERROR, "社区图谱查询失败", str(e))
 
+#=======================终端用户信息接口=======================
 
-@router.get("/read_end_user/profile", response_model=ApiResponse)
-async def get_end_user_profile(
+@router.get("/end_user_info", response_model=ApiResponse)
+async def get_end_user_info(
     end_user_id: str,
     current_user: User = Depends(get_current_user),
     db: Session = Depends(get_db),
 ) -> dict:
-    workspace_id = current_user.current_workspace_id
-    workspace_repo = WorkspaceRepository(db)
-    workspace_models = workspace_repo.get_workspace_models_configs(workspace_id)
+    """
+    查询终端用户信息记录
+
+    根据 end_user_id 查询单条终端用户信息记录。
+    """
+    workspace_id = current_user.current_workspace_id
 
-    if workspace_models:
-        model_id = workspace_models.get("llm", None)
-    else:
-        model_id = None
-    # 检查用户是否已选择工作空间
     if workspace_id is None:
-        api_logger.warning(f"用户 {current_user.username} 尝试查询用户信息但未选择工作空间")
+        api_logger.warning(f"用户 {current_user.username} 尝试查询终端用户信息但未选择工作空间")
         return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
 
     api_logger.info(
-        f"用户信息查询请求: end_user_id={end_user_id}, user={current_user.username}, "
+        f"查询终端用户信息请求: end_user_id={end_user_id}, user={current_user.username}, "
         f"workspace={workspace_id}"
     )
 
-    try:
-        # 查询终端用户
-        end_user = db.query(EndUser).filter(EndUser.id == end_user_id).first()
-
-        if not end_user:
-            api_logger.warning(f"终端用户不存在: end_user_id={end_user_id}")
-            return fail(BizCode.INVALID_PARAMETER, "终端用户不存在", f"end_user_id={end_user_id}")
-        # 构建响应数据
-        profile_data = EndUserProfileResponse(
-            id=end_user.id,
-            other_name=end_user.other_name,
-            position=end_user.position,
-            department=end_user.department,
-            contact=end_user.contact,
-            phone=end_user.phone,
-            hire_date=end_user.hire_date,
-            updatetime_profile=end_user.updatetime_profile
+    # 校验 end_user 是否属于当前工作空间
+    end_user_repo = EndUserRepository(db)
+    end_user = end_user_repo.get_end_user_by_id(end_user_id)
+    if end_user is None:
+        return fail(BizCode.USER_NOT_FOUND, "终端用户不存在", "end_user not found")
+    if str(end_user.workspace_id) != str(workspace_id):
+        api_logger.warning(
+            f"用户 {current_user.username} 尝试查询不属于工作空间 {workspace_id} 的终端用户 {end_user_id}"
         )
+        return fail(BizCode.PERMISSION_DENIED, "该终端用户不属于当前工作空间", "end_user workspace mismatch")
 
-        api_logger.info(f"成功获取用户信息: end_user_id={end_user_id}")
-        return success(data=UserMemoryService.convert_profile_to_dict_with_timestamp(profile_data), msg="查询成功")
+    result = user_memory_service.get_end_user_info(db, end_user_id)
 
-    except Exception as e:
-        api_logger.error(f"用户信息查询失败: end_user_id={end_user_id}, error={str(e)}")
-        return fail(BizCode.INTERNAL_ERROR, "用户信息查询失败", str(e))
+    if result["success"]:
+        api_logger.info(f"成功查询终端用户信息: end_user_id={end_user_id}")
+        return success(data=result["data"], msg="查询成功")
+    else:
+        error_msg = result["error"]
+        api_logger.error(f"查询终端用户信息失败: end_user_id={end_user_id}, error={error_msg}")
+        
+        if error_msg == "终端用户信息记录不存在":
+            return fail(BizCode.USER_NOT_FOUND, "终端用户信息记录不存在", error_msg)
+        elif error_msg == "无效的终端用户ID格式":
+            return fail(BizCode.INVALID_USER_ID, "无效的终端用户ID格式", error_msg)
+        else:
+            return fail(BizCode.INTERNAL_ERROR, "查询终端用户信息失败", error_msg)
 
 
-@router.post("/updated_end_user/profile", response_model=ApiResponse)
-async def update_end_user_profile(
-    profile_update: EndUserProfileUpdate,
+@router.post("/end_user_info/updated", response_model=ApiResponse)
+async def update_end_user_info(
+    info_update: EndUserInfoUpdate,
     current_user: User = Depends(get_current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     """
-    更新终端用户的基本信息
+    更新终端用户信息记录
 
-    该接口可以更新用户的姓名、职位、部门、联系方式、电话和入职日期等信息。
-    所有字段都是可选的，只更新提供的字段。
+    根据 end_user_id 更新终端用户信息记录，支持批量更新多个别名。
+    
+    示例请求体：
+    {
+      "end_user_id": "a1b2c3d4-e5f6-7890-abcd-ef1234567890",
+      "other_name": "张三1",
+      "aliases": ["小张", "张工"],
+      "meta_data": {"position": "工程师", "department": "技术部"}
+    }
     """
     workspace_id = current_user.current_workspace_id
-    end_user_id = profile_update.end_user_id
+    end_user_id = info_update.end_user_id
 
-    # 验证工作空间
     if workspace_id is None:
-        api_logger.warning(f"用户 {current_user.username} 尝试更新用户信息但未选择工作空间")
+        api_logger.warning(f"用户 {current_user.username} 尝试更新终端用户信息但未选择工作空间")
         return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
 
     api_logger.info(
-        f"用户信息更新请求: end_user_id={end_user_id}, user={current_user.username}, "
+        f"更新终端用户信息请求: end_user_id={end_user_id}, user={current_user.username}, "
         f"workspace={workspace_id}"
     )
 
-    # 调用 Service 层处理业务逻辑
-    result = user_memory_service.update_end_user_profile(db, end_user_id, profile_update)
+    # 校验 end_user 是否属于当前工作空间
+    end_user_repo = EndUserRepository(db)
+    end_user = end_user_repo.get_end_user_by_id(end_user_id)
+    if end_user is None:
+        return fail(BizCode.USER_NOT_FOUND, "终端用户不存在", "end_user not found")
+    if str(end_user.workspace_id) != str(workspace_id):
+        api_logger.warning(
+            f"用户 {current_user.username} 尝试更新不属于工作空间 {workspace_id} 的终端用户 {end_user_id}"
+        )
+        return fail(BizCode.PERMISSION_DENIED, "该终端用户不属于当前工作空间", "end_user workspace mismatch")
+
+    # 获取更新数据（排除 end_user_id）
+    update_data = info_update.model_dump(exclude_unset=True, exclude={'end_user_id'})
+    
+    result = user_memory_service.update_end_user_info(db, end_user_id, update_data)
 
     if result["success"]:
-        api_logger.info(f"成功更新用户信息: end_user_id={end_user_id}")
+        api_logger.info(f"成功更新终端用户信息: end_user_id={end_user_id}")
         return success(data=result["data"], msg="更新成功")
     else:
         error_msg = result["error"]
-        api_logger.error(f"用户信息更新失败: end_user_id={end_user_id}, error={error_msg}")
+        api_logger.error(f"终端用户信息更新失败: end_user_id={end_user_id}, error={error_msg}")
         
-        # 根据错误类型映射到合适的业务错误码
-        if error_msg == "终端用户不存在":
-            return fail(BizCode.USER_NOT_FOUND, "终端用户不存在", error_msg)
-        elif error_msg == "无效的用户ID格式":
-            return fail(BizCode.INVALID_USER_ID, "无效的用户ID格式", error_msg)
+        if error_msg == "终端用户信息记录不存在":
+            return fail(BizCode.USER_NOT_FOUND, "终端用户信息记录不存在", error_msg)
+        elif error_msg == "无效的终端用户ID格式":
+            return fail(BizCode.INVALID_USER_ID, "无效的终端用户ID格式", error_msg)
         else:
-            # 只有未预期的错误才使用 INTERNAL_ERROR
-            return fail(BizCode.INTERNAL_ERROR, "用户信息更新失败", error_msg)
+            return fail(BizCode.INTERNAL_ERROR, "终端用户信息更新失败", error_msg)
 
 @router.get("/memory_space/timeline_memories", response_model=ApiResponse)
-async def memory_space_timeline_of_shared_memories(id: str, label: str,language_type: str = Header(default=None, alias="X-Language-Type"),
-                                      current_user: User = Depends(get_current_user),
-                                      db: Session = Depends(get_db),
-                                      ):
+async def memory_space_timeline_of_shared_memories(
+        id: str, label: str,
+        language_type: str = Header(default=None, alias="X-Language-Type"),
+        current_user: User = Depends(get_current_user),
+        db: Session = Depends(get_db),
+):
     # 使用集中化的语言校验
     language = get_language_from_header(language_type)
-    
-    workspace_id=current_user.current_workspace_id
+
+    workspace_id = current_user.current_workspace_id
     workspace_repo = WorkspaceRepository(db)
     workspace_models = workspace_repo.get_workspace_models_configs(workspace_id)
 
@@ -447,11 +473,13 @@ async def memory_space_timeline_of_shared_memories(id: str, label: str,language_
     timeline_memories_result = await MemoryEntity.get_timeline_memories_server(model_id, language)
 
     return success(data=timeline_memories_result, msg="共同记忆时间线")
+
+
 @router.get("/memory_space/relationship_evolution", response_model=ApiResponse)
 async def memory_space_relationship_evolution(id: str, label: str,
-                                      current_user: User = Depends(get_current_user),
-                                      db: Session = Depends(get_db),
-                                      ):
+                                              current_user: User = Depends(get_current_user),
+                                              db: Session = Depends(get_db),
+                                              ):
     try:
         api_logger.info(f"关系演变查询请求: id={id}, table={label}, user={current_user.username}")
 

api/app/core/agent/langchain_agent.py

@@ -11,18 +11,14 @@ LangChain Agent 封装
 import time
 from typing import Any, AsyncGenerator, Dict, List, Optional, Sequence
 
-from app.core.memory.agent.langgraph_graph.write_graph import write_long_term
-from app.db import get_db
-from app.core.logging_config import get_business_logger
-from app.core.models import RedBearLLM, RedBearModelConfig
-from app.models.models_model import ModelType, ModelProvider
-from app.services.memory_agent_service import (
-    get_end_user_connected_config,
-)
 from langchain.agents import create_agent
 from langchain_core.messages import AIMessage, BaseMessage, HumanMessage, SystemMessage
 from langchain_core.tools import BaseTool
 
+from app.core.logging_config import get_business_logger
+from app.core.models import RedBearLLM, RedBearModelConfig
+from app.models.models_model import ModelType
+
 logger = get_business_logger()
 
 
@@ -226,10 +222,9 @@ class LangChainAgent:
         Returns:
             List[BaseMessage]: 消息列表
         """
-        messages = []
+        messages:list = [SystemMessage(content=self.system_prompt)]
 
         # 添加系统提示词
-        messages.append(SystemMessage(content=self.system_prompt))
 
         # 添加历史消息
         if history:
@@ -293,12 +288,7 @@ class LangChainAgent:
             message: str,
             history: Optional[List[Dict[str, str]]] = None,
             context: Optional[str] = None,
-            end_user_id: Optional[str] = None,
-            config_id: Optional[str] = None,  # 添加这个参数
-            storage_type: Optional[str] = None,
-            user_rag_memory_id: Optional[str] = None,
-            memory_flag: Optional[bool] = True,
-            files: Optional[List[Dict[str, Any]]] = None  # 新增：多模态文件
+            files: Optional[List[Dict[str, Any]]] = None
     ) -> Dict[str, Any]:
         """执行对话
 
@@ -306,32 +296,12 @@ class LangChainAgent:
             message: 用户消息
             history: 历史消息列表 [{"role": "user/assistant", "content": "..."}]
             context: 上下文信息（如知识库检索结果）
+            files: 多模态文件
 
         Returns:
             Dict: 包含 content 和元数据的字典
         """
-        message_chat = message
         start_time = time.time()
-        actual_config_id = config_id
-        # If config_id is None, try to get from end_user's connected config
-        if actual_config_id is None and end_user_id:
-            try:
-                from app.services.memory_agent_service import (
-                    get_end_user_connected_config,
-                )
-                db = next(get_db())
-                try:
-                    connected_config = get_end_user_connected_config(end_user_id, db)
-                    actual_config_id = connected_config.get("memory_config_id")
-                except Exception as e:
-                    logger.warning(f"Failed to get connected config for end_user {end_user_id}: {e}")
-                finally:
-                    db.close()
-            except Exception as e:
-                logger.warning(f"Failed to get db session: {e}")
-        actual_end_user_id = end_user_id if end_user_id is not None else "unknown"
-        logger.info(f'写入类型{storage_type, str(end_user_id), message, str(user_rag_memory_id)}')
-        print(f'写入类型{storage_type, str(end_user_id), message, str(user_rag_memory_id)}')
         try:
             # 准备消息列表（支持多模态）
             messages = self._prepare_messages(message, history, context, files)
@@ -419,9 +389,6 @@ class LangChainAgent:
             logger.info(f"最终提取的内容长度: {len(content)}")
 
             elapsed_time = time.time() - start_time
-            if memory_flag:
-                await write_long_term(storage_type, end_user_id, message_chat, content, user_rag_memory_id,
-                                      actual_config_id)
             response = {
                 "content": content,
                 "model": self.model_name,
@@ -452,12 +419,7 @@ class LangChainAgent:
             message: str,
             history: Optional[List[Dict[str, str]]] = None,
             context: Optional[str] = None,
-            end_user_id: Optional[str] = None,
-            config_id: Optional[str] = None,
-            storage_type: Optional[str] = None,
-            user_rag_memory_id: Optional[str] = None,
-            memory_flag: Optional[bool] = True,
-            files: Optional[List[Dict[str, Any]]] = None  # 新增：多模态文件
+            files: Optional[List[Dict[str, Any]]] = None
     ) -> AsyncGenerator[str, None]:
         """执行流式对话
 
@@ -465,6 +427,7 @@ class LangChainAgent:
             message: 用户消息
             history: 历史消息列表
             context: 上下文信息
+            files: 多模态文件
 
         Yields:
             str: 消息内容块
@@ -475,23 +438,6 @@ class LangChainAgent:
         logger.info(f"  Has tools: {bool(self.tools)}")
         logger.info(f"  Tool count: {len(self.tools) if self.tools else 0}")
         logger.info("=" * 80)
-        message_chat = message
-        actual_config_id = config_id
-        # If config_id is None, try to get from end_user's connected config
-        if actual_config_id is None and end_user_id:
-            try:
-                db = next(get_db())
-                try:
-                    connected_config = get_end_user_connected_config(end_user_id, db)
-                    actual_config_id = connected_config.get("memory_config_id")
-                except Exception as e:
-                    logger.warning(f"Failed to get connected config for end_user {end_user_id}: {e}")
-                finally:
-                    db.close()
-            except Exception as e:
-                logger.warning(f"Failed to get db session: {e}")
-
-            # 注意：不在这里写入用户消息，等 AI 回复后一起写入
         try:
             # 准备消息列表（支持多模态）
             messages = self._prepare_messages(message, history, context, files)
@@ -501,17 +447,18 @@ class LangChainAgent:
             )
 
             chunk_count = 0
-            yielded_content = False
 
             # 统一使用 agent 的 astream_events 实现流式输出
             logger.debug("使用 Agent astream_events 实现流式输出")
             full_content = ''
             try:
+                last_event = {}
                 async for event in self.agent.astream_events(
                         {"messages": messages},
                         version="v2",
                         config={"recursion_limit": self.max_iterations}
                 ):
+                    last_event = event
                     chunk_count += 1
                     kind = event.get("event")
 
@@ -525,7 +472,6 @@ class LangChainAgent:
                             if isinstance(chunk_content, str) and chunk_content:
                                 full_content += chunk_content
                                 yield chunk_content
-                                yielded_content = True
                             elif isinstance(chunk_content, list):
                                 # 多模态响应：提取文本部分
                                 for item in chunk_content:
@@ -536,18 +482,15 @@ class LangChainAgent:
                                             if text:
                                                 full_content += text
                                                 yield text
-                                                yielded_content = True
                                         # OpenAI 格式: {"type": "text", "text": "..."}
                                         elif item.get("type") == "text":
                                             text = item.get("text", "")
                                             if text:
                                                 full_content += text
                                                 yield text
-                                                yielded_content = True
                                     elif isinstance(item, str):
                                         full_content += item
                                         yield item
-                                        yielded_content = True
 
                     elif kind == "on_llm_stream":
                         # 另一种 LLM 流式事件
@@ -558,7 +501,6 @@ class LangChainAgent:
                                 if isinstance(chunk_content, str) and chunk_content:
                                     full_content += chunk_content
                                     yield chunk_content
-                                    yielded_content = True
                                 elif isinstance(chunk_content, list):
                                     # 多模态响应：提取文本部分
                                     for item in chunk_content:
@@ -569,22 +511,18 @@ class LangChainAgent:
                                                 if text:
                                                     full_content += text
                                                     yield text
-                                                    yielded_content = True
                                             # OpenAI 格式: {"type": "text", "text": "..."}
                                             elif item.get("type") == "text":
                                                 text = item.get("text", "")
                                                 if text:
                                                     full_content += text
                                                     yield text
-                                                    yielded_content = True
                                         elif isinstance(item, str):
                                             full_content += item
                                             yield item
-                                            yielded_content = True
                             elif isinstance(chunk, str):
                                 full_content += chunk
                                 yield chunk
-                                yielded_content = True
 
                     # 记录工具调用（可选）
                     elif kind == "on_tool_start":
@@ -594,17 +532,17 @@ class LangChainAgent:
 
                 logger.debug(f"Agent 流式完成，共 {chunk_count} 个事件")
                 # 统计token消耗
-                output_messages = event.get("data", {}).get("output", {}).get("messages", [])
+                output_messages = last_event.get("data", {}).get("output", {}).get("messages", [])
                 for msg in reversed(output_messages):
                     if isinstance(msg, AIMessage):
                         response_meta = msg.response_metadata if hasattr(msg, 'response_metadata') else None
-                        total_tokens = response_meta.get("token_usage", {}).get("total_tokens",
-                                                                                0) if response_meta else 0
+                        total_tokens = response_meta.get("token_usage", {}).get(
+                            "total_tokens",
+                            0
+                        ) if response_meta else 0
                         yield total_tokens
                         break
-                if memory_flag:
-                    await write_long_term(storage_type, end_user_id, message_chat, full_content, user_rag_memory_id,
-                                          actual_config_id)
+
             except Exception as e:
                 logger.error(f"Agent astream_events 失败: {str(e)}", exc_info=True)
                 raise

api/app/core/config.py

@@ -231,8 +231,8 @@ class Settings:
     # Celery configuration (internal)
     # NOTE: 变量名不以 CELERY_ 开头，避免被 Celery CLI 的前缀匹配机制劫持
     # 详见 docs/celery-env-bug-report.md
-    # 默认使用 Redis DB 3 (broker) 和 DB 4 (backend)，与业务缓存 (DB 1/2) 隔离
-    # 多人共用同一 Redis 时，每位开发者应在 .env 中配置不同的 DB 编号避免任务互相干扰
+    # 默认使用 Redis 作为 broker 和 backend，与业务缓存隔离
+    # 如需使用 RabbitMQ，在 .env 中设置 CELERY_BROKER_URL=amqp://user:pass@host:5672/vhost
     REDIS_DB_CELERY_BROKER: int = int(os.getenv("REDIS_DB_CELERY_BROKER", "3"))
     REDIS_DB_CELERY_BACKEND: int = int(os.getenv("REDIS_DB_CELERY_BACKEND", "4"))
 

api/app/core/memory/agent/langgraph_graph/read_graph.py

@@ -15,7 +15,7 @@ from app.core.memory.agent.langgraph_graph.nodes.problem_nodes import (
     Problem_Extension,
 )
 from app.core.memory.agent.langgraph_graph.nodes.retrieve_nodes import (
-    retrieve,
+    retrieve_nodes,
 )
 from app.core.memory.agent.langgraph_graph.nodes.summary_nodes import (
     Input_Summary,
@@ -53,8 +53,8 @@ async def make_read_graph():
         workflow.add_node("Split_The_Problem", Split_The_Problem)
         workflow.add_node("Problem_Extension", Problem_Extension)
         workflow.add_node("Input_Summary", Input_Summary)
-        # workflow.add_node("Retrieve", retrieve_nodes)
-        workflow.add_node("Retrieve", retrieve)
+        workflow.add_node("Retrieve", retrieve_nodes)
+        # workflow.add_node("Retrieve", retrieve)
         workflow.add_node("Verify", Verify)
         workflow.add_node("Retrieve_Summary", Retrieve_Summary)
         workflow.add_node("Summary", Summary)

api/app/core/memory/agent/langgraph_graph/routing/write_router.py

@@ -12,7 +12,6 @@ from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
 from app.db import get_db_context
 from app.repositories.memory_short_repository import LongTermMemoryRepository
 from app.schemas.memory_agent_schema import AgentMemory_Long_Term
-from app.services.memory_konwledges_server import write_rag
 from app.services.task_service import get_task_memory_write_result
 from app.tasks import write_message_task
 from app.utils.config_utils import resolve_config_id
@@ -21,25 +20,6 @@ logger = get_agent_logger(__name__)
 template_root = os.path.join(PROJECT_ROOT_, 'memory', 'agent', 'utils', 'prompt')
 
 
-async def write_rag_agent(end_user_id, user_message, ai_message, user_rag_memory_id):
-    """
-    Write messages to RAG storage system
-
-    Combines user and AI messages into a single string format and stores them
-    in the RAG (Retrieval-Augmented Generation) knowledge base for future retrieval.
-
-    Args:
-        end_user_id: User identifier for the conversation
-        user_message: User's input message content
-        ai_message: AI's response message content
-        user_rag_memory_id: RAG memory identifier for storage location
-    """
-    # RAG mode: combine messages into string format (maintain original logic)
-    combined_message = f"user: {user_message}\nassistant: {ai_message}"
-    await write_rag(end_user_id, combined_message, user_rag_memory_id)
-    logger.info(f'RAG_Agent:{end_user_id};{user_rag_memory_id}')
-
-
 async def write(
         storage_type,
         end_user_id,
@@ -118,7 +98,7 @@ async def write(
         logger.info(f'[WRITE] Task result - user={actual_end_user_id}, status={write_status}')
 
 
-async def term_memory_save(long_term_messages, actual_config_id, end_user_id, type, scope):
+async def term_memory_save(end_user_id, strategy_type, scope):
     """
     Save long-term memory data to database
 
@@ -127,10 +107,8 @@ async def term_memory_save(long_term_messages, actual_config_id, end_user_id, ty
     to long-term memory storage.
 
     Args:
-        long_term_messages: Long-term message data to be saved
-        actual_config_id: Configuration identifier for memory settings
         end_user_id: User identifier for memory association
-        type: Memory storage strategy type (STRATEGY_CHUNK or STRATEGY_AGGREGATE)
+        strategy_type: Memory storage strategy type (STRATEGY_CHUNK or STRATEGY_AGGREGATE)
         scope: Scope/window size for memory processing
     """
     with get_db_context() as db_session:
@@ -138,7 +116,10 @@ async def term_memory_save(long_term_messages, actual_config_id, end_user_id, ty
 
         from app.core.memory.agent.utils.redis_tool import write_store
         result = write_store.get_session_by_userid(end_user_id)
-        if type == AgentMemory_Long_Term.STRATEGY_CHUNK or AgentMemory_Long_Term.STRATEGY_AGGREGATE:
+        if not result:
+            logger.warning(f"No write data found for user {end_user_id}")
+            return
+        if strategy_type in [AgentMemory_Long_Term.STRATEGY_CHUNK, AgentMemory_Long_Term.STRATEGY_AGGREGATE]:
             data = await format_parsing(result, "dict")
             chunk_data = data[:scope]
             if len(chunk_data) == scope:
@@ -151,9 +132,6 @@ async def term_memory_save(long_term_messages, actual_config_id, end_user_id, ty
             logger.info(f'写入短长期：')
 
 
-"""Window-based dialogue processing"""
-
-
 async def window_dialogue(end_user_id, langchain_messages, memory_config, scope):
     """
     Process dialogue based on window size and write to Neo4j
@@ -167,40 +145,33 @@ async def window_dialogue(end_user_id, langchain_messages, memory_config, scope)
         langchain_messages: Original message data list
         scope: Window size determining when to trigger long-term storage
     """
-    scope = scope
-    is_end_user_id = count_store.get_sessions_count(end_user_id)
-    if is_end_user_id is not False:
-        is_end_user_id = count_store.get_sessions_count(end_user_id)[0]
-        redis_messages = count_store.get_sessions_count(end_user_id)[1]
-    if is_end_user_id and int(is_end_user_id) != int(scope):
-        is_end_user_id += 1
-        langchain_messages += redis_messages
-        count_store.update_sessions_count(end_user_id, is_end_user_id, langchain_messages)
-    elif int(is_end_user_id) == int(scope):
+    is_end_user_has_history = count_store.get_sessions_count(end_user_id)
+    if is_end_user_has_history:
+        end_user_visit_count, redis_messages = is_end_user_has_history
+    else:
+        count_store.save_sessions_count(end_user_id, 1, langchain_messages)
+        return
+    end_user_visit_count += 1
+    if end_user_visit_count < scope:
+        redis_messages.extend(langchain_messages)
+        count_store.update_sessions_count(end_user_id, end_user_visit_count, redis_messages)
+    else:
         logger.info('写入长期记忆NEO4J')
-        formatted_messages = (redis_messages)
+        redis_messages.extend(langchain_messages)
         # Get config_id (if memory_config is an object, extract config_id; otherwise use directly)
         if hasattr(memory_config, 'config_id'):
             config_id = memory_config.config_id
         else:
             config_id = memory_config
 
-        await write(
-            AgentMemory_Long_Term.STORAGE_NEO4J,
-            end_user_id,
-            "",
-            "",
-            None,
-            end_user_id,
-            config_id,
-            formatted_messages
+        write_message_task.delay(
+            end_user_id,  # end_user_id: User ID
+            redis_messages,  # message: JSON string format message list
+            config_id,  # config_id: Configuration ID string
+            AgentMemory_Long_Term.STORAGE_NEO4J,  # storage_type: "neo4j"
+            ""  # user_rag_memory_id: RAG memory ID (not used in Neo4j mode)
         )
-        count_store.update_sessions_count(end_user_id, 1, langchain_messages)
-    else:
-        count_store.save_sessions_count(end_user_id, 1, langchain_messages)
-
-
-"""Time-based memory processing"""
+        count_store.update_sessions_count(end_user_id, 0, [])
 
 
 async def memory_long_term_storage(end_user_id, memory_config, time):
@@ -291,9 +262,7 @@ async def aggregate_judgment(end_user_id: str, ori_messages: list, memory_config
         return result_dict
 
     except Exception as e:
-        print(f"[aggregate_judgment] 发生错误: {e}")
-        import traceback
-        traceback.print_exc()
+        logger.error(f"[aggregate_judgment] 发生错误: {e}", exc_info=True)
 
         return {
             "is_same_event": False,

api/app/core/memory/agent/langgraph_graph/write_graph.py

@@ -1,49 +1,25 @@
-import asyncio
-import json
-import sys
 import warnings
-from contextlib import asynccontextmanager
-from langgraph.constants import END, START
-from langgraph.graph import StateGraph
 
-from app.db import get_db, get_db_context
 from app.core.logging_config import get_agent_logger
-from app.core.memory.agent.utils.llm_tools import WriteState
-from app.core.memory.agent.langgraph_graph.nodes.write_nodes import write_node
+from app.core.memory.agent.langgraph_graph.routing.write_router import memory_long_term_storage, window_dialogue, \
+    aggregate_judgment
+from app.core.memory.agent.utils.redis_tool import write_store
+from app.db import get_db_context
 from app.schemas.memory_agent_schema import AgentMemory_Long_Term
 from app.services.memory_config_service import MemoryConfigService
+from app.services.memory_konwledges_server import write_rag
 
 warnings.filterwarnings("ignore", category=RuntimeWarning)
 logger = get_agent_logger(__name__)
 
-if sys.platform.startswith("win"):
-    asyncio.set_event_loop_policy(asyncio.WindowsSelectorEventLoopPolicy())
 
-
-@asynccontextmanager
-async def make_write_graph():
-    """
-    Create a write graph workflow for memory operations.
-
-    Args:
-        user_id: User identifier
-        tools: MCP tools loaded from session
-        apply_id: Application identifier
-        end_user_id: Group identifier
-        memory_config: MemoryConfig object containing all configuration
-    """
-    workflow = StateGraph(WriteState)
-    workflow.add_node("save_neo4j", write_node)
-    workflow.add_edge(START, "save_neo4j")
-    workflow.add_edge("save_neo4j", END)
-
-    graph = workflow.compile()
-
-    yield graph
-
-
-async def long_term_storage(long_term_type: str = "chunk", langchain_messages: list = [], memory_config: str = '',
-                            end_user_id: str = '', scope: int = 6):
+async def long_term_storage(
+        long_term_type: str,
+        langchain_messages: list,
+        memory_config_id: str,
+        end_user_id: str,
+        scope: int = 6
+):
     """
     Handle long-term memory storage with different strategies
 
@@ -53,33 +29,39 @@ async def long_term_storage(long_term_type: str = "chunk", langchain_messages: l
     Args:
         long_term_type: Storage strategy type ('chunk', 'time', 'aggregate')
         langchain_messages: List of messages to store
-        memory_config: Memory configuration identifier
+        memory_config_id: Memory configuration identifier
         end_user_id: User group identifier
         scope: Scope parameter for chunk-based storage (default: 6)
     """
-    from app.core.memory.agent.langgraph_graph.routing.write_router import memory_long_term_storage, window_dialogue, \
-        aggregate_judgment
-    from app.core.memory.agent.utils.redis_tool import write_store
+    if langchain_messages is None:
+        langchain_messages = []
+
     write_store.save_session_write(end_user_id, langchain_messages)
     # 获取数据库会话
     with get_db_context() as db_session:
         config_service = MemoryConfigService(db_session)
         memory_config = config_service.load_memory_config(
-            config_id=memory_config,  # 改为整数
+            config_id=memory_config_id,  # 改为整数
             service_name="MemoryAgentService"
         )
         if long_term_type == AgentMemory_Long_Term.STRATEGY_CHUNK:
-            '''Strategy 1: Dialogue window with 6 rounds of conversation'''
+            # Dialogue window with 6 rounds of conversation
             await window_dialogue(end_user_id, langchain_messages, memory_config, scope)
         if long_term_type == AgentMemory_Long_Term.STRATEGY_TIME:
-            """Time-based strategy"""
+            # Time-based strategy
             await memory_long_term_storage(end_user_id, memory_config, AgentMemory_Long_Term.TIME_SCOPE)
         if long_term_type == AgentMemory_Long_Term.STRATEGY_AGGREGATE:
-            """Strategy 3: Aggregate judgment"""
+            # Aggregate judgment
             await aggregate_judgment(end_user_id, langchain_messages, memory_config)
 
 
-async def write_long_term(storage_type, end_user_id, message_chat, aimessages, user_rag_memory_id, actual_config_id):
+async def write_long_term(
+        storage_type: str,
+        end_user_id: str,
+        messages: list[dict],
+        user_rag_memory_id: str,
+        actual_config_id: str
+):
     """
     Write long-term memory with different storage types
 
@@ -89,44 +71,24 @@ async def write_long_term(storage_type, end_user_id, message_chat, aimessages, u
     Args:
         storage_type: Type of storage (RAG or traditional)
         end_user_id: User group identifier
-        message_chat: User message content
-        aimessages: AI response messages
+        messages: message list
         user_rag_memory_id: RAG memory identifier
         actual_config_id: Actual configuration ID
     """
-    from app.core.memory.agent.langgraph_graph.routing.write_router import write_rag_agent
     from app.core.memory.agent.langgraph_graph.routing.write_router import term_memory_save
-    from app.core.memory.agent.langgraph_graph.tools.write_tool import agent_chat_messages
     if storage_type == AgentMemory_Long_Term.STORAGE_RAG:
-        await write_rag_agent(end_user_id, message_chat, aimessages, user_rag_memory_id)
+        message_content = []
+        for message in messages:
+            message_content.append(f'{message.get("role")}:{message.get("content")}')
+        messages_string = "\n".join(message_content)
+        await write_rag(end_user_id, messages_string, user_rag_memory_id)
     else:
         # AI reply writing (user messages and AI replies paired, written as complete dialogue at once)
         CHUNK = AgentMemory_Long_Term.STRATEGY_CHUNK
         SCOPE = AgentMemory_Long_Term.DEFAULT_SCOPE
-        long_term_messages = await agent_chat_messages(message_chat, aimessages)
-        await long_term_storage(long_term_type=CHUNK, langchain_messages=long_term_messages,
-                                memory_config=actual_config_id, end_user_id=end_user_id, scope=SCOPE)
-        await term_memory_save(long_term_messages, actual_config_id, end_user_id, CHUNK, scope=SCOPE)
-
-# async def main():
-#     """主函数 - 运行工作流"""
-#     langchain_messages = [
-#     {
-#       "role": "user",
-#       "content": "今天周五去爬山"
-#     },
-#     {
-#       "role": "assistant",
-#       "content": "好耶"
-#     }
-#
-#   ]
-#     end_user_id = '837fee1b-04a2-48ee-94d7-211488908940'  # 组ID
-#     memory_config="08ed205c-0f05-49c3-8e0c-a580d28f5fd4"
-#     await long_term_storage(long_term_type="chunk",langchain_messages=langchain_messages,memory_config=memory_config,end_user_id=end_user_id,scope=2)
-#
-#
-#
-# if __name__ == "__main__":
-#     import asyncio
-#     asyncio.run(main())
+        await long_term_storage(long_term_type=CHUNK,
+                                langchain_messages=messages,
+                                memory_config_id=actual_config_id,
+                                end_user_id=end_user_id,
+                                scope=SCOPE)
+        await term_memory_save(end_user_id, CHUNK, scope=SCOPE)

api/app/core/memory/agent/utils/get_dialogs.py

@@ -11,7 +11,7 @@ async def get_chunked_dialogs(
         chunker_strategy: str = "RecursiveChunker",
         end_user_id: str = "group_1",
         messages: list = None,
-        ref_id: str = "wyl_20251027",
+        ref_id: str = "",
         config_id: str = None
 ) -> List[DialogData]:
     """Generate chunks from structured messages using the specified chunker strategy.
@@ -40,12 +40,13 @@ async def get_chunked_dialogs(
 
         role = msg['role']
         content = msg['content']
+        files = msg.get("file_content", [])
 
         if role not in ['user', 'assistant']:
             raise ValueError(f"Message {idx} role must be 'user' or 'assistant', got: {role}")
 
         if content.strip():
-            conversation_messages.append(ConversationMessage(role=role, msg=content.strip()))
+            conversation_messages.append(ConversationMessage(role=role, msg=content.strip(), files=files))
 
     if not conversation_messages:
         raise ValueError("Message list cannot be empty after filtering")

api/app/core/memory/agent/utils/prompt/Problem_Extension_prompt.jinja2

@@ -39,6 +39,30 @@
     比如：输入历史信息内容:[{'Query': '4月27日，我和你推荐过一本书，书名是什么？', 'ANswer': '张曼玉推荐了《小王子》'}]
     拆分问题：4月27日，我和你推荐过一本书，书名是什么？，可以拆分为：4月27日，张曼玉推荐过一本书，书名是什么？
 
+## 指代消歧规则（Coreference Resolution）：
+在拆分问题时，必须解析并替换所有指代词和抽象称呼，使问题具体化：
+
+1. **"用户"的消歧**：
+   - "用户是谁？" → 分析历史记录，找出对话发起者的姓名
+   - 如果历史中有"我叫X"、"我的名字是X"、或多次提到某个人物，则"用户"指的就是这个人
+   - 示例：历史中有"老李的原名叫李建国"，则"用户是谁？"应拆分为"李建国是谁？"或"老李（李建国）是谁？"
+
+2. **"我"的消歧**：
+   - "我喜欢什么？" → 从历史中找出对话发起者的姓名，替换为"X喜欢什么？"
+   - 示例：历史中有"张曼玉推荐了《小王子》"，则"我推荐的书是什么？"应拆分为"张曼玉推荐的书是什么？"
+
+3. **"他/她/它"的消歧**：
+   - 从上下文或历史中找出最近提到的同类实体
+   - 示例：历史中有"老李的同事叫他建国哥"，则"他的同事怎么称呼他？"应拆分为"老李的同事怎么称呼他？"
+
+4. **"那个人/这个人"的消歧**：
+   - 从历史中找出最近提到的人物
+   - 示例：历史中有"李建国"，则"那个人的原名是什么？"应拆分为"李建国的原名是什么？"
+
+5. **优先级**：
+   - 如果历史记录中反复出现某个人物（如"老李"、"李建国"、"建国哥"），则"用户"很可能指的就是这个人
+   - 如果无法从历史中确定指代对象，保留原问题，但在reason中说明"无法确定指代对象"
+
 
 
 输出要求：
@@ -71,6 +95,34 @@
     "reason": "输出原问题的关键要素"
   }
 ]
+
+## 指代消歧示例（重要）：
+示例1 - "用户"的消歧：
+输入历史：[{'Query': '老李的原名叫什么？', 'Answer': '李建国'}, {'Query': '老李的同事叫他什么？', 'Answer': '建国哥'}]
+输入问题："用户是谁？"
+输出：
+[
+  {
+    "original_question": "用户是谁？",
+    "extended_question": "李建国是谁？",
+    "type": "单跳",
+    "reason": "历史中反复提到'老李/李建国/建国哥'，'用户'指的就是对话发起者李建国"
+  }
+]
+
+示例2 - "我"的消歧：
+输入历史：[{'Query': '张曼玉推荐了什么书？', 'Answer': '《小王子》'}]
+输入问题："我推荐的书是什么？"
+输出：
+[
+  {
+    "original_question": "我推荐的书是什么？",
+    "extended_question": "张曼玉推荐的书是什么？",
+    "type": "单跳",
+    "reason": "历史中提到张曼玉推荐了书，'我'指的就是张曼玉"
+  }
+]
+
 **Output format**
 **CRITICAL JSON FORMATTING REQUIREMENTS:**
 1. Use only standard ASCII double quotes (") for JSON structure - never use Chinese quotation marks ("") or other Unicode quotes

api/app/core/memory/agent/utils/prompt/problem_breakdown_prompt.jinja2

@@ -27,6 +27,30 @@
     比如：输入历史信息内容:[{'Query': '4月27日，我和你推荐过一本书，书名是什么？', 'ANswer': '张曼玉推荐了《小王子》'}]
     拆分问题：4月27日，我和你推荐过一本书，书名是什么？，可以拆分为：4月27日，张曼玉推荐过一本书，书名是什么？
 
+## 指代消歧规则（Coreference Resolution）：
+在拆分问题时，必须解析并替换所有指代词和抽象称呼，使问题具体化：
+
+1. **"用户"的消歧**：
+   - "用户是谁？" → 分析历史记录，找出对话发起者的姓名
+   - 如果历史中有"我叫X"、"我的名字是X"、或多次提到某个人物（如"老李"、"李建国"），则"用户"指的就是这个人
+   - 示例：历史中反复出现"老李/李建国/建国哥"，则"用户是谁？"应拆分为"李建国是谁？"或"老李（李建国）是谁？"
+
+2. **"我"的消歧**：
+   - "我喜欢什么？" → 从历史中找出对话发起者的姓名，替换为"X喜欢什么？"
+   - 示例：历史中有"张曼玉推荐了《小王子》"，则"我推荐的书是什么？"应拆分为"张曼玉推荐的书是什么？"
+
+3. **"他/她/它"的消歧**：
+   - 从上下文或历史中找出最近提到的同类实体
+   - 示例：历史中有"老李的同事叫他建国哥"，则"他的同事怎么称呼他？"应拆分为"老李的同事怎么称呼他？"
+
+4. **"那个人/这个人"的消歧**：
+   - 从历史中找出最近提到的人物
+   - 示例：历史中有"李建国"，则"那个人的原名是什么？"应拆分为"李建国的原名是什么？"
+
+5. **优先级**：
+   - 如果历史记录中反复出现某个人物（如"老李"、"李建国"、"建国哥"），则"用户"很可能指的就是这个人
+   - 如果无法从历史中确定指代对象，保留原问题，但在reason中说明"无法确定指代对象"
+
 ## 指令：
 你是一个智能数据拆分助手，请根据数据特性判断输入属于哪种类型：
 单跳（Single-hop）
@@ -151,6 +175,34 @@
 ]
 - 必须通过json.loads()的格式支持的形式输出
 - 必须通过json.loads()的格式支持的形式输出,响应必须是与此确切模式匹配的有效JSON对象。不要在JSON之前或之后包含任何文本。
+
+## 指代消歧示例（重要）：
+示例1 - "用户"的消歧：
+输入历史：[{'Query': '老李的原名叫什么？', 'Answer': '李建国'}, {'Query': '老李的同事叫他什么？', 'Answer': '建国哥'}]
+输入问题："用户是谁？"
+输出：
+[
+  {
+    "id": "Q1",
+    "question": "李建国是谁？",
+    "type": "单跳",
+    "reason": "历史中反复提到'老李/李建国/建国哥'，'用户'指的就是对话发起者李建国"
+  }
+]
+
+示例2 - "我"的消歧：
+输入历史：[{'Query': '张曼玉推荐了什么书？', 'Answer': '《小王子》'}]
+输入问题："我推荐的书是什么？"
+输出：
+[
+  {
+    "id": "Q1",
+    "question": "张曼玉推荐的书是什么？",
+    "type": "单跳",
+    "reason": "历史中提到张曼玉推荐了书，'我'指的就是张曼玉"
+  }
+]
+
 - 关键的JSON格式要求
 1.JSON结构仅使用标准ASCII双引号（“）-切勿使用中文引号（“”）或其他Unicode引号
 2.如果提取的语句文本包含引号，请使用反斜杠（\“）正确转义它们

api/app/core/memory/agent/utils/redis_tool.py

@@ -3,8 +3,9 @@ import uuid
 from app.core.config import settings
 from typing import List, Dict, Any, Optional, Union
 
+from app.core.logging_config import get_logger
 from app.core.memory.agent.utils.redis_base import (
-    serialize_messages, 
+    serialize_messages,
     deserialize_messages,
     fix_encoding,
     format_session_data,
@@ -14,12 +15,12 @@ from app.core.memory.agent.utils.redis_base import (
     get_current_timestamp
 )
 
-
+logger = get_logger(__name__)
 
 
 class RedisWriteStore:
     """Redis Write 类型存储类，用于管理 save_session_write 相关的数据"""
-    
+
     def __init__(self, host='localhost', port=6379, db=0, password=None, session_id=''):
         """
         初始化 Redis 连接
@@ -66,10 +67,10 @@ class RedisWriteStore:
             })
             result = pipe.execute()
 
-            print(f"[save_session_write] 保存结果: {result[0]}, session_id: {session_id}")
+            logger.debug(f"[save_session_write] 保存结果: {result[0]}, session_id: {session_id}")
             return session_id
         except Exception as e:
-            print(f"[save_session_write] 保存会话失败: {e}")
+            logger.error(f"[save_session_write] 保存会话失败: {e}")
             raise e
 
     def get_session_by_userid(self, userid: str) -> Union[List[Dict[str, str]], bool]:
@@ -99,7 +100,7 @@ class RedisWriteStore:
             for key, data in zip(keys, all_data):
                 if not data:
                     continue
-                
+
                 # 从 write 类型读取，匹配 sessionid 字段
                 if data.get('sessionid') == userid:
                     # 从 key 中提取 session_id: session:write:{session_id}
@@ -108,16 +109,16 @@ class RedisWriteStore:
                         "sessionid": session_id,
                         "messages": fix_encoding(data.get('messages', ''))
                     })
-            
+
             if not results:
                 return False
-            
-            print(f"[get_session_by_userid] userid={userid}, 找到 {len(results)} 条数据")
+
+            logger.debug(f"[get_session_by_userid] userid={userid}, 找到 {len(results)} 条数据")
             return results
         except Exception as e:
-            print(f"[get_session_by_userid] 查询失败: {e}")
+            logger.error(f"[get_session_by_userid] 查询失败: {e}")
             return False
-    
+
     def get_all_sessions_by_end_user_id(self, end_user_id: str) -> Union[List[Dict[str, Any]], bool]:
         """
         通过 end_user_id 获取所有 write 类型的会话数据
@@ -144,7 +145,7 @@ class RedisWriteStore:
             # 只查询 write 类型的 key
             keys = self.r.keys('session:write:*')
             if not keys:
-                print(f"[get_all_sessions_by_end_user_id] 没有找到任何 write 类型的会话")
+                logger.debug(f"[get_all_sessions_by_end_user_id] 没有找到任何 write 类型的会话")
                 return False
 
             # 批量获取数据
@@ -158,12 +159,12 @@ class RedisWriteStore:
             for key, data in zip(keys, all_data):
                 if not data:
                     continue
-                
+
                 # 从 write 类型读取，匹配 sessionid 字段
                 if data.get('sessionid') == end_user_id:
                     # 从 key 中提取 session_id: session:write:{session_id}
                     session_id = key.split(':')[-1]
-                    
+
                     # 构建完整的会话信息
                     session_info = {
                         "session_id": session_id,
@@ -173,23 +174,21 @@ class RedisWriteStore:
                         "starttime": data.get('starttime', '')
                     }
                     results.append(session_info)
-            
+
             if not results:
-                print(f"[get_all_sessions_by_end_user_id] end_user_id={end_user_id}, 没有找到数据")
+                logger.debug(f"[get_all_sessions_by_end_user_id] end_user_id={end_user_id}, 没有找到数据")
                 return False
-            
+
             # 按时间排序（最新的在前）
             results.sort(key=lambda x: x.get('starttime', ''), reverse=True)
-            
-            print(f"[get_all_sessions_by_end_user_id] end_user_id={end_user_id}, 找到 {len(results)} 条数据")
+
+            logger.debug(f"[get_all_sessions_by_end_user_id] end_user_id={end_user_id}, 找到 {len(results)} 条数据")
             return results
         except Exception as e:
-            print(f"[get_all_sessions_by_end_user_id] 查询失败: {e}")
-            import traceback
-            traceback.print_exc()
+            logger.error(f"[get_all_sessions_by_end_user_id] 查询失败: {e}", exc_info=True)
             return False
 
-    def find_user_recent_sessions(self, userid: str, 
+    def find_user_recent_sessions(self, userid: str,
                                   minutes: int = 5) -> List[Dict[str, str]]:
         """
         根据 userid 从 save_session_write 写入的数据中查询最近 N 分钟内的会话数据
@@ -203,11 +202,11 @@ class RedisWriteStore:
         """
         import time
         start_time = time.time()
-        
+
         # 只查询 write 类型的 key
         keys = self.r.keys('session:write:*')
         if not keys:
-            print(f"[find_user_recent_sessions] 查询耗时: {time.time() - start_time:.3f}秒, 结果数: 0")
+            logger.debug(f"[find_user_recent_sessions] 查询耗时: {time.time() - start_time:.3f}秒, 结果数: 0")
             return []
 
         # 批量获取数据
@@ -221,7 +220,7 @@ class RedisWriteStore:
         for data in all_data:
             if not data:
                 continue
-            
+
             # 从 write 类型读取，匹配 sessionid 字段
             if data.get('sessionid') == userid and data.get('starttime'):
                 # write 类型没有 aimessages，所以 Answer 为空
@@ -230,15 +229,14 @@ class RedisWriteStore:
                     "Answer": "",
                     "starttime": data.get('starttime', '')
                 })
-        
+
         # 根据时间范围过滤
         filtered_items = filter_by_time_range(matched_items, minutes)
         # 排序并移除时间字段
-        result_items = sort_and_limit_results(filtered_items, limit=None)
-        print(result_items)
+        result_items = sort_and_limit_results(filtered_items)
 
         elapsed_time = time.time() - start_time
-        print(f"[find_user_recent_sessions] userid={userid}, minutes={minutes}, "
+        logger.debug(f"[find_user_recent_sessions] userid={userid}, minutes={minutes}, "
               f"查询耗时: {elapsed_time:.3f}秒, 结果数: {len(result_items)}")
 
         return result_items
@@ -258,7 +256,7 @@ class RedisWriteStore:
 
 class RedisCountStore:
     """Redis Count 类型存储类，用于管理访问次数统计相关的数据"""
-    
+
     def __init__(self, host='localhost', port=6379, db=0, password=None, session_id=''):
         """
         初始化 Redis 连接
@@ -278,7 +276,7 @@ class RedisCountStore:
             decode_responses=True,
             encoding='utf-8'
         )
-        self.uudi = session_id
+        self.uuid = session_id
 
     def save_sessions_count(self, end_user_id: str, count: int, messages: Any) -> str:
         """
@@ -295,26 +293,26 @@ class RedisCountStore:
         session_id = str(uuid.uuid4())
         key = generate_session_key(session_id, key_type="count")
         index_key = f'session:count:index:{end_user_id}'  # 索引键
-        
+
         pipe = self.r.pipeline()
         pipe.hset(key, mapping={
-            "id": self.uudi,
+            "id": self.uuid,
             "end_user_id": end_user_id,
             "count": int(count),
             "messages": serialize_messages(messages),
             "starttime": get_current_timestamp()
         })
         pipe.expire(key, 30 * 24 * 60 * 60)  # 30天过期
-        
+
         # 创建索引：end_user_id -> session_id 映射
         pipe.set(index_key, session_id, ex=30 * 24 * 60 * 60)
-        
+
         result = pipe.execute()
-        
-        print(f"[save_sessions_count] 保存结果: {result}, session_id: {session_id}")
+
+        logger.debug(f"[save_sessions_count] 保存结果: {result}, session_id: {session_id}")
         return session_id
 
-    def get_sessions_count(self, end_user_id: str) -> Union[List[Any], bool]:
+    def get_sessions_count(self, end_user_id: str) -> tuple[int, list[dict]] | bool:
         """
         通过 end_user_id 查询访问次数统计
         
@@ -327,7 +325,7 @@ class RedisCountStore:
         try:
             # 使用索引键快速查找
             index_key = f'session:count:index:{end_user_id}'
-            
+
             # 检查索引键类型，避免 WRONGTYPE 错误
             try:
                 key_type = self.r.type(index_key)
@@ -335,35 +333,40 @@ class RedisCountStore:
                     self.r.delete(index_key)
                     return False
             except Exception as type_error:
-                print(f"[get_sessions_count] 检查键类型失败: {type_error}")
-            
+                logger.error(f"[get_sessions_count] 检查键类型失败: {type_error}")
+
             session_id = self.r.get(index_key)
-            
+
             if not session_id:
                 return False
-            
+
             # 直接获取数据
             key = generate_session_key(session_id, key_type="count")
             data = self.r.hgetall(key)
-            
+
             if not data:
                 # 索引存在但数据不存在，清理索引
                 self.r.delete(index_key)
                 return False
-            
+
             count = data.get('count')
             messages_str = data.get('messages')
-            
+
             if count is not None:
-                messages = deserialize_messages(messages_str)
-                return [int(count), messages]
-            
+                messages: list[dict] = deserialize_messages(messages_str)
+                return int(count), messages
+
             return False
         except Exception as e:
-            print(f"[get_sessions_count] 查询失败: {e}")
+            logger.error(f"[get_sessions_count] 查询失败: {e}")
             return False
-    def update_sessions_count(self, end_user_id: str, new_count: int, 
-                             messages: Any) -> bool:
+
+    def update_sessions_count(
+            self,
+            end_user_id: str,
+            new_count: int,
+            messages: Any
+    ) -> bool:
         """
         通过 end_user_id 修改访问次数统计（优化版：使用索引）
         
@@ -378,39 +381,39 @@ class RedisCountStore:
         try:
             # 使用索引键快速查找
             index_key = f'session:count:index:{end_user_id}'
-            
+
             # 检查索引键类型，避免 WRONGTYPE 错误
             try:
                 key_type = self.r.type(index_key)
                 if key_type != 'string' and key_type != 'none':
                     # 索引键类型错误，删除并返回 False
-                    print(f"[update_sessions_count] 索引键类型错误: {key_type}，删除索引")
+                    logger.warning(f"[update_sessions_count] 索引键类型错误: {key_type}，删除索引")
                     self.r.delete(index_key)
-                    print(f"[update_sessions_count] 未找到记录: end_user_id={end_user_id}")
+                    logger.debug(f"[update_sessions_count] 未找到记录: end_user_id={end_user_id}")
                     return False
             except Exception as type_error:
-                print(f"[update_sessions_count] 检查键类型失败: {type_error}")
-            
+                logger.error(f"[update_sessions_count] 检查键类型失败: {type_error}")
+
             session_id = self.r.get(index_key)
-            
+
             if not session_id:
-                print(f"[update_sessions_count] 未找到记录: end_user_id={end_user_id}")
+                logger.debug(f"[update_sessions_count] 未找到记录: end_user_id={end_user_id}")
                 return False
-            
+
             # 直接更新数据
             key = generate_session_key(session_id, key_type="count")
             messages_str = serialize_messages(messages)
-            
+
             pipe = self.r.pipeline()
-            pipe.hset(key, 'count', int(new_count))
+            pipe.hset(key, 'count', str(new_count))
             pipe.hset(key, 'messages', messages_str)
             result = pipe.execute()
-            
-            print(f"[update_sessions_count] 更新成功: end_user_id={end_user_id}, new_count={new_count}, key={key}")
+
+            logger.debug(f"[update_sessions_count] 更新成功: end_user_id={end_user_id}, new_count={new_count}, key={key}")
             return True
-            
+
         except Exception as e:
-            print(f"[update_sessions_count] 更新失败: {e}")
+            logger.debug(f"[update_sessions_count] 更新失败: {e}")
             return False
 
     def delete_all_count_sessions(self) -> int:
@@ -428,7 +431,7 @@ class RedisCountStore:
 
 class RedisSessionStore:
     """Redis 会话存储类，用于管理会话数据"""
-    
+
     def __init__(self, host='localhost', port=6379, db=0, password=None, session_id=''):
         """
         初始化 Redis 连接
@@ -451,9 +454,9 @@ class RedisSessionStore:
         self.uudi = session_id
 
     # ==================== 写入操作 ====================
-    
-    def save_session(self, userid: str, messages: str, aimessages: str, 
-                    apply_id: str, end_user_id: str) -> str:
+
+    def save_session(self, userid: str, messages: str, aimessages: str,
+                     apply_id: str, end_user_id: str) -> str:
         """
         写入一条会话数据，返回 session_id
         
@@ -483,14 +486,14 @@ class RedisSessionStore:
             })
             result = pipe.execute()
 
-            print(f"[save_session] 保存结果: {result[0]}, session_id: {session_id}")
+            logger.debug(f"[save_session] 保存结果: {result[0]}, session_id: {session_id}")
             return session_id
         except Exception as e:
-            print(f"[save_session] 保存会话失败: {e}")
+            logger.error(f"[save_session] 保存会话失败: {e}")
             raise e
 
     # ==================== 读取操作 ====================
-    
+
     def get_session(self, session_id: str) -> Optional[Dict[str, Any]]:
         """
         读取一条会话数据
@@ -520,8 +523,8 @@ class RedisSessionStore:
                 sessions[sid] = self.get_session(sid)
         return sessions
 
-    def find_user_apply_group(self, sessionid: str, apply_id: str, 
-                             end_user_id: str) -> List[Dict[str, str]]:
+    def find_user_apply_group(self, sessionid: str, apply_id: str,
+                              end_user_id: str) -> List[Dict[str, str]]:
         """
         根据 sessionid、apply_id 和 end_user_id 查询会话数据，返回最新的6条
         
@@ -535,10 +538,10 @@ class RedisSessionStore:
         """
         import time
         start_time = time.time()
-        
+
         keys = self.r.keys('session:*')
         if not keys:
-            print(f"[find_user_apply_group] 查询耗时: {time.time() - start_time:.3f}秒, 结果数: 0")
+            logger.debug(f"[find_user_apply_group] 查询耗时: {time.time() - start_time:.3f}秒, 结果数: 0")
             return []
 
         # 批量获取数据
@@ -556,21 +559,21 @@ class RedisSessionStore:
                 continue
 
             if (data.get('apply_id') == apply_id and
-                data.get('end_user_id') == end_user_id):
+                    data.get('end_user_id') == end_user_id):
                 # 支持模糊匹配或完全匹配 sessionid
                 if sessionid in data.get('sessionid', '') or data.get('sessionid') == sessionid:
                     matched_items.append(format_session_data(data, include_time=True))
-        
+
         # 排序、限制数量并移除时间字段
         result_items = sort_and_limit_results(matched_items, limit=6)
 
         elapsed_time = time.time() - start_time
-        print(f"[find_user_apply_group] 查询耗时: {elapsed_time:.3f}秒, 结果数: {len(result_items)}")
+        logger.debug(f"[find_user_apply_group] 查询耗时: {elapsed_time:.3f}秒, 结果数: {len(result_items)}")
 
         return result_items
 
     # ==================== 更新操作 ====================
-    
+
     def update_session(self, session_id: str, field: str, value: Any) -> bool:
         """
         更新单个字段
@@ -591,7 +594,7 @@ class RedisSessionStore:
         return bool(results[0])
 
     # ==================== 删除操作 ====================
-    
+
     def delete_session(self, session_id: str) -> int:
         """
         删除单条会话
@@ -632,7 +635,7 @@ class RedisSessionStore:
 
         keys = self.r.keys('session:*')
         if not keys:
-            print("[delete_duplicate_sessions] 没有会话数据")
+            logger.debug("[delete_duplicate_sessions] 没有会话数据")
             return 0
 
         # 批量获取所有数据
@@ -678,7 +681,7 @@ class RedisSessionStore:
                 deleted_count += len(batch)
 
         elapsed_time = time.time() - start_time
-        print(f"[delete_duplicate_sessions] 删除重复会话数量: {deleted_count}, 耗时: {elapsed_time:.3f}秒")
+        logger.debug(f"[delete_duplicate_sessions] 删除重复会话数量: {deleted_count}, 耗时: {elapsed_time:.3f}秒")
         return deleted_count
 
 

api/app/core/memory/agent/utils/write_tools.py

@@ -6,35 +6,37 @@ pipeline. Only MemoryConfig is needed - clients are constructed internally.
 """
 import asyncio
 import time
+import uuid
 from datetime import datetime
+from typing import List, Optional
 
 from dotenv import load_dotenv
 
 from app.core.logging_config import get_agent_logger
 from app.core.memory.agent.utils.get_dialogs import get_chunked_dialogs
 from app.core.memory.storage_services.extraction_engine.extraction_orchestrator import ExtractionOrchestrator
-from app.core.memory.storage_services.extraction_engine.knowledge_extraction.memory_summary import memory_summary_generation
+from app.core.memory.storage_services.extraction_engine.knowledge_extraction.memory_summary import \
+    memory_summary_generation
 from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
 from app.core.memory.utils.log.logging_utils import log_time
 from app.db import get_db_context
 from app.repositories.neo4j.add_edges import add_memory_summary_statement_edges
 from app.repositories.neo4j.add_nodes import add_memory_summary_nodes
-from app.repositories.neo4j.graph_saver import save_dialog_and_statements_to_neo4j, _trigger_clustering_sync
+from app.repositories.neo4j.graph_saver import save_dialog_and_statements_to_neo4j
 from app.repositories.neo4j.neo4j_connector import Neo4jConnector
 from app.schemas.memory_config_schema import MemoryConfig
 
-
 load_dotenv()
 
 logger = get_agent_logger(__name__)
 
 
 async def write(
-    end_user_id: str,
-    memory_config: MemoryConfig,
-    messages: list,
-    ref_id: str = "wyl20251027",
-    language: str = "zh",
+        end_user_id: str,
+        memory_config: MemoryConfig,
+        messages: list,
+        ref_id: str = "",
+        language: str = "zh",
 ) -> None:
     """
     Execute the complete knowledge extraction pipeline.
@@ -43,9 +45,11 @@ async def write(
         end_user_id: Group identifier
         memory_config: MemoryConfig object containing all configuration
         messages: Structured message list [{"role": "user", "content": "..."}, ...]
-        ref_id: Reference ID, defaults to "wyl20251027"
+        ref_id: Reference ID, defaults to ""
         language: 语言类型 ("zh" 中文, "en" 英文)，默认中文
     """
+    if not ref_id:
+        ref_id = uuid.uuid4().hex
     # Extract config values
     embedding_model_id = str(memory_config.embedding_model_id)
     chunker_strategy = memory_config.chunker_strategy
@@ -99,14 +103,14 @@ async def write(
     if memory_config.scene_id:
         try:
             from app.core.memory.ontology_services.ontology_type_loader import load_ontology_types_for_scene
-            
+
             with get_db_context() as db:
                 ontology_types = load_ontology_types_for_scene(
                     scene_id=memory_config.scene_id,
                     workspace_id=memory_config.workspace_id,
                     db=db
                 )
-                
+
                 if ontology_types:
                     logger.info(
                         f"Loaded {len(ontology_types.types)} ontology types for scene_id: {memory_config.scene_id}"
@@ -135,9 +139,11 @@ async def write(
         all_chunk_nodes,
         all_statement_nodes,
         all_entity_nodes,
+        all_perceptual_nodes,
         all_statement_chunk_edges,
         all_statement_entity_edges,
         all_entity_entity_edges,
+        all_perceptual_edges,
         all_dedup_details,
     ) = await orchestrator.run(chunked_dialogs, is_pilot_run=False)
 
@@ -145,11 +151,6 @@ async def write(
 
     # Step 3: Save all data to Neo4j database
     step_start = time.time()
-    from app.repositories.neo4j.create_indexes import create_fulltext_indexes
-    try:
-        await create_fulltext_indexes()
-    except Exception as e:
-        logger.error(f"Error creating indexes: {e}", exc_info=True)
 
     # 添加死锁重试机制
     max_retries = 3
@@ -162,19 +163,43 @@ async def write(
                 chunk_nodes=all_chunk_nodes,
                 statement_nodes=all_statement_nodes,
                 entity_nodes=all_entity_nodes,
+                perceptual_nodes=all_perceptual_nodes,
                 statement_chunk_edges=all_statement_chunk_edges,
                 statement_entity_edges=all_statement_entity_edges,
                 entity_edges=all_entity_entity_edges,
+                perceptual_edges=all_perceptual_edges,
                 connector=neo4j_connector,
             )
             if success:
                 logger.info("Successfully saved all data to Neo4j")
-                # 写入成功后，同步等待聚类完成（避免与 Memory Summary 并发冲突）
-                await _trigger_clustering_sync(
-                    all_entity_nodes,
-                    llm_model_id=str(memory_config.llm_model_id) if memory_config.llm_model_id else None,
-                    embedding_model_id=str(memory_config.embedding_model_id) if memory_config.embedding_model_id else None,
-                )
+                
+                # 使用 Celery 异步任务触发聚类（不阻塞主流程）
+                if all_entity_nodes:
+                    try:
+                        from app.tasks import run_incremental_clustering
+                        
+                        end_user_id = all_entity_nodes[0].end_user_id
+                        new_entity_ids = [e.id for e in all_entity_nodes]
+                        
+                        # 异步提交 Celery 任务
+                        task = run_incremental_clustering.apply_async(
+                            kwargs={
+                                "end_user_id": end_user_id,
+                                "new_entity_ids": new_entity_ids,
+                                "llm_model_id": str(memory_config.llm_model_id) if memory_config.llm_model_id else None,
+                                "embedding_model_id": str(memory_config.embedding_model_id) if memory_config.embedding_model_id else None,
+                            },
+                            # 设置任务优先级（低优先级，不影响主业务）
+                            priority=3,
+                        )
+                        logger.info(
+                            f"[Clustering] 增量聚类任务已提交到 Celery - "
+                            f"task_id={task.id}, end_user_id={end_user_id}, entity_count={len(new_entity_ids)}"
+                        )
+                    except Exception as e:
+                        # 聚类任务提交失败不影响主流程
+                        logger.error(f"[Clustering] 提交聚类任务失败（不影响主流程）: {e}", exc_info=True)
+                
                 break
             else:
                 logger.warning("Failed to save some data to Neo4j")
@@ -208,9 +233,8 @@ async def write(
         summaries = await memory_summary_generation(
             chunked_dialogs, llm_client=llm_client, embedder_client=embedder_client, language=language
         )
-
+        ms_connector = Neo4jConnector()
         try:
-            ms_connector = Neo4jConnector()
             await add_memory_summary_nodes(summaries, ms_connector)
             await add_memory_summary_statement_edges(summaries, ms_connector)
         finally:
@@ -250,5 +274,21 @@ async def write(
     except Exception as cache_err:
         logger.warning(f"[WRITE] 写入活动统计缓存失败（不影响主流程）: {cache_err}", exc_info=True)
 
+    # Close LLM/Embedder underlying httpx clients to prevent
+    # 'RuntimeError: Event loop is closed' during garbage collection
+    for client_obj in (llm_client, embedder_client):
+        try:
+            underlying = getattr(client_obj, 'client', None) or getattr(client_obj, 'model', None)
+            if underlying is None:
+                continue
+            # Unwrap RedBearLLM / RedBearEmbeddings to get the LangChain model
+            inner = getattr(underlying, '_model', underlying)
+            # LangChain OpenAI models expose async_client (httpx.AsyncClient)
+            http_client = getattr(inner, 'async_client', None)
+            if http_client is not None and hasattr(http_client, 'aclose'):
+                await http_client.aclose()
+        except Exception:
+            pass
+
     logger.info("=== Pipeline Complete ===")
-    logger.info(f"Total execution time: {total_time:.2f} seconds")
+    logger.info(f"Total execution time: {total_time:.2f} seconds")

api/app/core/memory/llm_tools/chunker_client.py

@@ -1,10 +1,10 @@
-from typing import Any, List
-import re
-import os
 import asyncio
 import json
-import numpy as np
 import logging
+import os
+from typing import Any, List
+
+import numpy as np
 
 # Fix tokenizer parallelism warning
 os.environ["TOKENIZERS_PARALLELISM"] = "false"
@@ -246,6 +246,7 @@ class ChunkerClient:
                             "total_sub_chunks": len(sub_chunks),
                             "chunker_strategy": self.chunker_config.chunker_strategy,
                         },
+                        files=msg.files
                     )
                     dialogue.chunks.append(chunk)
             else:
@@ -258,6 +259,7 @@ class ChunkerClient:
                         "message_role": msg.role,
                         "chunker_strategy": self.chunker_config.chunker_strategy,
                     },
+                    files=msg.files
                 )
                 dialogue.chunks.append(chunk)
 

api/app/core/memory/llm_tools/llm_client.py

@@ -56,7 +56,7 @@ class LLMClient(ABC):
         self.max_retries = self.config.max_retries
         self.timeout = self.config.timeout
 
-        logger.info(
+        logger.debug(
             f"初始化 LLM 客户端: provider={self.provider}, "
             f"model={self.model_name}, max_retries={self.max_retries}"
         )

api/app/core/memory/llm_tools/openai_client.py

@@ -65,7 +65,7 @@ class OpenAIClient(LLMClient):
             type=type_
         )
 
-        logger.info(f"OpenAI 客户端初始化完成: type={type_}")
+        logger.debug(f"OpenAI 客户端初始化完成: type={type_}")
 
     async def chat(self, messages: List[Dict[str, str]], **kwargs) -> Any:
         """

api/app/core/memory/llm_tools/openai_embedder.py

@@ -2,6 +2,7 @@
 OpenAI Embedder 客户端实现
 
 基于 LangChain 和 RedBearEmbeddings 的 OpenAI 嵌入模型客户端实现。
+自动支持火山引擎的多模态 Embedding。
 """
 
 from typing import List
@@ -13,6 +14,7 @@ from app.core.memory.llm_tools.embedder_client import (
 )
 from app.core.models.base import RedBearModelConfig
 from app.core.models.embedding import RedBearEmbeddings
+from app.models.models_model import ModelProvider
 
 logger = logging.getLogger(__name__)
 
@@ -25,6 +27,7 @@ class OpenAIEmbedderClient(EmbedderClient):
     - 批量文本嵌入
     - 自动重试机制
     - 错误处理
+    - 火山引擎多模态 Embedding（自动识别）
     """
 
     def __init__(self, model_config: RedBearModelConfig):
@@ -36,7 +39,7 @@ class OpenAIEmbedderClient(EmbedderClient):
         """
         super().__init__(model_config)
 
-        # 初始化 RedBearEmbeddings 模型
+        # 初始化 RedBearEmbeddings（自动支持火山引擎多模态）
         self.model = RedBearEmbeddings(
             RedBearModelConfig(
                 model_name=self.model_name,
@@ -47,8 +50,9 @@ class OpenAIEmbedderClient(EmbedderClient):
                 timeout=self.timeout,
             )
         )
+        self.is_multimodal = self.model.is_multimodal_supported()
 
-        logger.info("OpenAI Embedder 客户端初始化完成")
+        logger.info(f"OpenAI Embedder 客户端初始化完成 (provider={self.provider}, multimodal={self.is_multimodal})")
 
     async def response(
         self,
@@ -77,7 +81,14 @@ class OpenAIEmbedderClient(EmbedderClient):
                 return []
 
             # 生成嵌入向量
-            embeddings = await self.model.aembed_documents(texts)
+            if self.is_multimodal:
+                # 火山引擎多模态 Embedding
+                embeddings = await self.model.aembed_multimodal(
+                    [{"type": "text", "text": text} for text in texts]
+                )
+            else:
+                # 普通 Embedding
+                embeddings = await self.model.aembed_documents(texts)
 
             logger.debug(f"成功生成 {len(embeddings)} 个嵌入向量")
             return embeddings

api/app/core/memory/models/graph_models.py

@@ -44,21 +44,21 @@ def parse_historical_datetime(v):
     """
     if v is None:
         return v
-    
+
     # 处理 Neo4j DateTime 对象
     if hasattr(v, 'to_native'):
         return v.to_native()
-    
+
     # 处理 Python datetime 对象
     if isinstance(v, datetime):
         return v
-    
+
     if isinstance(v, str):
         # 匹配 ISO 8601 格式：YYYY-MM-DD 或 YYYY-MM-DDTHH:MM:SS[.ffffff][Z|±HH:MM]
         # 支持1-4位年份
         pattern = r'^(\d{1,4})-(\d{2})-(\d{2})(?:T(\d{2}):(\d{2}):(\d{2})(?:\.(\d+))?(?:Z|([+-]\d{2}:\d{2}))?)?'
         match = re.match(pattern, v)
-        
+
         if match:
             try:
                 year = int(match.group(1))
@@ -68,31 +68,31 @@ def parse_historical_datetime(v):
                 minute = int(match.group(5)) if match.group(5) else 0
                 second = int(match.group(6)) if match.group(6) else 0
                 microsecond = 0
-                
+
                 # 处理微秒
                 if match.group(7):
                     # 补齐或截断到6位
                     us_str = match.group(7).ljust(6, '0')[:6]
                     microsecond = int(us_str)
-                
+
                 # 处理时区
                 tzinfo = None
                 if 'Z' in v or match.group(8):
                     tzinfo = timezone.utc
-                
+
                 # 创建 datetime 对象
                 return datetime(year, month, day, hour, minute, second, microsecond, tzinfo=tzinfo)
-                
+
             except (ValueError, OverflowError):
                 # 日期值无效（如月份13、日期32等）
                 return None
-        
+
         # 如果不匹配模式，尝试使用 fromisoformat（用于标准格式）
         try:
             return datetime.fromisoformat(v.replace('Z', '+00:00'))
         except Exception:
             return None
-    
+
     return v
 
 
@@ -114,7 +114,7 @@ class Edge(BaseModel):
     end_user_id: str = Field(..., description="The end user ID of the edge.")
     run_id: str = Field(default_factory=lambda: uuid4().hex, description="Unique identifier for this pipeline run.")
     created_at: datetime = Field(..., description="The valid time of the edge from system perspective.")
-    expired_at: Optional[datetime] = Field(None, description="The expired time of the edge from system perspective.")
+    expired_at: Optional[datetime] = Field(default=None, description="The expired time of the edge from system perspective.")
 
 
 class ChunkEdge(Edge):
@@ -167,7 +167,7 @@ class EntityEntityEdge(Edge):
     source_statement_id: str = Field(..., description="Statement where this relationship was extracted")
     valid_at: Optional[datetime] = Field(None, description="Temporal validity start")
     invalid_at: Optional[datetime] = Field(None, description="Temporal validity end")
-    
+
     @field_validator('valid_at', 'invalid_at', mode='before')
     @classmethod
     def validate_datetime(cls, v):
@@ -175,6 +175,12 @@ class EntityEntityEdge(Edge):
         return parse_historical_datetime(v)
 
 
+class PerceptualEdge(Edge):
+    """Edge connecting perceptual nodes to their source chunks
+    """
+    pass
+
+
 class Node(BaseModel):
     """Base class for all graph nodes in the knowledge graph.
 
@@ -206,7 +212,8 @@ class DialogueNode(Node):
     ref_id: str = Field(..., description="Reference identifier of the dialog")
     content: str = Field(..., description="Dialogue content")
     dialog_embedding: Optional[List[float]] = Field(None, description="Dialog embedding vector")
-    config_id: Optional[int | str] = Field(None, description="Configuration ID used to process this dialogue (integer or string)")
+    config_id: Optional[int | str] = Field(None,
+                                           description="Configuration ID used to process this dialogue (integer or string)")
 
 
 class StatementNode(Node):
@@ -241,17 +248,17 @@ class StatementNode(Node):
     chunk_id: str = Field(..., description="ID of the parent chunk")
     stmt_type: str = Field(..., description="Type of the statement")
     statement: str = Field(..., description="The statement text content")
-    
+
     # Speaker identification
     speaker: Optional[str] = Field(
         None,
         description="Speaker identifier: 'user' for user messages, 'assistant' for AI responses"
     )
-    
+
     # Emotion fields (ordered as requested, emotion_intensity first for display)
     emotion_intensity: Optional[float] = Field(
-        None, 
-        ge=0.0, 
+        None,
+        ge=0.0,
         le=1.0,
         description="Emotion intensity: 0.0-1.0 (displayed on node)"
     )
@@ -264,25 +271,26 @@ class StatementNode(Node):
         description="Emotion subject: self/other/object"
     )
     emotion_type: Optional[str] = Field(
-        None, 
+        None,
         description="Emotion type: joy/sadness/anger/fear/surprise/neutral"
     )
     emotion_keywords: Optional[List[str]] = Field(
         default_factory=list,
         description="Emotion keywords list, max 3 items"
     )
-    
+
     # Temporal fields
     temporal_info: TemporalInfo = Field(..., description="Temporal information")
     valid_at: Optional[datetime] = Field(None, description="Temporal validity start")
     invalid_at: Optional[datetime] = Field(None, description="Temporal validity end")
-    
+
     # Embedding and other fields
     statement_embedding: Optional[List[float]] = Field(None, description="Statement embedding vector")
     chunk_embedding: Optional[List[float]] = Field(None, description="Chunk embedding vector")
     connect_strength: str = Field(..., description="Strong VS Weak classification of this statement")
-    config_id: Optional[int | str] = Field(None, description="Configuration ID used to process this statement (integer or string)")
-    
+    config_id: Optional[int | str] = Field(None,
+                                           description="Configuration ID used to process this statement (integer or string)")
+
     # ACT-R Memory Activation Properties
     importance_score: float = Field(
         default=0.5,
@@ -309,13 +317,13 @@ class StatementNode(Node):
         ge=0,
         description="Total number of times this node has been accessed"
     )
-    
+
     @field_validator('valid_at', 'invalid_at', mode='before')
     @classmethod
     def validate_datetime(cls, v):
         """使用通用的历史日期解析函数"""
         return parse_historical_datetime(v)
-    
+
     @field_validator('emotion_type', mode='before')
     @classmethod
     def validate_emotion_type(cls, v):
@@ -326,7 +334,7 @@ class StatementNode(Node):
         if v not in valid_types:
             raise ValueError(f"emotion_type must be one of {valid_types}, got {v}")
         return v
-    
+
     @field_validator('emotion_subject', mode='before')
     @classmethod
     def validate_emotion_subject(cls, v):
@@ -337,7 +345,7 @@ class StatementNode(Node):
         if v not in valid_subjects:
             raise ValueError(f"emotion_subject must be one of {valid_subjects}, got {v}")
         return v
-    
+
     @field_validator('emotion_keywords', mode='before')
     @classmethod
     def validate_emotion_keywords(cls, v):
@@ -405,19 +413,20 @@ class ExtractedEntityNode(Node):
     entity_type: str = Field(..., description="Type of the entity")
     description: str = Field(..., description="Entity description")
     example: str = Field(
-        default="", 
+        default="",
         description="A concise example (around 20 characters) to help understand the entity"
     )
     aliases: List[str] = Field(
-        default_factory=list, 
+        default_factory=list,
         description="Entity aliases - alternative names for this entity"
     )
     name_embedding: Optional[List[float]] = Field(default_factory=list, description="Name embedding vector")
     # TODO: fact_summary 功能暂时禁用，待后续开发完善后启用
     # fact_summary: str = Field(default="", description="Summary of the fact about this entity")
     connect_strength: str = Field(..., description="Strong VS Weak about this entity")
-    config_id: Optional[int | str] = Field(None, description="Configuration ID used to process this entity (integer or string)")
-    
+    config_id: Optional[int | str] = Field(None,
+                                           description="Configuration ID used to process this entity (integer or string)")
+
     # ACT-R Memory Activation Properties
     importance_score: float = Field(
         default=0.5,
@@ -444,16 +453,16 @@ class ExtractedEntityNode(Node):
         ge=0,
         description="Total number of times this node has been accessed"
     )
-    
+
     # Explicit Memory Classification
     is_explicit_memory: bool = Field(
         default=False,
         description="Whether this entity represents explicit/semantic memory (knowledge, concepts, definitions, theories, principles)"
     )
-    
+
     @field_validator('aliases', mode='before')
     @classmethod
-    def validate_aliases_field(cls, v): # 字段验证器 自动清理和验证 aliases 字段
+    def validate_aliases_field(cls, v):  # 字段验证器 自动清理和验证 aliases 字段
         """Validate and clean aliases field using utility function.
         
         This validator ensures that the aliases field is always a valid list of strings.
@@ -507,8 +516,9 @@ class MemorySummaryNode(Node):
     memory_type: Optional[str] = Field(None, description="Type/category of the episodic memory")
     summary_embedding: Optional[List[float]] = Field(None, description="Embedding vector for the summary")
     metadata: dict = Field(default_factory=dict, description="Additional metadata for the summary")
-    config_id: Optional[int | str] = Field(None, description="Configuration ID used to process this summary (integer or string)")
-    
+    config_id: Optional[int | str] = Field(None,
+                                           description="Configuration ID used to process this summary (integer or string)")
+
     # ACT-R Forgetting Engine Properties
     original_statement_id: Optional[str] = Field(
         None,
@@ -522,7 +532,7 @@ class MemorySummaryNode(Node):
         None,
         description="Timestamp when the nodes were merged"
     )
-    
+
     # ACT-R Memory Activation Properties
     importance_score: float = Field(
         default=0.5,
@@ -549,3 +559,18 @@ class MemorySummaryNode(Node):
         ge=0,
         description="Total number of times this node has been accessed (reset to 1 on creation)"
     )
+
+
+class PerceptualNode(Node):
+    """Node representing a multimodal message in the knowledge graph.
+    """
+    perceptual_type: int
+    file_path: str
+    file_name: str
+    file_ext: str
+    summary: str
+    keywords: list[str]
+    topic: str
+    domain: str
+    file_type: str
+    summary_embedding: list[float] | None

api/app/core/memory/models/message_models.py

@@ -30,6 +30,7 @@ class ConversationMessage(BaseModel):
     """
     role: str = Field(..., description="The role of the speaker (e.g., 'user', 'assistant').")
     msg: str = Field(..., description="The text content of the message.")
+    files: list[tuple] = Field(default_factory=list, description="The file content of the message", exclude=True)
 
 
 class TemporalValidityRange(BaseModel):
@@ -130,7 +131,8 @@ class Chunk(BaseModel):
     content: str = Field(..., description="The content of the chunk as a string.")
     speaker: Optional[str] = Field(None, description="The speaker/role for this chunk (user/assistant).")
     statements: List[Statement] = Field(default_factory=list, description="A list of statements in the chunk.")
-    chunk_embedding: Optional[List[float]] = Field(None, description="The embedding vector of the chunk.")
+    files: list[tuple] = Field(default_factory=list, description="List of files in the chunk.")
+    chunk_embedding: Optional[List[float]] = Field(default=None, description="The embedding vector of the chunk.")
     metadata: Dict[str, Any] = Field(default_factory=dict, description="Additional metadata for the chunk.")
 
     @classmethod

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

@@ -76,6 +76,9 @@ class LabelPropagationEngine:
         self.repo = CommunityRepository(connector)
         self.llm_model_id = llm_model_id
         self.embedding_model_id = embedding_model_id
+        # 缓存客户端实例，避免重复初始化
+        self._llm_client = None
+        self._embedder_client = None
 
     # ──────────────────────────────────────────────────────────────────────────
     # 公开接口
@@ -215,8 +218,17 @@ class LabelPropagationEngine:
         3. 若邻居无社区 → 创建新社区
         4. 若邻居分属多个社区 → 评估是否合并
         """
+        # 收集所有需要生成元数据的社区ID
+        communities_to_update = set()
+        
         for entity_id in new_entity_ids:
-            await self._process_single_entity(entity_id, end_user_id)
+            cid = await self._process_single_entity(entity_id, end_user_id)
+            if cid:
+                communities_to_update.add(cid)
+        
+        # 批量生成所有社区的元数据
+        if communities_to_update:
+            await self._generate_community_metadata(list(communities_to_update), end_user_id, force=True)
 
     # ──────────────────────────────────────────────────────────────────────────
     # 内部方法
@@ -224,8 +236,21 @@ class LabelPropagationEngine:
 
     async def _process_single_entity(
         self, entity_id: str, end_user_id: str
-    ) -> None:
-        """处理单个新实体的社区分配。"""
+    ) -> Optional[str]:
+        """
+        处理单个新实体的社区分配。
+        
+        该函数会为新实体分配社区，可能的情况包括：
+        1. 孤立实体（无邻居）：创建新的单成员社区
+        2. 邻居都没有社区：创建新社区并将实体和邻居都加入
+        3. 邻居有社区：通过加权投票选择最合适的社区加入
+        
+        Returns:
+            Optional[str]: 分配到的社区ID。当前实现总是返回一个有效的社区ID，
+                          但返回类型保留为Optional以支持未来可能的扩展场景
+                          （例如：实体无法分配到任何社区的情况）。
+                          调用方应检查返回值的真假性（truthiness）。
+        """
         neighbors = await self.repo.get_entity_neighbors(entity_id, end_user_id)
 
         # 查询自身 embedding（从邻居查询结果中无法获取，需单独查）
@@ -237,8 +262,7 @@ class LabelPropagationEngine:
             await self.repo.upsert_community(new_cid, end_user_id, member_count=1)
             await self.repo.assign_entity_to_community(entity_id, new_cid, end_user_id)
             logger.debug(f"[Clustering] 孤立实体 {entity_id} → 新社区 {new_cid}")
-            await self._generate_community_metadata([new_cid], end_user_id)
-            return
+            return new_cid
 
         # 统计邻居社区分布
         community_ids_in_neighbors = set(
@@ -260,7 +284,7 @@ class LabelPropagationEngine:
             logger.debug(
                 f"[Clustering] 新实体 {entity_id} 与 {len(neighbors)} 个无社区邻居 → 新社区 {new_cid}"
             )
-            await self._generate_community_metadata([new_cid], end_user_id)
+            return new_cid
         else:
             # 加入得票最多的社区
             await self.repo.assign_entity_to_community(entity_id, target_cid, end_user_id)
@@ -272,8 +296,8 @@ class LabelPropagationEngine:
                 await self._evaluate_merge(
                     list(community_ids_in_neighbors), end_user_id
                 )
-            # 新实体加入后成员变化，强制重新生成元数据
-            await self._generate_community_metadata([target_cid], end_user_id, force=True)
+            # 返回目标社区ID，稍后批量生成元数据
+            return target_cid
 
     async def _evaluate_merge(
         self, community_ids: List[str], end_user_id: str
@@ -456,20 +480,19 @@ class LabelPropagationEngine:
         self, community_ids: List[str], end_user_id: str, force: bool = False
     ) -> None:
         """
-        为一个或多个社区生成并写入元数据。
+        为一个或多个社区生成并写入元数据（优化版：批量 LLM 调用）。
 
         流程：
-        1. 逐个社区调 LLM 生成 name / summary（串行）
-        2. 收集所有 summary，一次性批量 embed
-        3. 单个社区用 update_community_metadata，多个用 batch_update_community_metadata
+        1. 批量准备所有社区的 prompt
+        2. 并发调用 LLM 生成所有社区的 name / summary
+        3. 批量 embed 所有 summary
+        4. 批量写入数据库
 
         Args:
             force: 为 True 时跳过完整性检查，强制重新生成（用于增量更新成员变化后）
         """
-        from app.db import get_db_context
-        from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
-
-        async def _build_one(cid: str) -> Optional[Dict]:
+        async def _prepare_one(cid: str) -> Optional[Dict]:
+            """准备单个社区的数据和 prompt"""
             try:
                 if not force:
                     check_embedding = bool(self.embedding_model_id)
@@ -489,42 +512,32 @@ class LabelPropagationEngine:
                 core_entities = [m["name"] for m in sorted_members[:CORE_ENTITY_LIMIT] if m.get("name")]
                 all_names = [m["name"] for m in members if m.get("name")]
 
+                # 默认值
                 name = "、".join(core_entities[:3]) if core_entities else cid[:8]
                 summary = f"包含实体：{', '.join(all_names)}"
 
+                # 准备 LLM prompt（如果配置了 LLM）
+                prompt = None
                 if self.llm_model_id:
-                    try:
-                        entity_list_str = "\n".join(self._build_entity_lines(members))
-                        relationships = await self.repo.get_community_relationships(cid, end_user_id)
-                        rel_lines = [
-                            f"- {r['subject']} → {r['predicate']} → {r['object']}"
-                            for r in relationships
-                            if r.get("subject") and r.get("predicate") and r.get("object")
-                        ]
-                        rel_section = (
-                            f"\n实体间关系：\n" + "\n".join(rel_lines)
-                            if rel_lines else ""
-                        )
-                        prompt = (
-                            f"以下是一组语义相关的实体：\n{entity_list_str}{rel_section}\n\n"
-                            f"请为这组实体所代表的主题：\n"
-                            f"1. 起一个简洁的中文名称（不超过10个字）\n"
-                            f"2. 写一句话摘要（不超过80个字）\n\n"
-                            f"严格按以下格式输出，不要有其他内容：\n"
-                            f"名称：<名称>\n摘要：<摘要>"
-                        )
-                        with get_db_context() as db:
-                            llm_client = MemoryClientFactory(db).get_llm_client(self.llm_model_id)
-                            response = await llm_client.chat([{"role": "user", "content": prompt}])
-                            text = response.content if hasattr(response, "content") else str(response)
-
-                        for line in text.strip().splitlines():
-                            if line.startswith("名称："):
-                                name = line[3:].strip()
-                            elif line.startswith("摘要："):
-                                summary = line[3:].strip()
-                    except Exception as e:
-                        logger.warning(f"[Clustering] 社区 {cid} LLM 生成失败，使用兜底值: {e}")
+                    entity_list_str = "\n".join(self._build_entity_lines(members))
+                    relationships = await self.repo.get_community_relationships(cid, end_user_id)
+                    rel_lines = [
+                        f"- {r['subject']} → {r['predicate']} → {r['object']}"
+                        for r in relationships
+                        if r.get("subject") and r.get("predicate") and r.get("object")
+                    ]
+                    rel_section = (
+                        f"\n实体间关系：\n" + "\n".join(rel_lines)
+                        if rel_lines else ""
+                    )
+                    prompt = (
+                        f"以下是一组语义相关的实体：\n{entity_list_str}{rel_section}\n\n"
+                        f"请为这组实体所代表的主题：\n"
+                        f"1. 起一个简洁的中文名称（不超过10个字）\n"
+                        f"2. 写一句话摘要（不超过80个字）\n\n"
+                        f"严格按以下格式输出，不要有其他内容：\n"
+                        f"名称：<名称>\n摘要：<摘要>"
+                    )
 
                 return {
                     "community_id": cid,
@@ -532,14 +545,16 @@ class LabelPropagationEngine:
                     "name": name,
                     "summary": summary,
                     "core_entities": core_entities,
+                    "prompt": prompt,
                     "summary_embedding": None,
                 }
             except Exception as e:
                 logger.error(f"[Clustering] 社区 {cid} 元数据准备失败: {e}", exc_info=True)
                 return None
 
+        # --- 阶段1：并发准备所有社区数据 ---
         results = await asyncio.gather(
-            *[_build_one(cid) for cid in community_ids],
+            *[_prepare_one(cid) for cid in community_ids],
             return_exceptions=True,
         )
         metadata_list = []
@@ -553,19 +568,77 @@ class LabelPropagationEngine:
             logger.warning(f"[Clustering] 无有效元数据可写入，community_ids={community_ids}")
             return
 
-        # --- 阶段2：批量生成 summary_embedding ---
-        if self.embedding_model_id:
-            try:
-                summaries = [m["summary"] for m in metadata_list]
-                with get_db_context() as db:
-                    embedder = MemoryClientFactory(db).get_embedder_client(self.embedding_model_id)
-                    embeddings = await embedder.response(summaries)
-                for i, meta in enumerate(metadata_list):
-                    meta["summary_embedding"] = embeddings[i] if i < len(embeddings) else None
-            except Exception as e:
-                logger.error(f"[Clustering] 批量生成 summary_embedding 失败: {e}", exc_info=True)
+        # --- 阶段2：批量调用 LLM 生成 name 和 summary ---
+        if self.llm_model_id:
+            llm_client = self._get_llm_client()
+            if not llm_client:
+                logger.warning(
+                    f"[Clustering] LLM 已配置（model_id={self.llm_model_id}）但客户端初始化失败，"
+                    f"将跳过社区元数据的 LLM 富化。请检查 model_id 是否正确或数据库连接是否正常。"
+                )
+            if llm_client:
+                prompts_to_process = [(i, m) for i, m in enumerate(metadata_list) if m.get("prompt")]
+                
+                if prompts_to_process:
+                    logger.info(f"[Clustering] 批量调用 LLM 生成 {len(prompts_to_process)} 个社区元数据")
+                    
+                    async def _call_llm(idx: int, meta: Dict) -> tuple:
+                        """单个 LLM 调用"""
+                        try:
+                            response = await llm_client.chat([{"role": "user", "content": meta["prompt"]}])
+                            text = response.content if hasattr(response, "content") else str(response)
+                            return (idx, text, None)
+                        except Exception as e:
+                            logger.warning(f"[Clustering] 社区 {meta['community_id']} LLM 生成失败: {e}")
+                            return (idx, None, e)
+                    
+                    # 并发调用所有 LLM 请求
+                    llm_results = await asyncio.gather(
+                        *[_call_llm(idx, meta) for idx, meta in prompts_to_process],
+                        return_exceptions=True
+                    )
+                    
+                    # 解析 LLM 响应
+                    for result in llm_results:
+                        if isinstance(result, Exception):
+                            continue
+                        idx, text, error = result
+                        if error or not text:
+                            continue
+                        
+                        meta = metadata_list[idx]
+                        for line in text.strip().splitlines():
+                            if line.startswith("名称："):
+                                meta["name"] = line[3:].strip()
+                            elif line.startswith("摘要："):
+                                meta["summary"] = line[3:].strip()
+                    
+                    logger.info(f"[Clustering] LLM 批量生成完成")
 
-        # --- 阶段3：写入（单个 or 批量）---
+        # --- 阶段3：批量生成 summary_embedding ---
+        if self.embedding_model_id:
+            embedder = self._get_embedder_client()
+            if not embedder:
+                logger.warning(
+                    f"[Clustering] Embedding 已配置（model_id={self.embedding_model_id}）但客户端初始化失败，"
+                    f"将跳过社区摘要的向量化。请检查 model_id 是否正确或数据库连接是否正常。"
+                )
+            if embedder:
+                try:
+                    summaries = [m["summary"] for m in metadata_list]
+                    logger.info(f"[Clustering] 批量生成 {len(summaries)} 个 summary embedding")
+                    embeddings = await embedder.response(summaries)
+                    for i, meta in enumerate(metadata_list):
+                        meta["summary_embedding"] = embeddings[i] if i < len(embeddings) else None
+                    logger.info(f"[Clustering] Embedding 批量生成完成")
+                except Exception as e:
+                    logger.error(f"[Clustering] 批量生成 summary_embedding 失败: {e}", exc_info=True)
+
+        # --- 阶段4：批量写入数据库 ---
+        # 移除 prompt 字段（不需要存储）
+        for m in metadata_list:
+            m.pop("prompt", None)
+        
         if len(metadata_list) == 1:
             m = metadata_list[0]
             result = await self.repo.update_community_metadata(
@@ -582,6 +655,28 @@ class LabelPropagationEngine:
             ok = await self.repo.batch_update_community_metadata(metadata_list)
             if not ok:
                 logger.error(f"[Clustering] 批量写入 {len(metadata_list)} 个社区元数据失败")
+            else:
+                logger.info(f"[Clustering] 批量写入 {len(metadata_list)} 个社区元数据成功")
+
+    def _get_llm_client(self):
+        """获取或创建 LLM 客户端（单例模式）"""
+        if self._llm_client is None and self.llm_model_id:
+            from app.db import get_db_context
+            from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
+            with get_db_context() as db:
+                self._llm_client = MemoryClientFactory(db).get_llm_client(self.llm_model_id)
+            logger.info(f"[Clustering] LLM 客户端初始化完成（单例）: model_id={self.llm_model_id}")
+        return self._llm_client
+
+    def _get_embedder_client(self):
+        """获取或创建 Embedder 客户端（单例模式）"""
+        if self._embedder_client is None and self.embedding_model_id:
+            from app.db import get_db_context
+            from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
+            with get_db_context() as db:
+                self._embedder_client = MemoryClientFactory(db).get_embedder_client(self.embedding_model_id)
+            logger.info(f"[Clustering] Embedder 客户端初始化完成（单例）: model_id={self.embedding_model_id}")
+        return self._embedder_client
 
     @staticmethod
     def _new_community_id() -> str:

api/app/core/memory/storage_services/extraction_engine/data_preprocessing/data_pruning.py

@@ -30,6 +30,18 @@ from app.core.memory.storage_services.extraction_engine.data_preprocessing.scene
 logger = logging.getLogger(__name__)
 
 
+def message_has_files(message: "ConversationMessage") -> bool:
+    """检查消息是否包含文件。
+    
+    Args:
+        message: 待检查的消息对象
+        
+    Returns:
+        bool: 如果消息包含文件则返回 True，否则返回 False
+    """
+    return message.files and len(message.files) > 0
+
+
 class DialogExtractionResponse(BaseModel):
     """对话级一次性抽取的结构化返回，用于加速剪枝。
 
@@ -128,7 +140,7 @@ class SemanticPruner:
         1. 空消息
         2. 场景特定填充词库精确匹配
         3. 常见寒暄精确匹配
-        4. 组合寒暄模式（前缀+后缀组合，如"好的谢谢"、"同学你好"、"明白了"）
+        4. 组合寒暄模式（前缀 + 后缀组合，如"好的谢谢"、"同学你好"、"明白了"）
         5. 纯表情/标点
         """
         t = message.msg.strip()
@@ -482,6 +494,11 @@ class SemanticPruner:
         """
         to_delete_ids: set = set()
         for m in msgs:
+            # 最高优先级保护：带有文件的消息一律保留，不参与任何剪枝判断
+            if message_has_files(m):
+                self._log(f"  [保护] 带文件的消息（不参与剪枝）：'{m.msg[:40]}'，文件数={len(m.files)}")
+                continue
+                
             # 填充检测优先：先判断是否为填充，再看 LLM 保护
             if self._is_filler_message(m):
                 to_delete_ids.add(id(m))
@@ -549,6 +566,11 @@ class SemanticPruner:
         to_delete_ids: set = set()
         for m in msgs:
             msg_text = m.msg.strip()
+            
+            # 最高优先级保护：带有文件的消息一律保留，不参与任何剪枝判断
+            if message_has_files(m):
+                self._log(f"  [保护] 带文件的消息（不参与剪枝）：'{msg_text[:40]}'，文件数={len(m.files)}")
+                continue
 
             # 第一优先级：填充消息无论模式直接删除，不参与后续场景判断
             if self._is_filler_message(m):
@@ -801,6 +823,12 @@ class SemanticPruner:
 
             for idx, m in enumerate(msgs):
                 msg_text = m.msg.strip()
+                
+                # 最高优先级保护：带有文件的消息一律保留，不参与分类
+                if message_has_files(m):
+                    self._log(f"  [保护] 带文件的消息（不参与分类，直接保留）：索引{idx}, '{msg_text[:40]}', 文件数={len(m.files)}")
+                    llm_protected_msgs.append((idx, m))  # 放入保护列表
+                    continue
 
                 if self._msg_matches_tokens(m, preserve_tokens):
                     llm_protected_msgs.append((idx, m))

api/app/core/memory/storage_services/extraction_engine/deduplication/deduped_and_disamb.py

@@ -203,6 +203,7 @@ def accurate_match(
 ) -> Tuple[List[ExtractedEntityNode], Dict[str, str], Dict[str, Dict]]:
     """
     精确匹配：按 (end_user_id, name, entity_type) 合并实体并建立重定向与合并记录。
+    同时检测某实体的 name 是否命中另一实体的 aliases，若命中则直接合并。
     返回: (deduped_entities, id_redirect, exact_merge_map)
     """
     exact_merge_map: Dict[str, Dict] = {}
@@ -240,6 +241,48 @@ def accurate_match(
             pass
 
     deduped_entities = list(canonical_map.values())
+
+    # 2) 第二轮：检测某实体的 name 是否命中另一实体的 aliases（alias-to-name 精确合并）
+    #    场景：LLM 把 aliases 中的词（如"齐齐"）又单独抽取为独立实体，需在此阶段合并掉
+    #    优化：先构建 (end_user_id, alias_lower) -> canonical 的反向索引，查找 O(1)
+    alias_index: Dict[tuple, ExtractedEntityNode] = {}
+    for canonical in deduped_entities:
+        uid = getattr(canonical, "end_user_id", None)
+        for alias in (getattr(canonical, "aliases", []) or []):
+            alias_lower = alias.strip().lower()
+            if alias_lower:
+                alias_index[(uid, alias_lower)] = canonical
+
+    i = 0
+    while i < len(deduped_entities):
+        ent = deduped_entities[i]
+        ent_name = (getattr(ent, "name", "") or "").strip().lower()
+        ent_uid = getattr(ent, "end_user_id", None)
+        canonical = alias_index.get((ent_uid, ent_name))
+        # 确保不是自身
+        if canonical is not None and canonical.id != ent.id:
+            _merge_attribute(canonical, ent)
+            id_redirect[ent.id] = canonical.id
+            for k, v in list(id_redirect.items()):
+                if v == ent.id:
+                    id_redirect[k] = canonical.id
+            try:
+                k = f"{canonical.end_user_id}|{(canonical.name or '').strip()}|{(canonical.entity_type or '').strip()}"
+                if k not in exact_merge_map:
+                    exact_merge_map[k] = {
+                        "canonical_id": canonical.id,
+                        "end_user_id": canonical.end_user_id,
+                        "name": canonical.name,
+                        "entity_type": canonical.entity_type,
+                        "merged_ids": set(),
+                    }
+                exact_merge_map[k]["merged_ids"].add(ent.id)
+            except Exception:
+                pass
+            deduped_entities.pop(i)
+        else:
+            i += 1
+
     return deduped_entities, id_redirect, exact_merge_map
 
 def fuzzy_match(

api/app/core/memory/storage_services/extraction_engine/deduplication/two_stage_dedup.py

@@ -25,17 +25,17 @@ from app.repositories.neo4j.neo4j_connector import Neo4jConnector
 
 
 async def dedup_layers_and_merge_and_return(
-    dialogue_nodes: List[DialogueNode],
-    chunk_nodes: List[ChunkNode],
-    statement_nodes: List[StatementNode],
-    entity_nodes: List[ExtractedEntityNode],
-    statement_chunk_edges: List[StatementChunkEdge],
-    statement_entity_edges: List[StatementEntityEdge],
-    entity_entity_edges: List[EntityEntityEdge],
-    dialog_data_list: List[DialogData],
-    pipeline_config: ExtractionPipelineConfig,
-    connector: Optional[Neo4jConnector] = None,
-    llm_client = None,
+        dialogue_nodes: List[DialogueNode],
+        chunk_nodes: List[ChunkNode],
+        statement_nodes: List[StatementNode],
+        entity_nodes: List[ExtractedEntityNode],
+        statement_chunk_edges: List[StatementChunkEdge],
+        statement_entity_edges: List[StatementEntityEdge],
+        entity_entity_edges: List[EntityEntityEdge],
+        dialog_data_list: List[DialogData],
+        pipeline_config: ExtractionPipelineConfig,
+        connector: Optional[Neo4jConnector] = None,
+        llm_client=None,
 ) -> Tuple[
     List[DialogueNode],
     List[ChunkNode],
@@ -44,7 +44,7 @@ async def dedup_layers_and_merge_and_return(
     List[StatementChunkEdge],
     List[StatementEntityEdge],
     List[EntityEntityEdge],
-    dict,  # 新增：返回去重详情
+    dict
 ]:
     """
     执行两层实体去重与融合：

api/app/core/memory/storage_services/extraction_engine/knowledge_extraction/memory_summary.py

@@ -188,7 +188,6 @@ async def _process_chunk_summary(
             response_model=MemorySummaryResponse,
         )
         summary_text = structured.summary.strip()
-
         # Generate title and type for the summary
         title = None
         episodic_type = None

api/app/core/memory/utils/prompt/prompts/extract_triplet.jinja2

@@ -5,6 +5,15 @@
 ===Task===
 Extract entities and knowledge triplets from the given statement.
 
+**⚠️ CRITICAL REQUIREMENTS:**
+1. **ALIASES ORDER IS CRITICAL**: The FIRST alias in the array will be used as the user's primary display name (other_name). You MUST put the most important/frequently used name FIRST.
+2. **ALWAYS include aliases field**: Even if empty, you MUST include "aliases": [] in EVERY entity.
+
+<!-- TODO: v0.2.10 - denied_aliases 功能暂时禁用，将通过 Cypher 查询实现
+2. **DENIED_ALIASES**: When user explicitly denies a name (e.g., "我不叫X", "I'm not called X"), you MUST put X in denied_aliases field, NOT in aliases.
+3. **ALWAYS include both fields**: Even if empty, you MUST include "aliases": [] and "denied_aliases": [] in EVERY entity.
+-->
+
 {% if language == "zh" %}
 **重要：请使用中文生成实体名称（name）、描述（description）和示例（example）。**
 {% else %}
@@ -18,34 +27,29 @@ Extract entities and knowledge triplets from the given statement.
 {% if ontology_types %}
 ===Ontology Type Guidance===
 
-**CRITICAL RULE: You MUST ONLY use the predefined ontology type names listed below for the entity "type" field. Do NOT use any other type names, even if they seem reasonable.**
+**CRITICAL: Use ONLY predefined type names below. If no exact match, use CLOSEST type. NEVER invent new types.**
 
-**If no predefined type fits an entity, use the CLOSEST matching predefined type. NEVER invent new type names.**
+**Type Priority:**
+1. [场景类型] Scene Types (domain-specific, prefer first)
+2. [通用类型] General Types (standard ontologies)
+3. [通用父类] Parent Types (hierarchy context)
 
-**Type Priority (from highest to lowest):**
-1. **[场景类型] Scene Types** - Domain-specific types, ALWAYS prefer these first
-2. **[通用类型] General Types** - Common types from standard ontologies (DBpedia)
-3. **[通用父类] Parent Types** - Provide type hierarchy context
+**Rules:**
+- Type MUST exactly match predefined names
+- Do NOT modify, translate, or abbreviate type names
+- Prefer scene types over general types
 
-**Type Matching Rules:**
-- Entity type MUST exactly match one of the predefined type names below
-- Do NOT use types like "Equipment", "Component", "Concept", "Action", "Condition", "Data", "Duration" unless they appear in the predefined list
-- Do NOT modify, translate, abbreviate, or create variations of type names
-- Prefer scene types (marked [场景类型]) over general types when both could apply
-- If uncertain, check the type description to find the best match
-
-**Predefined Ontology Types:**
+**Predefined Types:**
 {{ ontology_types }}
 
 {% if type_hierarchy_hints %}
-**Type Hierarchy Reference:**
-The following shows type inheritance relationships (Child → Parent → Grandparent):
+**Hierarchy:**
 {% for hint in type_hierarchy_hints %}
 - {{ hint }}
 {% endfor %}
 {% endif %}
 
-**ALLOWED Type Names (use EXACTLY one of these, no exceptions):**
+**ALLOWED Names:**
 {{ ontology_type_names | join(', ') }}
 
 {% endif %}
@@ -62,66 +66,94 @@ The following shows type inheritance relationships (Child → Parent → Grandpa
 - **Entity descriptions must be in English**
 - **Examples must be in English**
 {% endif %}
-- **Semantic Memory Classification (is_explicit_memory):**
-  * Set to `true` if the entity represents **explicit/semantic memory**:
-    - **Concepts:** "Machine Learning", "Photosynthesis", "Democracy"
-    - **Knowledge:** "Python Programming Language", "Theory of Relativity"
-    - **Definitions:** "API (Application Programming Interface)", "REST API"
-    - **Principles:** "SOLID Principles", "First Law of Thermodynamics"
-    - **Theories:** "Evolution Theory", "Quantum Mechanics"
-    - **Methods/Techniques:** "Agile Development", "Machine Learning Algorithm"
-    - **Technical Terms:** "Neural Network", "Database"
-  * Set to `false` for:
-    - **People:** "John Smith", "Dr. Wang"
-    - **Organizations:** "Microsoft", "Harvard University"
-    - **Locations:** "Beijing", "Central Park"
-    - **Events:** "2024 Conference", "Project Meeting"
-    - **Specific objects:** "iPhone 15", "Building A"
-- **Example Generation (IMPORTANT for semantic memory entities):**
-  * For entities where `is_explicit_memory=true`, generate a **concise example (around 20 characters)** to help understand the concept
-  * The example should be:
-    - **Specific and concrete**: Use real-world scenarios or applications
-    - **Brief**: Around 20 characters (can be slightly longer if needed for clarity)
+- **Semantic Memory (is_explicit_memory):**
+  * `true` for: Concepts, Knowledge, Definitions, Theories, Methods (e.g., "Machine Learning", "REST API")
+  * `false` for: People, Organizations, Locations, Events, Specific objects
+  * For `is_explicit_memory=true`, provide concise example (~20 chars{% if language == "zh" %}，使用中文{% endif %})
+
+**🚨🚨🚨 ALIASES & DENIED_ALIASES - MANDATORY FIELDS 🚨🚨🚨**
+
+**CRITICAL RULES (违反将导致提取失败):**
+
+1. **EVERY entity MUST have aliases field:**
+   - `"aliases": [...]` - REQUIRED, even if empty `[]`
+
+2. **ALIASES - 别名提取规则:**
 {% if language == "zh" %}
-    - **使用中文**
+   - 包含：昵称、全名、简称、别称、网名等
+   - 顺序：**第一个别名将作为用户的主显示名称（other_name），必须把最重要/最常用的名字放在第一位**
+   - 提取顺序：严格按照对话中首次出现的顺序
+   - 示例：
+     * "我叫张三，大家叫我小张" → aliases=["张三", "小张"]（张三是第一个，将成为 other_name）
+     * "大家叫我小李，我全名叫李明" → aliases=["小李", "李明"]（小李先出现，将成为 other_name）
+   - 空值：如果没有别名，使用 `[]`
+   - 重要：只提取本次对话中明确提到的别名，不要推测或添加未提及的名字
 {% else %}
-    - **In English**
+   - Include: nicknames, full names, abbreviations, alternative names
+   - Order: **The FIRST alias will be used as the user's primary display name (other_name). Put the most important/frequently used name FIRST**
+   - Extraction order: Strictly follow the order of first appearance in conversation
+   - Examples:
+     * "I'm John, people call me Johnny" → aliases=["John", "Johnny"] (John is first, will become other_name)
+     * "People call me Mike, my full name is Michael" → aliases=["Mike", "Michael"] (Mike appears first, will become other_name)
+   - Empty: If no aliases, use `[]`
+   - Important: Only extract aliases explicitly mentioned in current conversation, do not infer or add unmentioned names
 {% endif %}
-  * For non-semantic entities (`is_explicit_memory=false`), the example field can be empty
-- **Aliases Extraction:**
+
+
+
+3. **USER ENTITY SPECIAL HANDLING:**
 {% if language == "zh" %}
-  * 别名使用中文
+   - 用户实体的 name 字段：使用 "用户" 或 "我"
+   - 用户的真实姓名：放入 aliases
+   - **🚨 禁止将 "用户"、"我" 放入 aliases 中，aliases 只能包含用户的真实姓名、昵称等**
+   - 示例：
+     * "我叫李明" → name="用户", aliases=["李明"]
+     * ❌ 错误：aliases=["用户", "李明"]（"用户"不是真实姓名，禁止放入 aliases）
+     * ❌ 错误：aliases=["我", "李明"]（"我"不是真实姓名，禁止放入 aliases）
 {% else %}
-  * Aliases should be in English
+   - User entity name field: use "User" or "I"
+   - User's real name: put in aliases
+   - **🚨 NEVER put "User" or "I" in aliases. Aliases must only contain real names, nicknames, etc.**
+   - Examples:
+     * "I'm John" → name="User", aliases=["John"]
+     * ❌ Wrong: aliases=["User", "John"] ("User" is not a real name, FORBIDDEN in aliases)
+     * ❌ Wrong: aliases=["I", "John"] ("I" is not a real name, FORBIDDEN in aliases)
 {% endif %}
-  * Include common alternative names, abbreviations and full names
-  * If no aliases exist, use empty array: []
-- Exclude lengthy quotes, calendar dates, temporal ranges, and temporal expressions
-- For numeric values: extract as separate entities (instance_of: 'Numeric', name: units, numeric_value: value)
-  Example: £30 → name: 'GBP', numeric_value: 30, instance_of: 'Numeric'
+
+
+
+4. **ALIASES ORDER:**
+{% if language == "zh" %}
+   - 顺序优先级：按出现顺序，先出现的在前
+{% else %}
+   - Order priority: by appearance order, first mentioned comes first
+{% endif %}
+
+**EXAMPLES OF CORRECT EXTRACTION:**
+{% if language == "zh" %}
+- "我叫张三" → aliases=["张三"] （张三将成为 other_name）
+- "大家叫我小明，我全名叫李明" → aliases=["小明", "李明"] （小明先出现，将成为 other_name）
+- "我是李华，网名叫华仔" → aliases=["李华", "华仔"] （李华先出现，将成为 other_name）
+{% else %}
+- "I'm John" → aliases=["John"] (John will become other_name)
+- "People call me Mike, my full name is Michael" → aliases=["Mike", "Michael"] (Mike appears first, will become other_name)
+- "I'm John Smith, username JSmith" → aliases=["John Smith", "JSmith"] (John Smith appears first, will become other_name)
+{% endif %}
+  
+- Exclude lengthy quotes, dates, temporal expressions
+- Numeric values: extract as entities (instance_of: 'Numeric', name: units, numeric_value: value)
 
 **Triplet Extraction:**
-- Extract (subject, predicate, object) triplets where:
-  - Subject: main entity performing the action or being described
-  - Predicate: relationship between entities (e.g., 'is', 'works at', 'believes')
-  - Object: entity, value, or concept affected by the predicate
+- Extract (subject, predicate, object) where subject/object are entities, predicate is relationship
 {% if language == "zh" %}
-- subject_name 和 object_name 必须使用中文
+- subject_name 和 object_name 使用中文
 {% else %}
-- subject_name and object_name must be in English (translate if original is in another language)
+- subject_name and object_name in English
 {% endif %}
-- Exclude all temporal expressions from every field
-- Use ONLY the predicates listed in "Predicate Instructions" (uppercase English tokens)
-- Do NOT translate predicate tokens
-- Do NOT include `statement_id` field (assigned automatically)
-
-**When NOT to extract triplets:**
-- Non-propositional utterances (emotions, fillers, onomatopoeia)
-- No clear predicate from the given definitions applies
-- Standalone noun phrases or checklist items → extract as entities only
-- Do NOT invent generic predicates (e.g., "IS_DOING", "FEELS", "MENTIONS")
-
-**If no valid triplet exists:** Return triplets: [], extract entities if present, otherwise both arrays empty.
+- Use ONLY predicates from "Predicate Instructions" (uppercase tokens)
+- Exclude temporal expressions, do NOT include `statement_id`
+- **When NOT to extract:** emotions, fillers, no clear predicate, standalone nouns
+- **If no valid triplet:** Return triplets: []
 {%- if predicate_instructions -%}
 
 **Predicate Instructions:**
@@ -207,26 +239,44 @@ Output:
     {"entity_idx": 0, "name": "三脚架", "type": "Equipment", "description": "摄影器材配件", "example": "", "aliases": ["相机三脚架"], "is_explicit_memory": false}
   ]
 }
+
+**Example 4 (别名 - Chinese):** "我的名字是乐力齐，我的小名是齐齐，同事们都叫我小乐"
+Output:
+{
+  "triplets": [],
+  "entities": [
+    {"entity_idx": 0, "name": "用户", "type": "Person", "description": "用户本人", "example": "", "aliases": ["乐力齐", "齐齐", "小乐"], "is_explicit_memory": false}
+  ]
+}
+
+**Example 5 (别名顺序 - Chinese):** "我叫陈思远。对了，我的网名叫「远山」"
+Output:
+{
+  "triplets": [],
+  "entities": [
+    {"entity_idx": 0, "name": "用户", "type": "Person", "description": "用户本人", "example": "", "aliases": ["陈思远", "远山"], "is_explicit_memory": false}
+  ]
+}
+
+
 {% endif %}
 ===End of Examples===
 
 {% if ontology_types %}
-**⚠️ REMINDER: The examples above use generic type names for illustration only. You MUST use ONLY the predefined ontology type names from the "ALLOWED Type Names" list above. For example, use "PredictiveMaintenance" instead of "Concept", use "ProductionLine" instead of "Equipment", etc. Map each entity to the closest matching predefined type.**
+**⚠️ REMINDER: Examples use generic types for illustration. You MUST use predefined types from "ALLOWED Names" above.**
 {% endif %}
 
 ===Output Format===
 
 **JSON Requirements:**
-- Use only ASCII double quotes (") for JSON structure
-- Never use Chinese quotation marks ("") or Unicode quotes
-- Escape quotation marks in text with backslashes (\")
-- Ensure proper string closure and comma separation
-- No line breaks within JSON string values
+- Use ASCII double quotes ("), escape with \"
+- No Chinese quotes (""), no line breaks in strings
 {% if language == "zh" %}
-- **语言要求：实体名称（name）、描述（description）、示例（example）、subject_name、object_name 必须使用中文**
+- **语言：name、description、example、subject_name、object_name 使用中文**
 {% else %}
-- **Language Requirement: Entity names, descriptions, examples, subject_name, object_name must be in English**
-- **If the original text is in Chinese, translate all names to English**
+- **Language: names, descriptions, examples in English (translate if needed)**
 {% endif %}
+- **⚠️ ALIASES ORDER: preserve temporal order of appearance**
+- **🚨 MANDATORY FIELD: EVERY entity MUST include "aliases" field, even if empty array []**
 
 {{ json_schema }}

api/app/core/models/__init__.py

@@ -2,6 +2,7 @@ from .base import  RedBearModelConfig, get_provider_llm_class, RedBearModelFacto
 from .llm import RedBearLLM
 from .embedding import RedBearEmbeddings
 from .rerank import RedBearRerank
+from .generation import RedBearImageGenerator, RedBearVideoGenerator
 
 __all__ = [
     "RedBearModelConfig",
@@ -9,5 +10,7 @@ __all__ = [
     "RedBearEmbeddings",
     "RedBearRerank",
     "RedBearModelFactory",
-    "get_provider_llm_class"
+    "get_provider_llm_class",
+    "RedBearImageGenerator",
+    "RedBearVideoGenerator"
 ]

api/app/core/models/base.py

@@ -67,7 +67,7 @@ class RedBearModelFactory:
                 **config.extra_params
             }
 
-        if provider in [ModelProvider.OPENAI, ModelProvider.XINFERENCE, ModelProvider.GPUSTACK, ModelProvider.OLLAMA]:
+        if provider in [ModelProvider.OPENAI, ModelProvider.XINFERENCE, ModelProvider.GPUSTACK, ModelProvider.OLLAMA, ModelProvider.VOLCANO]:
             # 使用 httpx.Timeout 对象来设置详细的超时配置
             # 这样可以分别控制连接超时和读取超时
             import httpx
@@ -160,11 +160,13 @@ def get_provider_llm_class(config: RedBearModelConfig, type: ModelType = ModelTy
     # dashscope 的 omni 模型使用 OpenAI 兼容模式
     if provider == ModelProvider.DASHSCOPE and config.is_omni:
         return ChatOpenAI
-    if provider in [ModelProvider.OPENAI, ModelProvider.XINFERENCE, ModelProvider.GPUSTACK]:
+    if provider in [ModelProvider.OPENAI, ModelProvider.XINFERENCE, ModelProvider.GPUSTACK, ModelProvider.VOLCANO]:
         if type == ModelType.LLM:
             return OpenAI
         elif type == ModelType.CHAT:
             return ChatOpenAI
+        else:
+            raise BusinessException(f"不支持的模型提供商及类型: {provider}-{type}", code=BizCode.PROVIDER_NOT_SUPPORTED)
     elif provider == ModelProvider.DASHSCOPE:
         return ChatTongyi
     elif provider == ModelProvider.OLLAMA:

api/app/core/models/embedding.py

@@ -1,23 +1,190 @@
 
-from typing import Any, Dict, List, Optional, TypeVar, Callable
+from typing import Any, Dict, List, Optional, Union
 from langchain_core.embeddings import Embeddings
 
-from app.core.models.base import RedBearModelConfig,get_provider_embedding_class,RedBearModelFactory
+from app.core.models.base import RedBearModelConfig, get_provider_embedding_class, RedBearModelFactory
+from app.models.models_model import ModelProvider
+
 
 class RedBearEmbeddings(Embeddings):
-    """Embedding → 完全符合 LangChain Embeddings"""
+    """统一的 Embedding 类，自动支持多模态（根据 provider 判断）"""
+    
     def __init__(self, config: RedBearModelConfig):
-        self._model = self._create_model(config)
         self._config = config
+        self._is_volcano = config.provider.lower() == ModelProvider.VOLCANO
+        
+        if self._is_volcano:
+            # 火山引擎使用 Ark SDK
+            self._client = self._create_volcano_client(config)
+            self._model = None
+        else:
+            # 其他 provider 使用 LangChain
+            self._model = self._create_model(config)
+            self._client = None
 
     def _create_model(self, config: RedBearModelConfig) -> Embeddings:
-        """根据配置创建模型"""
+        """根据配置创建 LangChain 模型"""
         embedding_class = get_provider_embedding_class(config.provider)
         model_params = RedBearModelFactory.get_model_params(config)
         return embedding_class(**model_params)
+    
+    def _create_volcano_client(self, config: RedBearModelConfig):
+        """创建火山引擎客户端"""
+        from volcenginesdkarkruntime import Ark
+        return Ark(api_key=config.api_key, base_url=config.base_url)
 
+    # ==================== LangChain 标准接口 ====================
+    
     def embed_documents(self, texts: list[str]) -> list[list[float]]:
-        return self._model.embed_documents(texts)
+        """批量文本向量化（LangChain 标准接口）"""
+        if self._is_volcano:
+            # 火山引擎多模态 Embedding
+            contents = [{"type": "text", "text": text} for text in texts]
+            response = self._client.multimodal_embeddings.create(
+                model=self._config.model_name,
+                input=contents,
+                encoding_format="float"
+            )
+            return [response.data.embedding]
+        else:
+            # 其他 provider
+            return self._model.embed_documents(texts)
 
     def embed_query(self, text: str) -> List[float]:
-        return self._model.embed_query(text)
+        """单个文本向量化（LangChain 标准接口）"""
+        if self._is_volcano:
+            # 火山引擎多模态 Embedding
+            result = self.embed_documents([text])
+            return result[0] if result else []
+        else:
+            # 其他 provider
+            return self._model.embed_query(text)
+    
+    # ==================== 多模态扩展方法 ====================
+    
+    def embed_multimodal(
+        self,
+        contents: List[Dict[str, Any]],
+        **kwargs
+    ) -> List[List[float]]:
+        """
+        多模态向量化（仅火山引擎支持）
+        
+        Args:
+            contents: 内容列表，格式：
+                - 文本: {"type": "text", "text": "..."}
+                - 图片: {"type": "image_url", "image_url": {"url": "..."}}
+                - 视频: {"type": "video_url", "video_url": {"url": "..."}}
+            **kwargs: 其他参数
+            
+        Returns:
+            向量列表
+        """
+        if not self._is_volcano:
+            raise NotImplementedError(
+                f"多模态 Embedding 仅支持火山引擎，当前 provider: {self._config.provider}"
+            )
+        
+        response = self._client.multimodal_embeddings.create(
+            model=self._config.model_name,
+            input=contents,
+            **kwargs
+        )
+        return [response.data.embedding]
+    
+    async def aembed_multimodal(
+        self,
+        contents: List[Dict[str, Any]],
+        **kwargs
+    ) -> List[List[float]]:
+        """异步多模态向量化"""
+        # 火山引擎 SDK 暂不支持异步，使用同步方法
+        return self.embed_multimodal(contents, **kwargs)
+    
+    def embed_text(self, text: str, **kwargs) -> List[float]:
+        """文本向量化（便捷方法）"""
+        if self._is_volcano:
+            result = self.embed_multimodal(
+                [{"type": "text", "text": text}],
+                **kwargs
+            )
+            return result[0] if result else []
+        else:
+            return self.embed_query(text)
+    
+    def embed_image(self, image_url: str, **kwargs) -> List[float]:
+        """图片向量化（仅火山引擎支持）"""
+        if not self._is_volcano:
+            raise NotImplementedError(
+                f"图片向量化仅支持火山引擎，当前 provider: {self._config.provider}"
+            )
+        
+        result = self.embed_multimodal(
+            [{"type": "image_url", "image_url": {"url": image_url}}],
+            **kwargs
+        )
+        return result[0] if result else []
+    
+    def embed_video(self, video_url: str, **kwargs) -> List[float]:
+        """视频向量化（仅火山引擎支持）"""
+        if not self._is_volcano:
+            raise NotImplementedError(
+                f"视频向量化仅支持火山引擎，当前 provider: {self._config.provider}"
+            )
+        
+        result = self.embed_multimodal(
+            [{"type": "video_url", "video_url": {"url": video_url}}],
+            **kwargs
+        )
+        return result[0] if result else []
+    
+    def embed_batch(
+        self,
+        items: List[Union[str, Dict[str, Any]]],
+        **kwargs
+    ) -> List[List[float]]:
+        """
+        批量向量化（支持混合类型）
+        
+        Args:
+            items: 可以是字符串列表或内容字典列表
+            **kwargs: 其他参数
+            
+        Returns:
+            向量列表
+        """
+        # 如果全是字符串，使用标准方法
+        if all(isinstance(item, str) for item in items):
+            return self.embed_documents(items)
+        
+        # 如果包含字典，需要多模态支持
+        if not self._is_volcano:
+            raise NotImplementedError(
+                f"混合类型批量向量化仅支持火山引擎，当前 provider: {self._config.provider}"
+            )
+        
+        # 标准化输入格式
+        contents = []
+        for item in items:
+            if isinstance(item, str):
+                contents.append({"type": "text", "text": item})
+            elif isinstance(item, dict):
+                contents.append(item)
+            else:
+                raise ValueError(f"不支持的输入类型: {type(item)}")
+        
+        return self.embed_multimodal(contents, **kwargs)
+    
+    # ==================== 工具方法 ====================
+    
+    def is_multimodal_supported(self) -> bool:
+        """检查是否支持多模态"""
+        return self._is_volcano
+    
+    def get_provider(self) -> str:
+        """获取 provider"""
+        return self._config.provider
+
+
+# 保留 RedBearMultimodalEmbeddings 作为别名，向后兼容
+RedBearMultimodalEmbeddings = RedBearEmbeddings

api/app/core/models/generation.py

@@ -0,0 +1,344 @@
+"""
+图片和视频生成模型封装
+
+支持的 Provider:
+- Volcano (火山引擎): 使用 volcenginesdkarkruntime
+- OpenAI: 使用 openai SDK
+"""
+from typing import Any, Dict, Optional
+
+from volcenginesdkarkruntime import Ark
+from volcenginesdkarkruntime.types.images.images import (
+    SequentialImageGenerationOptions,
+    ContentGenerationTool,
+    OptimizePromptOptions
+)
+
+from app.core.models.base import RedBearModelConfig
+from app.core.exceptions import BusinessException
+from app.core.error_codes import BizCode
+from app.models.models_model import ModelProvider
+
+
+class RedBearImageGenerator:
+    """图片生成模型封装"""
+    
+    def __init__(self, config: RedBearModelConfig):
+        self._config = config
+        self._client = self._create_client(config)
+    
+    def _create_client(self, config: RedBearModelConfig):
+        """根据 provider 创建客户端"""
+        provider = config.provider.lower()
+        
+        if provider == ModelProvider.VOLCANO:
+            return Ark(api_key=config.api_key, base_url=config.base_url)
+        # elif provider == ModelProvider.OPENAI:
+        #     from openai import OpenAI
+        #     return OpenAI(api_key=config.api_key, base_url=config.base_url)
+        else:
+            raise BusinessException(
+                f"不支持的图片生成提供商: {provider}",
+                code=BizCode.PROVIDER_NOT_SUPPORTED
+            )
+    
+    def generate(
+        self,
+        prompt: str,
+        image: Optional[Any] = None,
+        size: Optional[str] = "2K",
+        output_format: str = "png",
+        response_format: str = "url",
+        watermark: bool = False,
+        sequential_image_generation: Optional[str] = None,
+        sequential_image_generation_options: Optional[Dict] = None,
+        tools: Optional[list] = None,
+        optimize_prompt_options: Optional[Dict] = None,
+        stream: bool = False,
+        **kwargs
+    ) -> Dict[str, Any]:
+        """
+        生成图片
+        
+        Args:
+            prompt: 提示词
+            image: 参考图片URL或URL列表（图文生图/多图融合）
+            size: 图片尺寸，支持 "2K", "2048x2048", "1920x1080" 等（至少3686400像素）
+            output_format: 输出格式，如 "png", "jpg"
+            response_format: 返回格式，"url" 或 "b64_json"
+            watermark: 是否添加水印
+            sequential_image_generation: 组图生成模式，"auto" 或 "disabled"
+            sequential_image_generation_options: 组图生成选项，如 {"max_images": 4}
+            tools: 工具列表，如 [{"type": "web_search"}] 用于联网搜索生图
+            optimize_prompt_options: 提示词优化选项，如 {"mode": "fast"}
+            stream: 是否使用流式生成
+            **kwargs: 其他参数
+            
+        Returns:
+            生成结果
+        """
+        provider = self._config.provider.lower()
+        
+        if provider == ModelProvider.VOLCANO:
+            params = {
+                "model": self._config.model_name,
+                "prompt": prompt,
+                "size": size,
+                "output_format": output_format,
+                "response_format": response_format,
+                "watermark": watermark,
+            }
+            
+            if image is not None:
+                params["image"] = image
+            
+            if sequential_image_generation:
+                params["sequential_image_generation"] = sequential_image_generation
+                if sequential_image_generation_options:
+                    params["sequential_image_generation_options"] = SequentialImageGenerationOptions(
+                        **sequential_image_generation_options
+                    )
+            
+            if tools:
+                params["tools"] = [ContentGenerationTool(**tool) if isinstance(tool, dict) else tool for tool in tools]
+            
+            if optimize_prompt_options:
+                params["optimize_prompt_options"] = OptimizePromptOptions(**optimize_prompt_options)
+            
+            if stream:
+                params["stream"] = True
+            
+            params.update(kwargs)
+            response = self._client.images.generate(**params)
+            
+        # elif provider == ModelProvider.OPENAI:
+        #     response = self._client.images.generate(
+        #         model=self._config.model_name,
+        #         prompt=prompt,
+        #         size=size,
+        #         n=n,
+        #         **kwargs
+        #     )
+        else:
+            raise BusinessException(
+                f"不支持的提供商: {provider}",
+                code=BizCode.PROVIDER_NOT_SUPPORTED
+            )
+        
+        return response.model_dump() if hasattr(response, 'model_dump') else response
+    
+    async def agenerate(
+        self,
+        prompt: str,
+        image: Optional[Any] = None,
+        size: Optional[str] = "2K",
+        output_format: str = "png",
+        response_format: str = "url",
+        watermark: bool = False,
+        **kwargs
+    ) -> Dict[str, Any]:
+        """异步生成图片"""
+        return self.generate(prompt, image, size, output_format, response_format, watermark, **kwargs)
+
+
+class RedBearVideoGenerator:
+    """视频生成模型封装"""
+    
+    def __init__(self, config: RedBearModelConfig):
+        self._config = config
+        self._client = self._create_client(config)
+    
+    def _create_client(self, config: RedBearModelConfig):
+        """根据 provider 创建客户端"""
+        provider = config.provider.lower()
+        
+        if provider == ModelProvider.VOLCANO:
+            return Ark(api_key=config.api_key, base_url=config.base_url)
+        else:
+            raise BusinessException(
+                f"不支持的视频生成提供商: {provider}",
+                code=BizCode.PROVIDER_NOT_SUPPORTED
+            )
+    
+    def generate(
+        self,
+        prompt: str,
+        image_url: Optional[str] = None,
+        first_frame_url: Optional[str] = None,
+        last_frame_url: Optional[str] = None,
+        reference_images: Optional[list] = None,
+        draft_task_id: Optional[str] = None,
+        duration: Optional[int] = None,
+        frames: Optional[int] = None,
+        ratio: Optional[str] = None,
+        resolution: Optional[str] = None,
+        generate_audio: bool = False,
+        watermark: bool = False,
+        camera_fixed: bool = False,
+        seed: Optional[int] = None,
+        return_last_frame: bool = False,
+        service_tier: str = "default",
+        execution_expires_after: Optional[int] = None,
+        draft: bool = False,
+        **kwargs
+    ) -> Dict[str, Any]:
+        """
+        生成视频
+        
+        Args:
+            prompt: 提示词
+            image_url: 首帧图片URL（图生视频-基于首帧）
+            first_frame_url: 首帧图片URL（图生视频-基于首尾帧）
+            last_frame_url: 尾帧图片URL（图生视频-基于首尾帧）
+            reference_images: 参考图片URL列表（图生视频-基于参考图）
+            draft_task_id: Draft任务ID（基于Draft生成正式视频）
+            duration: 视频时长（秒），与frames二选一
+            frames: 视频帧数，与duration二选一
+            ratio: 视频比例，如 "16:9", "9:16", "adaptive"
+            resolution: 视频分辨率，如 "720p", "1080p"
+            generate_audio: 是否生成音频
+            watermark: 是否添加水印
+            camera_fixed: 是否固定镜头
+            seed: 随机种子
+            return_last_frame: 是否返回最后一帧
+            service_tier: 服务层级，"default" 或 "flex"（离线推理）
+            execution_expires_after: 任务过期时间（秒）
+            draft: 是否生成样片
+            **kwargs: 其他参数
+            
+        Returns:
+            生成结果（包含任务ID，需要轮询获取结果）
+        """
+        provider = self._config.provider.lower()
+        
+        if provider == ModelProvider.VOLCANO:
+            content = [{"type": "text", "text": prompt}]
+            
+            if draft_task_id:
+                content = [{"type": "draft_task", "draft_task": {"id": draft_task_id}}]
+            else:
+                if image_url:
+                    content.append({"type": "image_url", "image_url": {"url": image_url}})
+                
+                if first_frame_url:
+                    content.append({"type": "image_url", "image_url": {"url": first_frame_url}, "role": "first_frame"})
+                if last_frame_url:
+                    content.append({"type": "image_url", "image_url": {"url": last_frame_url}, "role": "last_frame"})
+                
+                if reference_images:
+                    for ref_url in reference_images:
+                        content.append({"type": "image_url", "image_url": {"url": ref_url}, "role": "reference_image"})
+            
+            params = {"model": self._config.model_name, "content": content, "watermark": watermark}
+            
+            if duration:
+                params["duration"] = duration
+            if frames:
+                params["frames"] = frames
+            if ratio:
+                params["ratio"] = ratio
+            if resolution:
+                params["resolution"] = resolution
+            if generate_audio:
+                params["generate_audio"] = generate_audio
+            if camera_fixed:
+                params["camera_fixed"] = camera_fixed
+            if seed is not None:
+                params["seed"] = seed
+            if return_last_frame:
+                params["return_last_frame"] = return_last_frame
+            if service_tier != "default":
+                params["service_tier"] = service_tier
+            if execution_expires_after:
+                params["execution_expires_after"] = execution_expires_after
+            if draft:
+                params["draft"] = draft
+            
+            params.update(kwargs)
+            response = self._client.content_generation.tasks.create(**params)
+        else:
+            raise BusinessException(
+                f"不支持的提供商: {provider}",
+                code=BizCode.PROVIDER_NOT_SUPPORTED
+            )
+        
+        return response.model_dump() if hasattr(response, 'model_dump') else response
+    
+    async def agenerate(
+        self,
+        prompt: str,
+        image_url: Optional[str] = None,
+        duration: Optional[int] = None,
+        **kwargs
+    ) -> Dict[str, Any]:
+        """异步生成视频"""
+        return self.generate(prompt, image_url=image_url, duration=duration, **kwargs)
+    
+    def get_task_status(self, task_id: str) -> Dict[str, Any]:
+        """
+        查询视频生成任务状态
+        
+        Args:
+            task_id: 任务ID
+            
+        Returns:
+            任务状态信息
+        """
+        provider = self._config.provider.lower()
+        
+        if provider == ModelProvider.VOLCANO:
+            response = self._client.content_generation.tasks.get(task_id=task_id)
+            return response.model_dump() if hasattr(response, 'model_dump') else response
+        else:
+            raise BusinessException(
+                f"不支持的提供商: {provider}",
+                code=BizCode.PROVIDER_NOT_SUPPORTED
+            )
+    
+    async def aget_task_status(self, task_id: str) -> Dict[str, Any]:
+        """异步查询任务状态"""
+        return self.get_task_status(task_id)
+    
+    def list_tasks(self, page_size: int = 10, status: Optional[str] = None, **kwargs) -> Dict[str, Any]:
+        """
+        查询视频生成任务列表
+        
+        Args:
+            page_size: 每页数量
+            status: 任务状态筛选，如 "succeeded", "failed", "pending"
+            **kwargs: 其他参数
+            
+        Returns:
+            任务列表
+        """
+        provider = self._config.provider.lower()
+        
+        if provider == ModelProvider.VOLCANO:
+            params = {"page_size": page_size}
+            if status:
+                params["status"] = status
+            params.update(kwargs)
+            response = self._client.content_generation.tasks.list(**params)
+            return response.model_dump() if hasattr(response, 'model_dump') else response
+        else:
+            raise BusinessException(
+                f"不支持的提供商: {provider}",
+                code=BizCode.PROVIDER_NOT_SUPPORTED
+            )
+    
+    def delete_task(self, task_id: str) -> None:
+        """
+        删除或取消视频生成任务
+        
+        Args:
+            task_id: 任务ID
+        """
+        provider = self._config.provider.lower()
+        
+        if provider == ModelProvider.VOLCANO:
+            self._client.content_generation.tasks.delete(task_id=task_id)
+        else:
+            raise BusinessException(
+                f"不支持的提供商: {provider}",
+                code=BizCode.PROVIDER_NOT_SUPPORTED
+            )

api/app/core/models/scripts/volcano_models.yaml

@@ -0,0 +1,334 @@
+provider: volcano
+models:
+# Doubao-Seed 2.0 系列
+- name: doubao-seed-2-0-pro-260215
+  type: chat
+  provider: volcano
+  description: 旗舰级全能通用模型，面向 Agent 时代的复杂推理与长链路任务执行场景。强调多模态理解、长上下文推理、结构化生成与工具增强执行。复杂指令与多约束执行能力突出，可稳定应对多步复杂规划、复杂图文推理、视频内容理解与高难度分析等场景。侧重长链路推理能力与复杂任务稳定性，适配真实业务中的复杂场景。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+    - video
+  is_omni: false
+  tags:
+  - 大语言模型
+  logo: volcano
+
+- name: doubao-seed-2-0-lite-260215
+  type: chat
+  provider: volcano
+  description: 面向高频企业场景兼顾性能与成本的均衡型模型，综合能力超越上一代Doubao-Seed-1.8。胜任非结构化信息处理、内容创作、搜索推荐、数据分析等生产型工作，支持长上下文、多源信息融合、多步指令执行与高保真结构化输出。在保障稳定效果的同时显著优化成本。兼顾生成质量与响应速度，适合作为通用生产级模型。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+    - video
+  is_omni: false
+  tags:
+  - 大语言模型
+  logo: volcano
+
+- name: doubao-seed-2-0-mini-260215
+  type: chat
+  provider: volcano
+  description: 面向低时延、高并发与成本敏感场景，提供极致的模型推理速度。模型效果与Doubao-Seed-1.6相当。支持256k上下文、4档思考长度和多模态理解，适合成本和速度优先的轻量级任务。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+    - video
+  is_omni: false
+  tags:
+  - 大语言模型
+  logo: volcano
+
+- name: doubao-seed-2-0-code-preview-260215
+  type: chat
+  provider: volcano
+  description: 面向真实编程环境优化的 Coding 模型，能稳定调用 Claude Code 等常见 IDE 中的工具。模型特别优化了前端能力，在使用常见的前端框架时能有良好表现。模型支持使用 Skills，可以配合多种自定义技能使用。Seed 2.0 的编程加强版，更适合 Agentic Coding。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+    - video
+  is_omni: false
+  tags:
+  - 大语言模型
+  - 代码模型
+  logo: volcano
+
+# Doubao-Seed 1.x 系列
+- name: doubao-seed-1-8-251228
+  type: chat
+  provider: volcano
+  description: Doubao-Seed-1.8 面向多模态 Agent 场景定向优化。Agent 能力上，Tool Use、复杂指令遵循等能力均大幅增强。多模态理解方面，视觉基础能力显著提升，可低帧率理解超长视频，视频运动理解、复杂空间理解及文档结构化解析能力也有所优化，还原生支持智能上下文管理，用户可配置上下文策略。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+    - video
+  is_omni: false
+  tags:
+  - 大语言模型
+  logo: volcano
+
+- name: doubao-seed-1-6-251015
+  type: chat
+  provider: volcano
+  description: Doubao-Seed-1.6全新多模态深度思考模型，同时支持minimal/low/medium/high 四种reasoning effort。 更强模型效果，服务复杂任务和有挑战场景。支持 256k 上下文窗口，输出长度支持最大 32k tokens。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+    - video
+  is_omni: false
+  tags:
+  - 大语言模型
+  logo: volcano
+
+- name: doubao-seed-1-6-lite-251015
+  type: chat
+  provider: volcano
+  description: 更高性价比，常见任务的最佳选择，支持minimal、low、medium、high 四种reasoning_effort思考深度
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+    - video
+  is_omni: false
+  tags:
+  - 大语言模型
+  logo: volcano
+
+- name: doubao-seed-1-6-flash-250828
+  type: chat
+  provider: volcano
+  description: Doubao-Seed-1.6-flash推理速度极致的多模态深度思考模型，TPOT低至10ms； 同时支持文本和视觉理解，文本理解能力超过上一代lite，视觉理解比肩友商pro系列模型。支持 256k 上下文窗口，输出长度支持最大 16k tokens。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+    - video
+  is_omni: false
+  tags:
+  - 大语言模型
+  logo: volcano
+
+- name: doubao-seed-code-preview-251028
+  type: chat
+  provider: volcano
+  description: 面向Agentic编程任务进行了深度优化。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+    - video
+  is_omni: false
+  tags:
+  - 大语言模型
+  - 代码模型
+  logo: volcano
+
+- name: doubao-seed-1-6-vision-250815
+  type: chat
+  provider: volcano
+  description: 全新Doubao-Seed-1.6系列视觉深度思考模型，视觉理解能力显著增强，并支持image_process视觉工具
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+    - video
+  is_omni: false
+  tags:
+  - 大语言模型
+  - 多模态模型
+  logo: volcano
+
+# Doubao 1.5 系列
+- name: doubao-1-5-vision-pro-32k-250115
+  type: chat
+  provider: volcano
+  description: 全新升级的多模态大模型，支持任意分辨率和极端长宽比图像识别，增强视觉推理、文档识别、细节信息理解和指令遵循能力。支持 32k 上下文窗口，输出长度支持最大 12k tokens。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+  is_omni: false
+  tags:
+  - 大语言模型
+  - 多模态模型
+  logo: volcano
+
+- name: doubao-1-5-pro-32k-250115
+  type: chat
+  provider: volcano
+  description: 全新一代主力模型，性能全面升级，在知识、代码、推理等方面表现卓越。最大支持 128k 上下文窗口，输出长度支持最大 12k tokens。
+  is_deprecated: false
+  is_official: true
+  capability: []
+  is_omni: false
+  tags:
+  - 大语言模型
+  logo: volcano
+
+- name: doubao-1-5-lite-32k-250115
+  type: chat
+  provider: volcano
+  description: 全新一代轻量版模型，极致响应速度，效果与时延均达到全球一流水平。支持 32k 上下文窗口，输出长度支持最大 12k tokens。
+  is_deprecated: false
+  is_official: true
+  capability: []
+  is_omni: false
+  tags:
+  - 大语言模型
+  logo: volcano
+
+# Doubao-Seedance 视频生成系列
+- name: doubao-seedance-1-5-pro-251215
+  type: video
+  provider: volcano
+  description: 豆包视频生成模型Seedance 1.5 pro 作为全球领先的视频生成模型，可生成音画高精同步的视频内容。支持多人多语言对白，全面覆盖环境音、动作音、合成音、乐器音、背景音及人声，支持首尾帧，实现影视级叙事效果，满足影视、漫剧、电商及广告领域的高阶创作需求。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+  is_omni: false
+  tags:
+  - 视频生成
+  logo:  volcano
+
+- name: doubao-seedance-1-0-pro-250528
+  type: video
+  provider: volcano
+  description: 一款支持多镜头叙事的视频生成基础模型，在各维度表现出色。它在语义理解与指令遵循能力上取得突破，能生成运动流畅、细节丰富、风格多样且具备影视级美感的 1080P 高清视频
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+  is_omni: false
+  tags:
+  - 视频生成
+  logo: volcano
+
+- name: doubao-seedance-1-0-pro-fast-251015
+  type: video
+  provider: volcano
+  description: 一款价格触底、效能封顶的全面模型，在视频生成质量、速度、价格之间取得了卓越平衡。它继承了Seedance 1.0 pro 核心优势，同时生成速度提升、价格更具竞争力，为创作者带来效率与成本双重优化的体验。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+  is_omni: false
+  tags:
+  - 视频生成
+  logo: volcano
+
+- name: doubao-seedance-1-0-lite-i2v-250428
+  type: video
+  provider: volcano
+  description: 基于首帧图片、尾帧图片（可选）、参考图片（可选）和文本提示词（可选）相结合的方式生成视频
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+  is_omni: false
+  tags:
+  - 视频生成
+  - 图生视频
+  logo: volcano
+
+- name: doubao-seedance-1-0-lite-t2v-250428
+  type: video
+  provider: volcano
+  description: 基于文本提示词生成视频
+  is_deprecated: false
+  is_official: true
+  capability: []
+  is_omni: false
+  tags:
+  - 视频生成
+  - 文生视频
+  logo: volcano
+
+# Doubao-Seedream 图像生成系列
+- name: doubao-seedream-5-0-260128
+  type: image
+  provider: volcano
+  description: 字节跳动发布的最新图像创作模型。该模型首次搭载联网检索功能，能融合实时网络信息，提升生图时效性。同时，模型的聪明度进一步升级，能够精准解析复杂指令和视觉内容。此外，模型在世界知识广度、参考一致性及专业场景生成质量上均有增强，可更好地满足企业级视觉创作需求。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+  is_omni: false
+  tags:
+  - 图像生成
+  logo: volcano
+
+- name: doubao-seedream-4-5-251128
+  type: image
+  provider: volcano
+  description: 字节跳动最新推出的图像多模态模型，整合了文生图、图生图、组图输出等能力，融合常识和推理能力。相比前代4.0模型生成效果大幅提升，具备更好的编辑一致性和多图融合效果，能更精准的控制画面细节，小字、小人脸生成更自然，图片排版、色彩更和谐，美感提升。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+  is_omni: false
+  tags:
+  - 图像生成
+  logo: volcano
+
+- name: doubao-seedream-4-0-250828
+  type: image
+  provider: volcano
+  description: 基于领先架构的SOTA级多模态图像创作模型，其生成美感、指令遵循、结构完整度、主体保持一致性处于世界头部水平。模型采用同一套架构实现文生图与编辑能力的统一，原生支持文本 、单图和多图输入，并能通过对提示词的深度推理，自动适配最优的图像比例尺寸与生成数量，可一次性连续输出最多 15 张内容关联的图像，支持 4K 超高清输出。
+  is_deprecated: false
+  is_official: true
+  capability:
+    - vision
+  is_omni: false
+  tags:
+  - 图像生成
+  logo: volcano
+
+- name: doubao-seedream-3-0-t2i-250415
+  type: image
+  provider: volcano
+  description: 一款支持原生高分辨率的中英双语图像生成基础模型，综合能力媲美GPT-4o，处于世界第一梯队。支持原生 2K 分辨率输出；响应速度更快；小字生成更准确，文本排版效果增强；指令遵循能力强，美感&结构提升，保真度和细节表现较好。
+  is_deprecated: false
+  is_official: true
+  capability: []
+  is_omni: false
+  tags:
+  - 图像生成
+  - 文生图
+  logo: volcano
+
+# Doubao 翻译系列
+- name: doubao-seed-translation-250915
+  type: chat
+  provider: volcano
+  description: 通用多语言翻译模型，支持30余种语言互译，支持 4K 上下文窗口，输出长度支持最大 3K tokens
+  is_deprecated: false
+  is_official: true
+  capability: []
+  is_omni: false
+  tags:
+  - 翻译模型
+  logo: volcano
+
+# Doubao Embedding 系列
+- name: doubao-embedding-vision-251215
+  type: embedding
+  provider: volcano
+  description: 主要面向图文多模向量检索的使用场景，支持图片输入及中、英双语文本输入，最长 128K 上下文长度。
+  is_deprecated: false
+  is_official: true
+  capability:
+  - vision
+  - video
+  is_omni: false
+  tags:
+  - 向量模型
+  - 多模态模型
+  logo: volcano

api/app/core/rag/vdb/elasticsearch/elasticsearch_vector.py

@@ -61,24 +61,16 @@ class ElasticSearchConfig(BaseModel):
 class ElasticSearchVector(BaseVector):
     def __init__(self, index_name: str, config: ElasticSearchConfig, embedding_config: ModelApiKey, reranker_config: ModelApiKey):
         super().__init__(index_name.lower())
-        # self.embeddings = XinferenceEmbeddings(
-        #     server_url=os.getenv("XINFERENCE_URL", "http://127.0.0.1"),  # Default Xinference port
-        #     model_uid="bge-m3"  # replace model_uid with the model UID return from launching the model
-        # )
-        # Remove debug printing to avoid leaking sensitive information
-        # print("embedding:" + embedding_config.model_name + "|" + embedding_config.provider + "|" + embedding_config.api_key + "|" + embedding_config.api_base)
+        
+        # 初始化 Embedding 模型（自动支持火山引擎多模态）
         self.embeddings = RedBearEmbeddings(RedBearModelConfig(
             model_name=embedding_config.model_name,
             provider=embedding_config.provider,
             api_key=embedding_config.api_key,
             base_url=embedding_config.api_base
         ))
-        # self.reranker = XinferenceRerank(
-        #     server_url=os.getenv("XINFERENCE_URL", "http://127.0.0.1"),
-        #     model_uid="bge-reranker-large"
-        # )
-        # Remove debug printing to avoid leaking sensitive information
-        # print("reranker:"+ reranker_config.model_name + "|" + reranker_config.provider + "|" + reranker_config.api_key + "|" + reranker_config.api_base)
+        self.is_multimodal_embedding = self.embeddings.is_multimodal_supported()
+        
         self.reranker = RedBearRerank(RedBearModelConfig(
             model_name=reranker_config.model_name,
             provider=reranker_config.provider,
@@ -144,7 +136,11 @@ class ElasticSearchVector(BaseVector):
     def add_chunks(self, chunks: list[DocumentChunk], **kwargs):
         # 实现 Elasticsearch 保存向量
         texts = [chunk.page_content for chunk in chunks]
-        embeddings = self.embeddings.embed_documents(list(texts))
+        if self.is_multimodal_embedding:
+            # 火山引擎多模态 Embedding
+            embeddings = self.embeddings.embed_batch(texts)
+        else:
+            embeddings = self.embeddings.embed_documents(list(texts))
         self.create(chunks, embeddings, **kwargs)
 
     def create(self, chunks: list[DocumentChunk], embeddings: list[list[float]], **kwargs):
@@ -394,7 +390,11 @@ class ElasticSearchVector(BaseVector):
             updated count.
         """
         indices = kwargs.get("indices", self._collection_name)  # Default single index, multi-index available，etc "index1,index2,index3"
-        chunk.vector = self.embeddings.embed_query(chunk.page_content)
+        if self.is_multimodal_embedding:
+            # 火山引擎多模态 Embedding
+            chunk.vector = self.embeddings.embed_text(chunk.page_content)
+        else:
+            chunk.vector = self.embeddings.embed_query(chunk.page_content)
 
         body = {
             "script": {
@@ -454,7 +454,11 @@ class ElasticSearchVector(BaseVector):
 
     def search_by_vector(self, query: str, **kwargs: Any) -> list[DocumentChunk]:
         """Search the nearest neighbors to a vector."""
-        query_vector = self.embeddings.embed_query(query)
+        if self.is_multimodal_embedding:
+            # 火山引擎多模态 Embedding
+            query_vector = self.embeddings.embed_text(query)
+        else:
+            query_vector = self.embeddings.embed_query(query)
         top_k = kwargs.get("top_k", 1024)
         score_threshold = float(kwargs.get("score_threshold") or 0.3)
         indices = kwargs.get("indices", self._collection_name)  # Default single index, multi-index available，etc "index1,index2,index3"

api/app/core/storage/base.py

@@ -109,17 +109,13 @@ class StorageBackend(ABC):
         pass
 
     @abstractmethod
-    async def get_url(self, file_key: str, expires: int = 3600) -> str:
-        """
-        Get an access URL for the file.
-
-        Args:
-            file_key: Unique identifier for the file in the storage system.
-            expires: URL validity period in seconds (default: 1 hour).
-
-        Returns:
-            URL for accessing the file.
-        """
+    async def get_url(
+        self,
+        file_key: str,
+        expires: int = 3600,
+        file_name: Optional[str] = None
+    ) -> str:
+        """Get an access URL for the file."""
         pass
 
     async def get_permanent_url(self, file_key: str) -> Optional[str]:

api/app/core/storage/local.py

@@ -210,7 +210,12 @@ class LocalStorage(StorageBackend):
                 cause=e,
             )
 
-    async def get_url(self, file_key: str, expires: int = 3600) -> str:
+    async def get_url(
+        self,
+        file_key: str,
+        expires: int = 3600,
+        file_name: Optional[str] = None
+    ) -> str:
         """
         Get an access URL for the file.
 
@@ -220,6 +225,7 @@ class LocalStorage(StorageBackend):
         Args:
             file_key: Unique identifier for the file in the storage system.
             expires: URL validity period in seconds (not used for local storage).
+            file_name: If set, adds Content-Disposition: attachment to force download.
 
         Returns:
             A relative URL path for accessing the file.

api/app/core/storage/oss.py

@@ -7,6 +7,7 @@ Storage Service (OSS) using the oss2 SDK.
 
 import io
 import logging
+import urllib.parse
 from typing import AsyncIterator, Optional
 
 import oss2
@@ -43,6 +44,8 @@ class OSSStorage(StorageBackend):
         access_key_id: str,
         access_key_secret: str,
         bucket_name: str,
+        connect_timeout: int = 30,
+        multipart_threshold: int = 10 * 1024 * 1024,  # 10MB
     ):
         """
         Initialize the OSSStorage backend.
@@ -52,6 +55,8 @@ class OSSStorage(StorageBackend):
             access_key_id: The Aliyun access key ID.
             access_key_secret: The Aliyun access key secret.
             bucket_name: The name of the OSS bucket.
+            connect_timeout: Connection timeout in seconds (default: 30).
+            multipart_threshold: File size threshold for multipart upload (default: 10MB).
 
         Raises:
             StorageConfigError: If any required configuration is missing.
@@ -68,10 +73,17 @@ class OSSStorage(StorageBackend):
 
         self.endpoint = endpoint
         self.bucket_name = bucket_name
+        self.multipart_threshold = multipart_threshold
 
         try:
             auth = oss2.Auth(access_key_id, access_key_secret)
-            self.bucket = oss2.Bucket(auth, endpoint, bucket_name)
+            # 设置超时和重试
+            self.bucket = oss2.Bucket(
+                auth, 
+                endpoint, 
+                bucket_name,
+                connect_timeout=connect_timeout
+            )
             logger.info(
                 f"OSSStorage initialized with endpoint: {endpoint}, bucket: {bucket_name}"
             )
@@ -107,21 +119,38 @@ class OSSStorage(StorageBackend):
             if content_type:
                 headers["Content-Type"] = content_type
 
-            self.bucket.put_object(file_key, content, headers=headers if headers else None)
+            # 大文件使用分片上传
+            if len(content) > self.multipart_threshold:
+                logger.info(f"Using multipart upload for large file: {file_key} ({len(content)} bytes)")
+                upload_id = self.bucket.init_multipart_upload(file_key, headers=headers if headers else None).upload_id
+                parts = []
+                part_size = 5 * 1024 * 1024  # 5MB per part
+                part_num = 1
+                
+                for offset in range(0, len(content), part_size):
+                    chunk = content[offset:offset + part_size]
+                    result = self.bucket.upload_part(file_key, upload_id, part_num, chunk)
+                    parts.append(oss2.models.PartInfo(part_num, result.etag))
+                    part_num += 1
+                
+                self.bucket.complete_multipart_upload(file_key, upload_id, parts)
+            else:
+                self.bucket.put_object(file_key, content, headers=headers if headers else None)
+            
             logger.info(f"File uploaded to OSS successfully: {file_key}")
             return file_key
 
         except OssError as e:
             logger.error(f"OSS error uploading file {file_key}: {e}")
             raise StorageUploadError(
-                message=f"Failed to upload file to OSS: {e.message}",
+                message=f"Failed to upload file to OSS: {str(e)}",
                 file_key=file_key,
                 cause=e,
             )
         except Exception as e:
             logger.error(f"Failed to upload file to OSS {file_key}: {e}")
             raise StorageUploadError(
-                message=f"Failed to upload file to OSS: {e}",
+                message=f"Failed to upload file to OSS: {str(e)}",
                 file_key=file_key,
                 cause=e,
             )
@@ -134,28 +163,73 @@ class OSSStorage(StorageBackend):
     ) -> int:
         """Upload from async stream to OSS. Returns total bytes written."""
         buf = io.BytesIO()
+        headers = {"Content-Type": content_type} if content_type else None
+        upload_id = None
+        
         try:
+            # 收集流数据
+            total_size = 0
             async for chunk in stream:
+                if not chunk:
+                    continue
                 buf.write(chunk)
+                total_size += len(chunk)
+            
             content = buf.getvalue()
-            headers = {"Content-Type": content_type} if content_type else None
-            self.bucket.put_object(file_key, content, headers=headers)
-            logger.info(f"File stream uploaded to OSS successfully: {file_key}")
-            return len(content)
+            
+            if not content:
+                raise StorageUploadError(
+                    message="Empty stream content",
+                    file_key=file_key,
+                )
+            
+            # 大文件使用分片上传
+            if len(content) > self.multipart_threshold:
+                logger.info(f"Using multipart upload for stream: {file_key} ({len(content)} bytes)")
+                upload_id = self.bucket.init_multipart_upload(file_key, headers=headers).upload_id
+                parts = []
+                part_size = 5 * 1024 * 1024  # 5MB
+                part_num = 1
+                
+                for offset in range(0, len(content), part_size):
+                    chunk = content[offset:offset + part_size]
+                    result = self.bucket.upload_part(file_key, upload_id, part_num, chunk)
+                    parts.append(oss2.models.PartInfo(part_num, result.etag))
+                    part_num += 1
+                
+                self.bucket.complete_multipart_upload(file_key, upload_id, parts)
+            else:
+                self.bucket.put_object(file_key, content, headers=headers)
+            
+            logger.info(f"File stream uploaded to OSS successfully: {file_key} ({total_size} bytes)")
+            return total_size
+            
         except OssError as e:
+            if upload_id:
+                try:
+                    self.bucket.abort_multipart_upload(file_key, upload_id)
+                except:
+                    pass
             logger.error(f"OSS error stream uploading file {file_key}: {e}")
             raise StorageUploadError(
-                message=f"Failed to stream upload file to OSS: {e.message}",
+                message=f"Failed to stream upload file to OSS: {str(e)}",
                 file_key=file_key,
                 cause=e,
             )
         except Exception as e:
+            if upload_id:
+                try:
+                    self.bucket.abort_multipart_upload(file_key, upload_id)
+                except:
+                    pass
             logger.error(f"Failed to stream upload file to OSS {file_key}: {e}")
             raise StorageUploadError(
-                message=f"Failed to stream upload file to OSS: {e}",
+                message=f"Failed to stream upload file to OSS: {str(e)}",
                 file_key=file_key,
                 cause=e,
             )
+        finally:
+            buf.close()
 
     async def download(self, file_key: str) -> bytes:
         """
@@ -181,14 +255,14 @@ class OSSStorage(StorageBackend):
         except OssError as e:
             logger.error(f"OSS error downloading file {file_key}: {e}")
             raise StorageDownloadError(
-                message=f"Failed to download file from OSS: {e.message}",
+                message=f"Failed to download file from OSS: {str(e)}",
                 file_key=file_key,
                 cause=e,
             )
         except Exception as e:
             logger.error(f"Failed to download file from OSS {file_key}: {e}")
             raise StorageDownloadError(
-                message=f"Failed to download file from OSS: {e}",
+                message=f"Failed to download file from OSS: {str(e)}",
                 file_key=file_key,
                 cause=e,
             )
@@ -214,14 +288,14 @@ class OSSStorage(StorageBackend):
         except OssError as e:
             logger.error(f"OSS error deleting file {file_key}: {e}")
             raise StorageDeleteError(
-                message=f"Failed to delete file from OSS: {e.message}",
+                message=f"Failed to delete file from OSS: {str(e)}",
                 file_key=file_key,
                 cause=e,
             )
         except Exception as e:
             logger.error(f"Failed to delete file from OSS {file_key}: {e}")
             raise StorageDeleteError(
-                message=f"Failed to delete file from OSS: {e}",
+                message=f"Failed to delete file from OSS: {str(e)}",
                 file_key=file_key,
                 cause=e,
             )
@@ -242,24 +316,33 @@ class OSSStorage(StorageBackend):
             logger.error(f"Failed to check file existence in OSS {file_key}: {e}")
             return False
 
-    async def get_url(self, file_key: str, expires: int = 3600) -> str:
+    async def get_url(
+        self,
+        file_key: str,
+        expires: int = 3600,
+        file_name: Optional[str] = None,
+    ) -> str:
         """
         Get a presigned URL for accessing the file.
 
         Args:
             file_key: Unique identifier for the file in the storage system.
             expires: URL validity period in seconds (default: 1 hour).
+            file_name: If set, adds Content-Disposition: attachment to force download.
 
         Returns:
             A presigned URL for accessing the file.
         """
         try:
-            url = self.bucket.sign_url("GET", file_key, expires)
+            params = {}
+            if file_name:
+                filename_encoded = urllib.parse.quote(file_name.encode("utf-8"))
+                params["response-content-disposition"] = f"attachment; filename*=UTF-8''{filename_encoded}"
+            url = self.bucket.sign_url("GET", file_key, expires, params=params if params else None)
             logger.debug(f"Generated presigned URL for {file_key}, expires in {expires}s")
             return url
         except Exception as e:
             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:

api/app/core/storage/s3.py

@@ -6,6 +6,7 @@ using the boto3 SDK.
 """
 
 import io
+import urllib.parse
 import logging
 from typing import AsyncIterator, Optional
 
@@ -352,31 +353,37 @@ class S3Storage(StorageBackend):
             logger.error(f"Failed to check file existence in S3 {file_key}: {e}")
             return False
 
-    async def get_url(self, file_key: str, expires: int = 3600) -> str:
+    async def get_url(
+        self,
+        file_key: str,
+        expires: int = 3600,
+        file_name: Optional[str] = None,
+    ) -> str:
         """
         Get a presigned URL for accessing the file.
 
         Args:
             file_key: Unique identifier for the file in the storage system.
             expires: URL validity period in seconds (default: 1 hour).
+            file_name: If set, adds Content-Disposition: attachment to force download.
 
         Returns:
             A presigned URL for accessing the file.
         """
         try:
+            params = {"Bucket": self.bucket_name, "Key": file_key}
+            if file_name:
+                filename_encoded = urllib.parse.quote(file_name.encode("utf-8"))
+                params["ResponseContentDisposition"] = f"attachment; filename*=UTF-8''{filename_encoded}"
             url = self.client.generate_presigned_url(
                 "get_object",
-                Params={
-                    "Bucket": self.bucket_name,
-                    "Key": file_key,
-                },
+                Params=params,
                 ExpiresIn=expires,
             )
             logger.debug(f"Generated presigned URL for {file_key}, expires in {expires}s")
             return url
         except Exception as e:
             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:

api/app/core/tools/mcp/client.py

@@ -99,7 +99,7 @@ class SimpleMCPClient:
             # 建立 SSE 连接
             response = await self._session.get(self.server_url)
             
-            if response.status != 200:
+            if response.status not in (200, 202):
                 error_text = await response.text()
                 raise MCPConnectionError(f"SSE 连接失败 {response.status}: {error_text}")
             
@@ -190,7 +190,9 @@ class SimpleMCPClient:
         
         try:
             async with self._session.post(self._endpoint_url, json=request) as response:
-                if response.status != 200:
+                # MCP SSE 协议：POST 请求返回 200 或 202 均为正常
+                # 202 Accepted 表示请求已接受，结果通过 SSE 流异步返回
+                if response.status not in (200, 202):
                     error_text = await response.text()
                     raise MCPConnectionError(f"请求失败 {response.status}: {error_text}")
             
@@ -205,7 +207,7 @@ class SimpleMCPClient:
             raise MCPConnectionError("endpoint URL 未初始化")
         
         async with self._session.post(self._endpoint_url, json=notification) as response:
-            if response.status != 200:
+            if response.status not in (200, 202):
                 logger.warning(f"通知发送失败: {response.status}")
     
     async def _initialize_modelscope_session(self):

api/app/core/workflow/adapters/base_adapter.py

@@ -9,7 +9,7 @@ from typing import Any
 
 from pydantic import BaseModel, Field
 
-from app.core.workflow.adapters.errors import ExceptionDefineition
+from app.core.workflow.adapters.errors import ExceptionDefinition
 from app.schemas.workflow_schema import (
     EdgeDefinition,
     NodeDefinition,
@@ -40,8 +40,8 @@ class WorkflowParserResult(BaseModel):
     edges: list[EdgeDefinition] = Field(default_factory=list)
     nodes: list[NodeDefinition] = Field(default_factory=list)
     variables: list[VariableDefinition] = Field(default_factory=list)
-    warnings: list[ExceptionDefineition] = Field(default_factory=list)
-    errors: list[ExceptionDefineition] = Field(default_factory=list)
+    warnings: list[ExceptionDefinition] = Field(default_factory=list)
+    errors: list[ExceptionDefinition] = Field(default_factory=list)
 
 
 class WorkflowImportResult(BaseModel):
@@ -51,8 +51,8 @@ class WorkflowImportResult(BaseModel):
     edges: list[EdgeDefinition] = Field(default_factory=list)
     nodes: list[NodeDefinition] = Field(default_factory=list)
     variables: list[VariableDefinition] = Field(default_factory=list)
-    warnings: list[ExceptionDefineition] = Field(default_factory=list)
-    errors: list[ExceptionDefineition] = Field(default_factory=list)
+    warnings: list[ExceptionDefinition] = Field(default_factory=list)
+    errors: list[ExceptionDefinition] = Field(default_factory=list)
 
 
 class BasePlatformAdapter(ABC):

api/app/core/workflow/adapters/dify/converter.py

@@ -9,9 +9,9 @@ from urllib.parse import quote
 
 from app.core.workflow.adapters.base_converter import BaseConverter
 from app.core.workflow.adapters.errors import (
-    UnsupportVariableType,
-    UnknowModelWarning,
-    ExceptionDefineition,
+    UnsupportedVariableType,
+    UnknownModelWarning,
+    ExceptionDefinition,
     ExceptionType
 )
 from app.core.workflow.nodes.assigner.config import AssignmentItem
@@ -54,7 +54,7 @@ from app.core.workflow.nodes.http_request.config import (
     HttpFormData,
     HttpTimeOutConfig,
     HttpRetryConfig,
-    HttpErrorDefaultTamplete,
+    HttpErrorDefaultTemplate,
     HttpErrorHandleConfig
 )
 from app.core.workflow.nodes.if_else.config import ConditionDetail, ConditionBranchConfig
@@ -108,7 +108,7 @@ class DifyConverter(BaseConverter):
         try:
             return config.model_validate(value)
         except Exception as e:
-            self.errors.append(ExceptionDefineition(
+            self.errors.append(ExceptionDefinition(
                 type=ExceptionType.CONFIG,
                 node_id=node_id,
                 node_name=node_name,
@@ -138,7 +138,7 @@ class DifyConverter(BaseConverter):
         var_selector = mapping.get(var_selector, var_selector)
         return var_selector
 
-    def _process_list_variable_litearl(self, variable_selector: list) -> str | None:
+    def _process_list_variable_literal(self, variable_selector: list) -> str | None:
         if not self.process_var_selector(".".join(variable_selector)):
             return None
         return "{{" + self.process_var_selector(".".join(variable_selector)) + "}}"
@@ -269,7 +269,7 @@ class DifyConverter(BaseConverter):
             var_type = self.variable_type_map(var["type"])
             if not var_type:
                 self.errors.append(
-                    UnsupportVariableType(
+                    UnsupportedVariableType(
                         scope=node["id"],
                         name=var["variable"],
                         var_type=var["type"],
@@ -281,7 +281,7 @@ class DifyConverter(BaseConverter):
 
             if var_type in ["file", "array[file]"]:
                 self.errors.append(
-                    ExceptionDefineition(
+                    ExceptionDefinition(
                         type=ExceptionType.VARIABLE,
                         node_id=node["id"],
                         node_name=node_data["title"],
@@ -311,7 +311,7 @@ class DifyConverter(BaseConverter):
     def convert_question_classifier_node_config(self, node: dict) -> dict:
         node_data = node["data"]
         self.warnings.append(
-            UnknowModelWarning(
+            UnknownModelWarning(
                 node_id=node["id"],
                 node_name=node_data["title"],
                 model_name=node_data["model"].get("name")
@@ -327,7 +327,7 @@ class DifyConverter(BaseConverter):
             )
 
         result = QuestionClassifierNodeConfig.model_construct(
-            input_variable=self._process_list_variable_litearl(node_data.get("query_variable_selector")),
+            input_variable=self._process_list_variable_literal(node_data.get("query_variable_selector")),
             user_supplement_prompt=self.trans_variable_format(node_data.get("instructions", "")),
             categories=categories,
         ).model_dump()
@@ -337,13 +337,13 @@ class DifyConverter(BaseConverter):
     def convert_llm_node_config(self, node: dict) -> dict:
         node_data = node["data"]
         self.warnings.append(
-            UnknowModelWarning(
+            UnknownModelWarning(
                 node_id=node["id"],
                 node_name=node_data["title"],
                 model_name=node_data["model"].get("name")
             )
         )
-        context = self._process_list_variable_litearl(node_data["context"]["variable_selector"])
+        context = self._process_list_variable_literal(node_data["context"]["variable_selector"])
         memory = MemoryWindowSetting(
             enable=bool(node_data.get("memory")),
             enable_window=bool(node_data.get("memory", {}).get("window", {}).get("enabled", False)),
@@ -367,7 +367,7 @@ class DifyConverter(BaseConverter):
                 )
             )
         vision = node_data["vision"]["enabled"]
-        vision_input = self._process_list_variable_litearl(
+        vision_input = self._process_list_variable_literal(
             node_data["vision"]["configs"]["variable_selector"]
         ) if vision else None
         result = LLMNodeConfig.model_construct(
@@ -433,7 +433,7 @@ class DifyConverter(BaseConverter):
             conditions.append(
                 LoopConditionDetail.model_construct(
                     operator=self.convert_compare_operator(condition["comparison_operator"]),
-                    left=self._process_list_variable_litearl(condition["variable_selector"]),
+                    left=self._process_list_variable_literal(condition["variable_selector"]),
                     right=self.trans_variable_format(
                         right_value
                     ) if isinstance(right_value, str) and self.is_variable(right_value) else self.convert_variable_type(
@@ -453,7 +453,7 @@ class DifyConverter(BaseConverter):
             right_input_type = variable["value_type"]
             right_value_type = self.variable_type_map(variable["var_type"])
             if right_input_type == ValueInputType.VARIABLE:
-                right_value = self._process_list_variable_litearl(variable.get("value", ""))
+                right_value = self._process_list_variable_literal(variable.get("value", ""))
             else:
                 right_value = self.convert_variable_type(right_value_type, variable.get("value", ""))
             loop_variables.append(
@@ -475,10 +475,10 @@ class DifyConverter(BaseConverter):
     def convert_iteration_node_config(self, node: dict) -> dict:
         node_data = node["data"]
         result = IterationNodeConfig.model_construct(
-            input=self._process_list_variable_litearl(node_data["iterator_selector"]),
+            input=self._process_list_variable_literal(node_data["iterator_selector"]),
             parallel=node_data["is_parallel"],
             parallel_count=node_data["parallel_nums"],
-            output=self._process_list_variable_litearl(node_data["output_selector"]),
+            output=self._process_list_variable_literal(node_data["output_selector"]),
             output_type=self.variable_type_map(node_data.get("output_type")),
             flatten=node_data["flatten_output"],
         ).model_dump()
@@ -494,8 +494,8 @@ class DifyConverter(BaseConverter):
                 continue
             assignments.append(
                 AssignmentItem(
-                    variable_selector=self._process_list_variable_litearl(assignment["variable_selector"]),
-                    value=self._process_list_variable_litearl(
+                    variable_selector=self._process_list_variable_literal(assignment["variable_selector"]),
+                    value=self._process_list_variable_literal(
                         assignment["value"]
                     ) if assignment["input_type"] == ValueInputType.VARIABLE else assignment["value"],
                     operation=self.convert_assignment_operator(assignment["operation"])
@@ -514,7 +514,7 @@ class DifyConverter(BaseConverter):
             input_variables.append(
                 InputVariable.model_construct(
                     name=input_variable["variable"],
-                    variable=self._process_list_variable_litearl(input_variable["value_selector"]),
+                    variable=self._process_list_variable_literal(input_variable["value_selector"]),
                 )
             )
 
@@ -570,7 +570,7 @@ class DifyConverter(BaseConverter):
         else:
             if node_data["body"]["data"]:
                 body_content = (node_data["body"]["data"][0].get("value") or
-                                self._process_list_variable_litearl(node_data["body"]["data"][0].get("file")))
+                                self._process_list_variable_literal(node_data["body"]["data"][0].get("file")))
             else:
                 body_content = ""
 
@@ -585,7 +585,7 @@ class DifyConverter(BaseConverter):
                     self.trans_variable_format(key_value[0])
                 ] = self.trans_variable_format(key_value[1])
             else:
-                self.warnings.append(ExceptionDefineition(
+                self.warnings.append(ExceptionDefinition(
                     type=ExceptionType.CONFIG,
                     node_id=node["id"],
                     node_name=node_data["title"],
@@ -603,7 +603,7 @@ class DifyConverter(BaseConverter):
                     self.trans_variable_format(key_value[0])
                 ] = self.trans_variable_format(key_value[1])
             else:
-                self.warnings.append(ExceptionDefineition(
+                self.warnings.append(ExceptionDefinition(
                     type=ExceptionType.CONFIG,
                     node_id=node["id"],
                     node_name=node_data["title"],
@@ -625,7 +625,7 @@ class DifyConverter(BaseConverter):
                     default_header = var["value"]
                 elif var["key"] == "status_code":
                     default_status_code = var["value"]
-            default_value = HttpErrorDefaultTamplete(
+            default_value = HttpErrorDefaultTemplate(
                 body=default_body,
                 headers=default_header,
                 status_code=default_status_code,
@@ -668,7 +668,7 @@ class DifyConverter(BaseConverter):
         for variable in node_data["variables"]:
             mapping.append(VariablesMappingConfig.model_construct(
                 name=variable["variable"],
-                value=self._process_list_variable_litearl(variable["value_selector"])
+                value=self._process_list_variable_literal(variable["value_selector"])
             ))
         result = JinjaRenderNodeConfig.model_construct(
             template=node_data["template"],
@@ -679,14 +679,14 @@ class DifyConverter(BaseConverter):
 
     def convert_knowledge_node_config(self, node: dict) -> dict:
         node_data = node["data"]
-        self.warnings.append(ExceptionDefineition(
+        self.warnings.append(ExceptionDefinition(
             node_id=node["id"],
             node_name=node_data["title"],
             type=ExceptionType.CONFIG,
             detail=f"Please reconfigure the Knowledge Retrieval node.",
         ))
         result = KnowledgeRetrievalNodeConfig.model_construct(
-            query=self._process_list_variable_litearl(node_data["query_variable_selector"]),
+            query=self._process_list_variable_literal(node_data["query_variable_selector"]),
         ).model_dump()
 
         self.config_validate(node["id"], node["data"]["title"], KnowledgeRetrievalNodeConfig, result)
@@ -695,7 +695,7 @@ class DifyConverter(BaseConverter):
     def convert_parameter_extractor_node_config(self, node: dict) -> dict:
         node_data = node["data"]
         self.warnings.append(
-            UnknowModelWarning(
+            UnknownModelWarning(
                 node_id=node["id"],
                 node_name=node_data["title"],
                 model_name=node_data["model"].get("name")
@@ -712,7 +712,7 @@ class DifyConverter(BaseConverter):
                 )
             )
         result = ParameterExtractorNodeConfig.model_construct(
-            text=self._process_list_variable_litearl(node_data["query"]),
+            text=self._process_list_variable_literal(node_data["query"]),
             params=params,
             prompt=node_data.get("instruction")
         ).model_dump()
@@ -727,14 +727,14 @@ class DifyConverter(BaseConverter):
         group_type = {}
         if not advanced_settings or not advanced_settings["group_enabled"]:
             group_variables = [
-                self._process_list_variable_litearl(variable)
+                self._process_list_variable_literal(variable)
                 for variable in node_data["variables"]
             ]
             group_type["output"] = node_data["output_type"]
         else:
             for group in advanced_settings["groups"]:
                 group_variables[group["group_name"]] = [
-                    self._process_list_variable_litearl(variable)
+                    self._process_list_variable_literal(variable)
                     for variable in group["variables"]
                 ]
                 group_type[group["group_name"]] = group["output_type"]
@@ -751,7 +751,7 @@ class DifyConverter(BaseConverter):
 
     def convert_tool_node_config(self, node: dict) -> dict:
         node_data = node["data"]
-        self.warnings.append(ExceptionDefineition(
+        self.warnings.append(ExceptionDefinition(
             node_id=node["id"],
             node_name=node_data["title"],
             type=ExceptionType.CONFIG,

api/app/core/workflow/adapters/dify/dify_adapter.py

@@ -12,7 +12,7 @@ from app.core.workflow.adapters.base_adapter import (
     WorkflowParserResult
 )
 from app.core.workflow.adapters.dify.converter import DifyConverter
-from app.core.workflow.adapters.errors import ExceptionDefineition, ExceptionType
+from app.core.workflow.adapters.errors import ExceptionDefinition, ExceptionType
 from app.core.workflow.nodes.enums import NodeType
 from app.schemas.workflow_schema import (
     NodeDefinition,
@@ -85,7 +85,7 @@ class DifyAdapter(BasePlatformAdapter, DifyConverter):
         if not all(field in self.config for field in require_fields):
             return False
         if self.config.get("app", {}).get("mode") == "workflow":
-            self.errors.append(ExceptionDefineition(
+            self.errors.append(ExceptionDefinition(
                 type=ExceptionType.PLATFORM,
                 detail="workflow mode is not supported"
             ))
@@ -111,12 +111,12 @@ class DifyAdapter(BasePlatformAdapter, DifyConverter):
             edge = self._convert_edge(edge)
             if edge:
                 self.edges.append(edge)
-        #
+
         for variable in self.config.get("workflow").get("conversation_variables"):
             con_var = self._convert_variable(variable)
             if variable:
                 self.conv_variables.append(con_var)
-        #
+
         # for variables in config.get("workflow").get("environment_variables"):
         #     variable = self._convert_variable(variables)
         #     conv_variables.append(variable)
@@ -152,7 +152,7 @@ class DifyAdapter(BasePlatformAdapter, DifyConverter):
                     "y": node["position"]["y"] + position["y"]
                 }
         self.errors.append(
-            ExceptionDefineition(
+            ExceptionDefinition(
                 type=ExceptionType.NODE,
                 node_id=node_id,
                 detail="parent cycle node not found"
@@ -189,7 +189,7 @@ class DifyAdapter(BasePlatformAdapter, DifyConverter):
             node_data = node["data"]
             converter = self.get_node_convert(node_type)
             if node_type == NodeType.UNKNOWN:
-                self.errors.append(ExceptionDefineition(
+                self.errors.append(ExceptionDefinition(
                     type=ExceptionType.NODE,
                     node_id=node["id"],
                     node_name=node["data"]["title"],
@@ -197,7 +197,7 @@ class DifyAdapter(BasePlatformAdapter, DifyConverter):
                 ))
             return converter(node)
         except Exception as e:
-            self.errors.append(ExceptionDefineition(
+            self.errors.append(ExceptionDefinition(
                 type=ExceptionType.NODE,
                 node_id=node["id"],
                 node_name=node["data"]["title"],
@@ -207,7 +207,6 @@ class DifyAdapter(BasePlatformAdapter, DifyConverter):
 
     def _convert_edge(self, edge: dict[str, Any]) -> EdgeDefinition | None:
         try:
-
             source = edge["source"]
             target = edge["target"]
             label = None
@@ -230,7 +229,7 @@ class DifyAdapter(BasePlatformAdapter, DifyConverter):
                 label=label,
             )
         except Exception as e:
-            self.errors.append(ExceptionDefineition(
+            self.errors.append(ExceptionDefinition(
                 type=ExceptionType.EDGE,
                 detail=f"convert edge error - {e}",
             ))
@@ -246,7 +245,7 @@ class DifyAdapter(BasePlatformAdapter, DifyConverter):
                 description=variable.get("description")
             )
         except Exception as e:
-            self.errors.append(ExceptionDefineition(
+            self.errors.append(ExceptionDefinition(
                 type=ExceptionType.VARIABLE,
                 name=variable.get("name"),
                 detail=f"convert variable error - {e}",

api/app/core/workflow/adapters/errors.py

@@ -18,7 +18,7 @@ class ExceptionType(StrEnum):
     UNKNOWN = "unknown"
 
 
-class ExceptionDefineition(BaseModel):
+class ExceptionDefinition(BaseModel):
     type: ExceptionType
     detail: str
 
@@ -29,7 +29,7 @@ class ExceptionDefineition(BaseModel):
     name: str | None = None
 
 
-class UnknowModelWarning(ExceptionDefineition):
+class UnknownModelWarning(ExceptionDefinition):
     type: ExceptionType = ExceptionType.NODE
 
     def __init__(self, node_id, node_name, model_name):
@@ -40,36 +40,36 @@ class UnknowModelWarning(ExceptionDefineition):
         )
 
 
-class UnknowError(ExceptionDefineition):
+class UnknownError(ExceptionDefinition):
     type: ExceptionType = ExceptionType.UNKNOWN
 
     def __init__(self, detail: str, **kwargs):
         super().__init__(detail=detail, **kwargs)
 
 
-class UnsupportPlatform(ExceptionDefineition):
+class UnsupportedPlatform(ExceptionDefinition):
     type: ExceptionType = ExceptionType.PLATFORM
 
     def __init__(self, platform: str):
-        super().__init__(detail=f"Unsupport platform {platform}")
+        super().__init__(detail=f"Unsupported platform {platform}")
 
 
-class UnsupportVariableType(ExceptionDefineition):
+class UnsupportedVariableType(ExceptionDefinition):
     type: ExceptionType = ExceptionType.VARIABLE
 
     def __init__(self, scope, name, var_type: str, **kwargs):
-        super().__init__(scope=scope, name=name, detail=f"Unsupport variable type：[{var_type}]", **kwargs)
+        super().__init__(scope=scope, name=name, detail=f"Unsupported variable type: [{var_type}]", **kwargs)
 
 
-class InvalidConfiguration(ExceptionDefineition):
+class InvalidConfiguration(ExceptionDefinition):
     type: ExceptionType = ExceptionType.CONFIG
 
     def __init__(self):
         super().__init__(detail="Invalid workflow configuration format")
 
 
-class UnsupportNodeType(ExceptionDefineition):
+class UnsupportedNodeType(ExceptionDefinition):
     type: ExceptionType = ExceptionType.NODE
 
     def __init__(self, node_id: str, node_type: str):
-        super().__init__(node_id=node_id, detail=f"Unsupport node Type {node_type}")
+        super().__init__(node_id=node_id, detail=f"Unsupported node type {node_type}")

api/app/core/workflow/adapters/memory_bear/memory_bear_adapter.py

@@ -11,7 +11,7 @@ from app.core.workflow.adapters.base_adapter import (
     BasePlatformAdapter,
     WorkflowParserResult
 )
-from app.core.workflow.adapters.errors import ExceptionDefineition, ExceptionType, UnsupportNodeType
+from app.core.workflow.adapters.errors import ExceptionDefinition, ExceptionType, UnsupportedNodeType
 from app.core.workflow.adapters.memory_bear.memory_bear_converter import MemoryBearConverter
 from app.core.workflow.nodes.enums import NodeType
 from app.schemas.workflow_schema import ExecutionConfig, NodeDefinition, EdgeDefinition, VariableDefinition
@@ -73,7 +73,7 @@ class MemoryBearAdapter(BasePlatformAdapter, MemoryBearConverter):
         try:
             node_type = self.map_node_type(node["type"])
             if node_type == NodeType.UNKNOWN:
-                self.errors.append(UnsupportNodeType(
+                self.errors.append(UnsupportedNodeType(
                     node_id=node_id,
                     node_type=node["type"]
                 ))
@@ -85,7 +85,7 @@ class MemoryBearAdapter(BasePlatformAdapter, MemoryBearConverter):
 
             return NodeDefinition(**node)
         except Exception as e:
-            self.errors.append(ExceptionDefineition(
+            self.errors.append(ExceptionDefinition(
                 type=ExceptionType.NODE,
                 node_id=node_id,
                 node_name=node_name,
@@ -97,14 +97,14 @@ class MemoryBearAdapter(BasePlatformAdapter, MemoryBearConverter):
     def _convert_edge(self, edge: dict[str, Any], valid_node_ids: set) -> EdgeDefinition | None:
         try:
             if edge.get("source") not in valid_node_ids or edge.get("target") not in valid_node_ids:
-                self.warnings.append(ExceptionDefineition(
+                self.warnings.append(ExceptionDefinition(
                     type=ExceptionType.EDGE,
                     detail=f"edge {edge.get('id')} skipped: source or target node not found"
                 ))
                 return None
             return EdgeDefinition(**edge)
         except Exception as e:
-            self.errors.append(ExceptionDefineition(
+            self.errors.append(ExceptionDefinition(
                 type=ExceptionType.EDGE,
                 detail=f"convert edge error - {e}"
             ))
@@ -115,7 +115,7 @@ class MemoryBearAdapter(BasePlatformAdapter, MemoryBearConverter):
         try:
             return VariableDefinition(**variable)
         except Exception as e:
-            self.warnings.append(ExceptionDefineition(
+            self.warnings.append(ExceptionDefinition(
                 type=ExceptionType.VARIABLE,
                 name=variable.get("name"),
                 detail=f"convert variable error - {e}"

api/app/core/workflow/adapters/memory_bear/memory_bear_converter.py

@@ -1,6 +1,6 @@
 # -*- coding: UTF-8 -*-
 from app.core.workflow.adapters.base_converter import BaseConverter
-from app.core.workflow.adapters.errors import ExceptionDefineition, ExceptionType
+from app.core.workflow.adapters.errors import ExceptionDefinition, ExceptionType
 from app.core.workflow.nodes.base_config import BaseNodeConfig
 from app.core.workflow.nodes.configs import (
     StartNodeConfig,
@@ -65,7 +65,7 @@ class MemoryBearConverter(BaseConverter):
         try:
             return config_cls.model_validate(value)
         except Exception as e:
-            self.errors.append(ExceptionDefineition(
+            self.errors.append(ExceptionDefinition(
                 type=ExceptionType.CONFIG,
                 node_id=node_id,
                 node_name=node_name,

api/app/core/workflow/engine/graph_builder.py

@@ -7,7 +7,7 @@ import re
 import uuid
 from collections import defaultdict
 from functools import lru_cache
-from typing import Any, Iterable
+from typing import Any, Iterable, Callable
 
 from langgraph.checkpoint.memory import InMemorySaver
 from langgraph.graph import START, END
@@ -41,48 +41,31 @@ class GraphBuilder:
             self,
             workflow_config: dict[str, Any],
             stream: bool = False,
-            subgraph: bool = False,
+            cycle: str = '',
             variable_pool: VariablePool | None = None
     ):
         self.workflow_config = workflow_config
 
         self.stream = stream
-        self.subgraph = subgraph
+        self.cycle = cycle
 
         self.start_node_id: str | None = None
 
-        self.node_map = {node["id"]: node for node in self.nodes}
+        self.node_map: dict[str, dict] = {}
         self.end_node_map: dict[str, StreamOutputConfig] = {}
-        self._find_upstream_activation_dep = lru_cache(
-            maxsize=len(self.nodes) * 2
-        )(self._find_upstream_activation_dep)
+        self._find_upstream_activation_dep: Callable = self._find_upstream_activation_dep
         if variable_pool:
             self.variable_pool = variable_pool
         else:
             self.variable_pool = VariablePool()
 
-        self.graph = StateGraph(WorkflowState)
-        self.add_nodes()
-        self.reachable_nodes = WorkflowValidator.get_reachable_nodes(self.start_node_id, self.edges)
-        self.end_nodes = [
-            node
-            for node in self.nodes
-            if node.get("type") == "end" and node.get("id") in self.reachable_nodes
-        ]
-        self.add_edges()
-        # EDGES MUST BE ADDED AFTER NODES ARE ADDED.
-
+        self.graph: StateGraph | None = None
+        self.nodes: list = []
+        self.edges: list = []
+        self.reachable_nodes: set[str] | None = None
+        self.end_nodes: list[dict] = []
         self._reverse_adj: dict[str, list[dict]] = defaultdict(list)
-        self._build_reverse_adj()
-        self._analyze_end_node_output()
-
-    @property
-    def nodes(self) -> list[dict[str, Any]]:
-        return self.workflow_config.get("nodes", [])
-
-    @property
-    def edges(self) -> list[dict[str, Any]]:
-        return self.workflow_config.get("edges", [])
+        self._adj: dict[str, list[str]] = defaultdict(list)
 
     def get_node_type(self, node_id: str) -> str:
         """Retrieve the type of node given its ID.
@@ -108,13 +91,14 @@ class GraphBuilder:
             result[node[0]].append(node[1])
         return result
 
-    def _build_reverse_adj(self):
+    def _build_adj(self):
         for edge in self.edges:
             if edge["source"] not in self.reachable_nodes:
                 continue
             self._reverse_adj[edge.get("target")].append({
                 "id": edge["source"], "branch": edge.get("label")
             })
+            self._adj[edge.get("source")].append(edge["target"])
 
     def _find_upstream_activation_dep(
             self,
@@ -302,22 +286,13 @@ class GraphBuilder:
         """
         for node in self.nodes:
             node_type = node.get("type")
-            if node_type == NodeType.NOTES:
-                continue
             node_id = node.get("id")
-            cycle_node = node.get("cycle")
-            if cycle_node:
-                # Nodes within a loop subgraph are constructed by CycleGraphNode
-                if not self.subgraph:
-                    continue
-
-            # Record start and end node IDs
-            if node_type in [NodeType.START, NodeType.CYCLE_START]:
-                self.start_node_id = node_id
+            if node_id not in self.reachable_nodes:
+                continue
 
             # Create node instance (start and end nodes are also created)
             # NOTE:Loop node creation automatically removes the nodes and edges of the subgraph from the current graph
-            node_instance = NodeFactory.create_node(node, self.workflow_config)
+            node_instance = NodeFactory.create_node(node, self.workflow_config, self._adj[node_id])
 
             if node_type in BRANCH_NODES:
 
@@ -390,6 +365,8 @@ class GraphBuilder:
         for edge in self.edges:
             source = edge.get("source")
             target = edge.get("target")
+            if source not in self.reachable_nodes or target not in self.reachable_nodes:
+                continue
             condition = edge.get("condition")
             edge_type = edge.get("type")
 
@@ -411,11 +388,12 @@ class GraphBuilder:
         # Add conditional edges
         for source_node, branches in conditional_edges.items():
             def make_router(src, branch_list):
-                """reate a router function for each source node that routes to a NOP node for later merging."""
+                """Create a router function for each source node that routes to a NOP node for later merging."""
 
                 def make_branch_node(node_name, targets):
                     def node(s):
-                        # NOTE: NOP NODE MUST NOT MODIFY STATE
+                        # NOTE: NOP NODE USED FOR ROUTING ONLY.
+                        # MUST NOT MUTATE STATE DIRECTLY; ONLY EMIT ACTIVATE SIGNALS.
                         return {
                             "activate": {
                                 node_id: s["activate"][node_name]
@@ -502,14 +480,52 @@ class GraphBuilder:
                 logger.debug(f"Added waiting edge: {sources} -> {target}")
 
         # Connect End nodes to the global END node
-        for end_node in self.end_nodes:
-            end_node_id = end_node.get("id")
-            if end_node_id:
-                self.graph.add_edge(end_node_id, END)
-                logger.debug(f"Added edge: {end_node_id} -> END")
+        for node in self.reachable_nodes:
+            if not self._adj[node]:
+                self.graph.add_edge(node, END)
         return
 
     def build(self) -> CompiledStateGraph:
+        nodes = self.workflow_config.get("nodes", [])
+        edges = self.workflow_config.get("edges", [])
+
+        for node in nodes:
+            if (node.get("cycle") or '') == self.cycle:
+                node_type = node.get("type")
+                if node_type in [NodeType.START, NodeType.CYCLE_START]:
+                    self.start_node_id = node.get("id")
+                elif node_type == NodeType.NOTES:
+                    continue
+                self.nodes.append(node)
+                self.node_map[node.get("id")] = node
+
+        for edge in edges:
+            source_in = edge.get("source") in self.node_map
+            target_in = edge.get("target") in self.node_map
+            if source_in ^ target_in:
+                raise ValueError(
+                    f"Cycle node is connected to external node, "
+                    f"source: {edge.get('source')}, target: {edge.get('target')}"
+                )
+
+            if source_in and target_in:
+                self.edges.append(edge)
+
+        self.reachable_nodes = WorkflowValidator.get_reachable_nodes(self.start_node_id, self.edges)
+        self.end_nodes = [
+            node
+            for node in self.nodes
+            if node.get("type") == "end" and node.get("id") in self.reachable_nodes
+        ]
+        self._build_adj()
+        self._find_upstream_activation_dep: Callable = lru_cache(
+            maxsize=len(self.nodes)*2
+        )(self._find_upstream_activation_dep)
+
+        self.graph = StateGraph(WorkflowState)
+        self.add_nodes()
+        self.add_edges()
+
+        self._analyze_end_node_output()
         checkpointer = InMemorySaver()
-        self.graph = self.graph.compile(checkpointer=checkpointer)
-        return self.graph
+        return self.graph.compile(checkpointer=checkpointer)

api/app/core/workflow/engine/result_builder.py

@@ -2,6 +2,7 @@
 # Author: Eternity
 # @Email: 1533512157@qq.com
 # @Time : 2026/2/10 13:33
+from app.core.workflow.engine.runtime_schema import ExecutionContext
 from app.core.workflow.engine.variable_pool import VariablePool
 
 
@@ -9,6 +10,7 @@ class WorkflowResultBuilder:
     def build_final_output(
             self,
             result: dict,
+            execution_context: ExecutionContext,
             variable_pool: VariablePool,
             elapsed_time: float,
             final_output: str,
@@ -26,6 +28,8 @@ class WorkflowResultBuilder:
                     - "node_outputs" (dict): Outputs of executed nodes.
                     - "messages" (list): Conversation messages exchanged during execution.
                     - "error" (str, optional): Error message if any node failed.
+            execution_context (ExecutionContext): The execution context containing metadata like
+                execution ID, workspace ID, and user ID.)
             variable_pool (VariablePool): Variable Pool
             elapsed_time (float): Total execution time in seconds.
             final_output (Any): The aggregated or final output content of the workflow
@@ -48,18 +52,23 @@ class WorkflowResultBuilder:
         """
         node_outputs = result.get("node_outputs", {})
         token_usage = self.aggregate_token_usage(node_outputs)
-        conversation_id = variable_pool.get_value("sys.conversation_id")
+        conversation_vars = {}
+        sys_vars = {}
+
+        if variable_pool:
+            conversation_vars = variable_pool.get_all_conversation_vars()
+            sys_vars = variable_pool.get_all_system_vars()
 
         return {
             "status": "completed" if success else "failed",
             "output": final_output,
             "variables": {
-                "conv": variable_pool.get_all_conversation_vars(),
-                "sys": variable_pool.get_all_system_vars()
+                "conv": conversation_vars,
+                "sys": sys_vars
             },
             "node_outputs": node_outputs,
             "messages": result.get("messages", []),
-            "conversation_id": conversation_id,
+            "conversation_id": execution_context.conversation_id,
             "elapsed_time": elapsed_time,
             "token_usage": token_usage,
             "error": result.get("error"),

api/app/core/workflow/engine/runtime_schema.py

@@ -12,14 +12,29 @@ class ExecutionContext(BaseModel):
     execution_id: str
     workspace_id: str
     user_id: str
+    conversation_id: str
+    memory_storage_type: str
+    user_rag_memory_id: str
     checkpoint_config: RunnableConfig
 
     @classmethod
-    def create(cls, execution_id: str, workspace_id: str, user_id: str):
+    def create(
+            cls,
+            execution_id: str,
+            workspace_id: str,
+            user_id: str,
+            conversation_id: str,
+            memory_storage_type: str,
+            user_rag_memory_id: str
+    ):
         return cls(
             execution_id=execution_id,
             workspace_id=workspace_id,
             user_id=user_id,
+            conversation_id=conversation_id,
+            memory_storage_type=memory_storage_type,
+            user_rag_memory_id=user_rag_memory_id,
+
             checkpoint_config=RunnableConfig(
                 configurable={
                     "thread_id": uuid.uuid4(),

api/app/core/workflow/engine/state_manager.py

@@ -9,10 +9,10 @@ from app.core.workflow.nodes.enums import NodeType
 
 
 def merge_activate_state(x, y):
-    return {
-        k: x.get(k, False) or y.get(k, False)
-        for k in set(x) | set(y)
-    }
+    merged = dict(x)
+    for k, v in y.items():
+        merged[k] = merged.get(k, False) or v
+    return merged
 
 
 def merge_looping_state(x, y):
@@ -33,6 +33,8 @@ class WorkflowState(dict):
         "workspace_id",
         "user_id",
         "activate",
+        "memory_storage_type",
+        "user_rag_memory_id"
     })
     __optional_keys__ = frozenset({
         "error",
@@ -62,6 +64,9 @@ class WorkflowState(dict):
     # node activate status
     activate: Annotated[dict[str, bool], merge_activate_state]
 
+    memory_storage_type: str
+    user_rag_memory_id: str
+
 
 class WorkflowStateManager:
     def create_initial_state(
@@ -85,7 +90,9 @@ class WorkflowStateManager:
             looping=0,
             activate={
                 start_node_id: True
-            }
+            },
+            memory_storage_type=execution_context.memory_storage_type,
+            user_rag_memory_id=execution_context.user_rag_memory_id
         )
 
     @staticmethod

api/app/core/workflow/engine/stream_output_coordinator.py

@@ -3,7 +3,7 @@
 # @Email: 1533512157@qq.com
 # @Time : 2026/2/9 15:11
 import re
-from queue import Queue
+from collections import deque
 from typing import AsyncGenerator
 
 from pydantic import BaseModel, Field, PrivateAttr
@@ -256,7 +256,7 @@ class StreamOutputCoordinator:
     def __init__(self):
         self.end_outputs: dict[str, StreamOutputConfig] = {}
         self.activate_end: str | None = None
-        self.output_queue: Queue = Queue()
+        self.output_queue: deque[str] = deque()
         self.processed_outputs = []
 
     def initialize_end_outputs(
@@ -266,7 +266,7 @@ class StreamOutputCoordinator:
         self.end_outputs = end_node_map
         self.processed_outputs = []
         self.activate_end = None
-        self.output_queue = Queue()
+        self.output_queue = deque()
 
     @property
     def current_activate_end_info(self):
@@ -296,13 +296,13 @@ class StreamOutputCoordinator:
             scope (str): The node ID or scope that has completed execution.
             status (str | None): Optional status of the node (used for branch/control nodes).
         """
-        for node in self.end_outputs.keys():
+        for node in self.end_outputs:
             self.end_outputs[node].update_activate(scope, status)
             if self.end_outputs[node].activate and node not in self.processed_outputs:
-                self.output_queue.put(node)
+                self.output_queue.append(node)
                 self.processed_outputs.append(node)
-        if self.activate_end is None and not self.output_queue.empty():
-            self.activate_end = self.output_queue.get_nowait()
+        if self.activate_end is None and self.output_queue:
+            self.activate_end = self.output_queue.popleft()
 
     async def emit_activate_chunk(
             self,
@@ -414,8 +414,8 @@ class StreamOutputCoordinator:
                 async for msg_event in self.emit_activate_chunk(variable_pool, force=True):
                     yield msg_event
 
-                if not self.output_queue.empty():
-                    self.activate_end = self.output_queue.get_nowait()
+                if self.output_queue:
+                    self.activate_end = self.output_queue.popleft()
                 # Move to next active End node if current one is done
                 if not self.activate_end and self.end_outputs:
                     self.activate_end = list(self.end_outputs.keys())[0]

api/app/core/workflow/engine/variable_pool.py

@@ -13,10 +13,55 @@ from pydantic import BaseModel
 
 from app.core.workflow.engine.runtime_schema import ExecutionContext
 from app.core.workflow.variable.base_variable import VariableType, DEFAULT_VALUE
-from app.core.workflow.variable.variable_objects import T, create_variable_instance
+from app.core.workflow.variable.variable_objects import T, create_variable_instance, ArrayVariable, FileVariable
 
 logger = logging.getLogger(__name__)
 
+VARIABLE_PATTERN = re.compile(r"\{\{\s*(.*?)\s*}}")
+
+
+class LazyVariableDict:
+    def __init__(self, source, literal):
+        self._source: dict[str, VariableStruct[Any]] = source
+        self._literal: bool = literal
+        self._cache = {}
+
+    def keys(self):
+        return self._source.keys()
+
+    def _resolve(self, key):
+        if key in self._cache:
+            return self._cache[key]
+        var_struct = self._source.get(key)
+        if var_struct is None:
+            raise KeyError(key)
+        value = var_struct.instance.to_literal() if self._literal else var_struct.instance.get_value()
+        self._cache[key] = value
+        return value
+
+    def get(self, key, default=None):
+        try:
+            return self._resolve(key)
+        except KeyError:
+            return default
+
+    def __getitem__(self, key):
+        return self._resolve(key)
+
+    def __getattr__(self, key):
+        if key.startswith('_'):
+            raise AttributeError(f"'{self.__class__.__name__}' object has no attribute '{key}'")
+        return self._resolve(key)
+
+    def __contains__(self, key):
+        return key in self._source
+
+    def __iter__(self):
+        return iter(self._source)
+
+    def __len__(self):
+        return len(self._source)
+
 
 class VariableSelector:
     """变量选择器
@@ -117,8 +162,7 @@ class VariablePool:
 
     @staticmethod
     def transform_selector(selector):
-        pattern = r"\{\{\s*(.*?)\s*\}\}"
-        variable_literal = re.sub(pattern, r"\1", selector).strip()
+        variable_literal = VARIABLE_PATTERN.sub(r"\1", selector).strip()
         selector = VariableSelector.from_string(variable_literal).path
         if len(selector) != 2:
             raise ValueError(f"Selector not valid - {selector}")
@@ -303,6 +347,16 @@ class VariablePool:
         """
         return self._get_variable_struct(selector) is not None
 
+    def lazy_namespace(self, namespace: str, literal: bool = False) -> LazyVariableDict:
+        return LazyVariableDict(self.variables.get(namespace, {}), literal)
+
+    def lazy_all_node_outputs(self, literal: bool = False) -> dict[str, LazyVariableDict]:
+        return {
+            ns: LazyVariableDict(vars_dict, literal)
+            for ns, vars_dict in self.variables.items()
+            if ns not in ("sys", "conv")
+        }
+
     def get_all_system_vars(self, literal=False) -> dict[str, Any]:
         """获取所有系统变量
         
@@ -373,6 +427,16 @@ class VariablePool:
     def copy(self, pool: 'VariablePool'):
         self.variables = deepcopy(pool.variables)
 
+    def is_file_variable(self, selector):
+        variable_struct = self.get_instance(selector, default=None, strict=False)
+        if variable_struct is None:
+            return False
+        if isinstance(variable_struct, FileVariable):
+            return True
+        elif isinstance(variable_struct, ArrayVariable) and variable_struct.child_type == FileVariable:
+            return True
+        return False
+
     def to_dict(self) -> dict[str, Any]:
         """导出为字典
         
@@ -469,5 +533,3 @@ class VariablePoolInitializer:
                 var_type=var_type,
                 mut=False
             )
-
-

api/app/core/workflow/executor.py

@@ -3,6 +3,7 @@
 # @Email: 1533512157@qq.com
 # @Time : 2026/2/9 13:51
 import datetime
+import time
 import logging
 from typing import Any
 
@@ -82,13 +83,15 @@ class WorkflowExecutor:
             CompiledStateGraph: The compiled and ready-to-run state graph.
         """
         logger.info(f"Starting workflow graph build: execution_id={self.execution_context.execution_id}")
+        start_time = time.time()
         builder = GraphBuilder(
             self.workflow_config,
             stream=stream,
         )
+
+        self.graph = builder.build()
         self.start_node_id = builder.start_node_id
         self.variable_pool = builder.variable_pool
-        self.graph = builder.build()
 
         self.stream_coordinator.initialize_end_outputs(builder.end_node_map)
         self.event_handler = EventStreamHandler(
@@ -96,7 +99,8 @@ class WorkflowExecutor:
             variable_pool=self.variable_pool,
             execution_id=self.execution_context.execution_id
         )
-        logger.info(f"Workflow graph build completed: execution_id={self.execution_context.execution_id}")
+        logger.info(f"Workflow graph build completed: execution_id={self.execution_context.execution_id}, "
+                    f"cost: {time.time() - start_time:.4f}s")
 
         return self.graph
 
@@ -134,94 +138,12 @@ class WorkflowExecutor:
                 return event.get("data")
         return self.result_builder.build_final_output(
             {"error": "Workflow execution did not end as expected"},
+            self.execution_context,
             self.variable_pool,
             (datetime.datetime.now() - start).total_seconds(),
             "",
             success=False
         )
-        # logger.info(f"Starting workflow execution: execution_id={self.execution_context.execution_id}")
-        #
-        # start_time = datetime.datetime.now()
-        #
-        # # Execute the workflow
-        # try:
-        #     # Build the workflow graph
-        #     graph = self.build_graph()
-        #
-        #     # Initialize the variable pool with input data
-        #     await self.variable_initializer.initialize(
-        #         variable_pool=self.variable_pool,
-        #         input_data=input_data,
-        #         execution_context=self.execution_context
-        #     )
-        #     initial_state = self.state_manager.create_initial_state(
-        #         workflow_config=self.workflow_config,
-        #         input_data=input_data,
-        #         execution_context=self.execution_context,
-        #         start_node_id=self.start_node_id
-        #     )
-        #
-        #     result = await graph.ainvoke(initial_state, config=self.execution_context.checkpoint_config)
-        #
-        #     # Aggregate output from all End nodes
-        #     full_content = ''
-        #     for end_id in self.stream_coordinator.end_outputs.keys():
-        #         full_content += self.variable_pool.get_value(f"{end_id}.output", default="", strict=False)
-        #
-        #     # Append messages for user and assistant
-        #     if input_data.get("files"):
-        #         result["messages"].extend(
-        #             [
-        #                 {
-        #                     "role": "user",
-        #                     "content": input_data.get("message", '')
-        #                 },
-        #                 {
-        #                     "role": "user",
-        #                     "content": input_data.get("files")
-        #                 },
-        #                 {
-        #                     "role": "assistant",
-        #                     "content": full_content
-        #                 }
-        #             ]
-        #         )
-        #     else:
-        #         result["messages"].extend(
-        #             [
-        #                 {
-        #                     "role": "user",
-        #                     "content": input_data.get("message", '')
-        #                 },
-        #                 {
-        #                     "role": "assistant",
-        #                     "content": full_content
-        #                 }
-        #             ]
-        #         )
-        #     # Calculate elapsed time
-        #     end_time = datetime.datetime.now()
-        #     elapsed_time = (end_time - start_time).total_seconds()
-        #
-        #     logger.info(
-        #         f"Workflow execution completed: execution_id={self.execution_context.execution_id}, elapsed_time={elapsed_time:.2f}ms")
-        #
-        #     return self.result_builder.build_final_output(result, self.variable_pool, elapsed_time, full_content)
-        #
-        # except Exception as e:
-        #     end_time = datetime.datetime.now()
-        #     elapsed_time = (end_time - start_time).total_seconds()
-        #
-        #     logger.error(f"Workflow execution failed: execution_id={self.execution_context.execution_id}, error={e}",
-        #                  exc_info=True)
-        #     return {
-        #         "status": "failed",
-        #         "error": str(e),
-        #         "output": None,
-        #         "node_outputs": {},
-        #         "elapsed_time": elapsed_time,
-        #         "token_usage": None
-        #     }
 
     async def execute_stream(
             self,
@@ -255,7 +177,7 @@ class WorkflowExecutor:
             "data": {
                 "execution_id": self.execution_context.execution_id,
                 "workspace_id": self.execution_context.workspace_id,
-                "conversation_id": input_data.get("conversation_id"),
+                "conversation_id": self.execution_context.conversation_id,
                 "timestamp": int(start_time.timestamp() * 1000)
             }
         }
@@ -376,6 +298,7 @@ class WorkflowExecutor:
                 "event": "workflow_end",
                 "data": self.result_builder.build_final_output(
                     result,
+                    self.execution_context,
                     self.variable_pool,
                     elapsed_time,
                     full_content,
@@ -396,6 +319,7 @@ class WorkflowExecutor:
                 "event": "workflow_end",
                 "data": self.result_builder.build_final_output(
                     result,
+                    self.execution_context,
                     self.variable_pool,
                     elapsed_time,
                     full_content,
@@ -409,7 +333,9 @@ async def execute_workflow(
         input_data: dict[str, Any],
         execution_id: str,
         workspace_id: str,
-        user_id: str
+        user_id: str,
+        memory_storage_type: str,
+        user_rag_memory_id: str
 ) -> dict[str, Any]:
     """
     Execute a workflow (convenience function, non-streaming).
@@ -420,6 +346,8 @@ async def execute_workflow(
         execution_id (str): Execution ID.
         workspace_id (str): Workspace ID.
         user_id (str): User ID.
+        user_rag_memory_id: rag knowledge db id
+        memory_storage_type: neo4j / rag
 
     Returns:
         dict: Workflow execution result.
@@ -427,7 +355,10 @@ async def execute_workflow(
     execution_context = ExecutionContext.create(
         execution_id=execution_id,
         workspace_id=workspace_id,
-        user_id=user_id
+        user_id=user_id,
+        conversation_id=input_data.get("conversation_id"),
+        memory_storage_type=memory_storage_type,
+        user_rag_memory_id=user_rag_memory_id
     )
     executor = WorkflowExecutor(
         workflow_config=workflow_config,
@@ -441,7 +372,9 @@ async def execute_workflow_stream(
         input_data: dict[str, Any],
         execution_id: str,
         workspace_id: str,
-        user_id: str
+        user_id: str,
+        memory_storage_type: str,
+        user_rag_memory_id: str
 ):
     """
     Execute a workflow in streaming mode (convenience function).
@@ -452,6 +385,8 @@ async def execute_workflow_stream(
         execution_id (str): Execution ID.
         workspace_id (str): Workspace ID.
         user_id (str): User ID.
+        user_rag_memory_id: rag knowledge db id
+        memory_storage_type: neo4j / rag
 
     Yields:
         dict: Streaming workflow events, e.g. node start, node end, chunk messages, workflow end.
@@ -459,7 +394,10 @@ async def execute_workflow_stream(
     execution_context = ExecutionContext.create(
         execution_id=execution_id,
         workspace_id=workspace_id,
-        user_id=user_id
+        user_id=user_id,
+        memory_storage_type=memory_storage_type,
+        conversation_id=input_data.get("conversation_id"),
+        user_rag_memory_id=user_rag_memory_id
     )
     executor = WorkflowExecutor(
         workflow_config=workflow_config,

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

@@ -64,9 +64,7 @@ class AgentNode(BaseNode):
         
         if not release:
             raise ValueError(f"Agent 不存在: {agent_id}")
-        
 
-        
         return release, message
     
     async def execute(self, state: WorkflowState, variable_pool: VariablePool) -> dict[str, Any]:

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

@@ -14,8 +14,8 @@ logger = logging.getLogger(__name__)
 
 
 class AssignerNode(BaseNode):
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
         self.variable_updater = True
         self.typed_config: AssignerNodeConfig | None = None
 

api/app/core/workflow/nodes/base_node.py

@@ -28,7 +28,7 @@ class BaseNode(ABC):
     All node types should inherit from this class and implement the `execute` method.
     """
 
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
         """Initialize the node.
 
         Args:
@@ -41,6 +41,7 @@ class BaseNode(ABC):
         self.node_type = node_config["type"]
         self.cycle = node_config.get("cycle")
         self.node_name = node_config.get("name", self.node_id)
+        self.down_stream_nodes = down_stream_nodes
         # 使用 or 运算符处理 None 值
         self.config = node_config.get("config") or {}
         self.error_handling = node_config.get("error_handling") or {}
@@ -93,18 +94,16 @@ class BaseNode(ABC):
             dict: A dict with a single key 'activate', mapping node IDs to
                   their activation status (True/False).
         """
-        edges = self.workflow_config.get("edges")
-        under_stream_nodes = [
-            edge.get("target")
-            for edge in edges
-            if edge.get("source") == self.node_id and self.node_type not in BRANCH_NODES
-        ]
-        return {
-            "activate": {
-                            node_id: self.check_activate(state)
-                            for node_id in under_stream_nodes
-                        } | {self.node_id: self.check_activate(state)}
-        }
+        activate_flag = self.check_activate(state)
+
+        if self.node_type not in BRANCH_NODES:
+            activate = {node_id: activate_flag for node_id in self.down_stream_nodes}
+        else:
+            activate = {}
+
+        activate[self.node_id] = activate_flag
+
+        return {"activate": activate}
 
     @abstractmethod
     async def execute(self, state: WorkflowState, variable_pool: VariablePool) -> Any:
@@ -315,8 +314,8 @@ class BaseNode(ABC):
 
             elapsed_time = (time.time() - start_time) * 1000
 
-            logger.info(f"Node {self.node_id} streaming execution finished, "
-                        f"time elapsed: {elapsed_time:.2f}ms, chunks: {chunk_count}")
+            logger.debug(f"Node {self.node_id} streaming execution finished, "
+                         f"time elapsed: {elapsed_time:.2f}ms, chunks: {chunk_count}")
 
             # Extract processed output (call subclass's _extract_output)
             extracted_output = self._extract_output(final_result)
@@ -428,8 +427,8 @@ class BaseNode(ABC):
             when an error edge exists. If no error edge exists, this method
             raises an exception to stop the workflow.
         """
-        # Check if the node has an error edge defined
-        error_edge = self._find_error_edge()
+        # # Check if the node has an error edge defined
+        # error_edge = self._find_error_edge()
 
         # Extract input data (for logging or audit purposes)
         input_data = self._extract_input(state, variable_pool)
@@ -447,27 +446,26 @@ class BaseNode(ABC):
             "error": error_message
         }
 
-        if error_edge:
-            # If an error edge exists, log a warning and continue to error node
-            logger.warning(
-                f"Node {self.node_id} execution failed, redirecting to error node: {error_edge['target']}"
-            )
-            return {
-                "node_outputs": {
-                    self.node_id: node_output
-                },
-                "error": error_message,
-                "error_node": self.node_id
-            }
-        else:
-            # If no error edge, send the error via stream writer and stop the workflow
-            writer = get_stream_writer()
-            writer({
-                "type": "node_error",
-                **node_output
-            })
-            logger.error(f"Node {self.node_id} execution failed, stopping workflow: {error_message}")
-            raise Exception(f"Node {self.node_id} execution failed: {error_message}")
+        # if error_edge:
+        #     # If an error edge exists, log a warning and continue to error node
+        #     logger.warning(
+        #         f"Node {self.node_id} execution failed, redirecting to error node: {error_edge['target']}"
+        #     )
+        #     return {
+        #         "node_outputs": {
+        #             self.node_id: node_output
+        #         },
+        #         "error": error_message,
+        #         "error_node": self.node_id
+        #     }
+        # else:
+        writer = get_stream_writer()
+        writer({
+            "type": "node_error",
+            **node_output
+        })
+        logger.error(f"Node {self.node_id} execution failed, stopping workflow: {error_message}")
+        raise Exception(f"Node {self.node_id} execution failed: {error_message}")
 
     def _extract_input(self, state: WorkflowState, variable_pool: VariablePool) -> dict[str, Any]:
         """Extracts the input data for this node (used for logging or audit).
@@ -554,9 +552,9 @@ class BaseNode(ABC):
 
         return render_template(
             template=template,
-            conv_vars=variable_pool.get_all_conversation_vars(literal=True),
-            node_outputs=variable_pool.get_all_node_outputs(literal=True),
-            system_vars=variable_pool.get_all_system_vars(literal=True),
+            conv_vars=variable_pool.lazy_namespace("conv", literal=True),
+            node_outputs=variable_pool.lazy_all_node_outputs(literal=True),
+            system_vars=variable_pool.lazy_namespace("sys", literal=True),
             strict=strict
         )
 
@@ -581,9 +579,9 @@ class BaseNode(ABC):
 
         return evaluate_condition(
             expression=expression,
-            conv_var=variable_pool.get_all_conversation_vars(),
-            node_outputs=variable_pool.get_all_node_outputs(),
-            system_vars=variable_pool.get_all_system_vars()
+            conv_var=variable_pool.lazy_namespace("conv"),
+            node_outputs=variable_pool.lazy_all_node_outputs(),
+            system_vars=variable_pool.lazy_namespace("sys")
         )
 
     @staticmethod
@@ -623,7 +621,6 @@ class BaseNode(ABC):
     async def process_message(
             api_config: ModelInfo,
             content: str | dict | FileObject,
-            end_user_id: str,
             enable_file=False
     ) -> list | str | None:
         provider = api_config.provider
@@ -642,10 +639,10 @@ class BaseNode(ABC):
             return content
 
         elif isinstance(content, FileObject):
-            if content.content_cache.get(provider):
-                return content.content_cache[provider]
+            if content.content_cache.get(f"{provider}_{api_config.is_omni}"):
+                return content.content_cache[f"{provider}_{api_config.is_omni}"]
             with get_db_read() as db:
-                multimodel_service = MultimodalService(db, api_config=api_config)
+                multimodal_service = MultimodalService(db, api_config=api_config)
                 file_obj = FileInput(
                     type=content.type,
                     url=content.url,
@@ -654,16 +651,15 @@ class BaseNode(ABC):
                     upload_file_id=uuid.UUID(content.file_id) if content.file_id else None,
                 )
                 file_obj.set_content(content.get_content())
-                message = await multimodel_service.process_files(
-                    end_user_id,
+                message = await multimodal_service.process_files(
                     [file_obj],
                 )
                 content.set_content(file_obj.get_content())
                 if message:
-                    content.content_cache[provider] = message
+                    content.content_cache[f"{provider}_{api_config.is_omni}"] = message
                     return message
                 return None
-        raise TypeError(f'Unexpect input value type - {type(content)}')
+        raise TypeError(f'Unexpected input value type - {type(content)}')
 
     @staticmethod
     def process_model_output(content) -> str:

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

@@ -51,8 +51,8 @@ console.log(result)
 
 
 class CodeNode(BaseNode):
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
         self.typed_config: CodeNodeConfig | None = None
 
     def _output_types(self) -> dict[str, VariableType]:

api/app/core/workflow/nodes/cycle_graph/loop.py

@@ -11,7 +11,6 @@ from app.core.workflow.engine.variable_pool import VariablePool
 from app.core.workflow.nodes.cycle_graph import LoopNodeConfig
 from app.core.workflow.nodes.enums import ValueInputType, ComparisonOperator, LogicOperator, NodeType
 from app.core.workflow.nodes.operators import TypeTransformer, ConditionExpressionResolver, CompareOperatorInstance
-from app.core.workflow.utils.expression_evaluator import evaluate_expression
 
 logger = logging.getLogger(__name__)
 
@@ -85,12 +84,7 @@ class LoopRuntime:
 
         for variable in self.typed_config.cycle_vars:
             if variable.input_type == ValueInputType.VARIABLE:
-                value = evaluate_expression(
-                    expression=variable.value,
-                    conv_var=self.variable_pool.get_all_conversation_vars(),
-                    node_outputs=self.variable_pool.get_all_node_outputs(),
-                    system_vars=self.variable_pool.get_all_system_vars(),
-                )
+                value = self.variable_pool.get_value(variable.value)
             else:
                 value = TypeTransformer.transform(variable.value, variable.type)
             await self.child_variable_pool.new(self.node_id, variable.name, value, variable.type, mut=True)
@@ -98,12 +92,7 @@ class LoopRuntime:
             **self.state
         )
         loopstate["node_outputs"][self.node_id] = {
-            variable.name: evaluate_expression(
-                expression=variable.value,
-                conv_var=self.variable_pool.get_all_conversation_vars(),
-                node_outputs=self.variable_pool.get_all_node_outputs(),
-                system_vars=self.variable_pool.get_all_system_vars(),
-            )
+            variable.name: self.variable_pool.get_value(variable.value)
             if variable.input_type == ValueInputType.VARIABLE
             else TypeTransformer.transform(variable.value, variable.type)
             for variable in self.typed_config.cycle_vars

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

@@ -30,17 +30,13 @@ class CycleGraphNode(BaseNode):
     It acts as a container and execution controller for a subgraph.
     """
 
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
-
-        self.cycle_nodes = list()  # Nodes belonging to this cycle
-        self.cycle_edges = list()  # Edges connecting nodes within the cycle
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
+        self.cycle_nodes, self.cycle_edges = self.pure_cycle_graph()
         self.start_node_id = None  # ID of the start node within the cycle
 
         self.graph: StateGraph | CompiledStateGraph | None = None
         self.child_variable_pool: VariablePool | None = None
-        self.build_graph()
-        self.iteration_flag = True
 
     def _output_types(self) -> dict[str, VariableType]:
         outputs = {"__child_state": VariableType.ARRAY_OBJECT}
@@ -119,11 +115,11 @@ class CycleGraphNode(BaseNode):
             else:
                 remain_edges.append(edge)
 
-        # Update workflow_config by removing cycle nodes and internal edges
-        self.workflow_config["nodes"] = [
-            node for node in nodes if node.get("cycle") != self.node_id
-        ]
-        self.workflow_config["edges"] = remain_edges
+        # # Update workflow_config by removing cycle nodes and internal edges
+        # self.workflow_config["nodes"] = [
+        #     node for node in nodes if node.get("cycle") != self.node_id
+        # ]
+        # self.workflow_config["edges"] = remain_edges
 
         return cycle_nodes, cycle_edges
 
@@ -137,18 +133,18 @@ class CycleGraphNode(BaseNode):
         3. Compile the graph for runtime execution
         """
         from app.core.workflow.engine.graph_builder import GraphBuilder
-        self.cycle_nodes, self.cycle_edges = self.pure_cycle_graph()
+
         self.child_variable_pool = VariablePool()
         builder = GraphBuilder(
             {
                 "nodes": self.cycle_nodes,
                 "edges": self.cycle_edges,
             },
-            subgraph=True,
-            variable_pool=self.child_variable_pool
+            variable_pool=self.child_variable_pool,
+            cycle=self.node_id
         )
-        self.start_node_id = builder.start_node_id
         self.graph = builder.build()
+        self.start_node_id = builder.start_node_id
         self.child_variable_pool = builder.variable_pool
 
     async def execute(self, state: WorkflowState, variable_pool: VariablePool) -> Any:
@@ -169,6 +165,7 @@ class CycleGraphNode(BaseNode):
         Raises:
             RuntimeError: If the node type is unsupported.
         """
+        self.build_graph()
         if self.node_type == NodeType.LOOP:
             return await LoopRuntime(
                 start_id=self.start_node_id,
@@ -194,6 +191,7 @@ class CycleGraphNode(BaseNode):
         raise RuntimeError("Unknown cycle node type")
 
     async def execute_stream(self, state: WorkflowState, variable_pool: VariablePool):
+        self.build_graph()
         if self.node_type == NodeType.LOOP:
             yield {
                 "__final__": True,

api/app/core/workflow/nodes/document_extractor/__init__.py

@@ -0,0 +1,4 @@
+from .config import DocExtractorNodeConfig
+from .node import DocExtractorNode
+
+__all__ = ["DocExtractorNode", "DocExtractorNodeConfig"]

api/app/core/workflow/nodes/document_extractor/config.py

@@ -0,0 +1,18 @@
+from pydantic import Field
+from app.core.workflow.nodes.base_config import BaseNodeConfig
+
+
+class DocExtractorNodeConfig(BaseNodeConfig):
+    file_selector: str = Field(
+        ...,
+        description="File variable selector, e.g. {{ sys.files }} or {{ node_id.file }}"
+    )
+
+    class Config:
+        json_schema_extra = {
+            "examples": [
+                {
+                    "file_selector": "{{ sys.files }}"
+                }
+            ]
+        }

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

@@ -0,0 +1,103 @@
+import logging
+from typing import Any
+
+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
+from app.core.workflow.nodes.document_extractor.config import DocExtractorNodeConfig
+from app.core.workflow.variable.base_variable import VariableType, FileObject
+from app.db import get_db_read
+from app.schemas.app_schema import FileInput, FileType, TransferMethod
+
+logger = logging.getLogger(__name__)
+
+
+def _file_object_to_file_input(f: FileObject) -> FileInput:
+    """Convert workflow FileObject to multimodal FileInput."""
+    return FileInput(
+        type=FileType.DOCUMENT,
+        transfer_method=TransferMethod(f.transfer_method),
+        url=f.url or None,
+        upload_file_id=f.file_id or None,
+        file_type=f.origin_file_type or "",
+    )
+
+
+def _normalise_files(val: Any) -> list[FileObject]:
+    if isinstance(val, FileObject):
+        return [val]
+    if isinstance(val, dict) and val.get("is_file"):
+        return [FileObject(**val)]
+    if isinstance(val, list):
+        result: list[FileObject] = []
+        for item in val:
+            if isinstance(item, FileObject):
+                result.append(item)
+            elif isinstance(item, dict) and item.get("is_file"):
+                result.append(FileObject(**item))
+            else:
+                logger.warning("Ignoring non-file entry in file list for document extractor: %r", item)
+        return result
+    return []
+
+
+class DocExtractorNode(BaseNode):
+    """Document Extractor Node.
+
+    Reads one or more file variables and extracts their text content
+    by delegating to MultimodalService._extract_document_text.
+
+    Outputs:
+        text   (string)        – full concatenated text of all input files
+        chunks (array[string]) – per-file extracted text
+    """
+
+    def _output_types(self) -> dict[str, VariableType]:
+        return {
+            "text": VariableType.STRING,
+            "chunks": VariableType.ARRAY_STRING,
+        }
+
+    def _extract_output(self, business_result: Any) -> Any:
+        return business_result
+
+    def _extract_input(self, state: WorkflowState, variable_pool: VariablePool) -> dict[str, Any]:
+        return {"file_selector": self.config.get("file_selector")}
+
+    async def execute(self, state: WorkflowState, variable_pool: VariablePool) -> Any:
+        config = DocExtractorNodeConfig(**self.config)
+
+        raw_val = self.get_variable(config.file_selector, variable_pool, strict=False)
+        if raw_val is None:
+            logger.warning(f"Node {self.node_id}: file variable '{config.file_selector}' is empty")
+            return {"text": "", "chunks": []}
+
+        files = _normalise_files(raw_val)
+        if not files:
+            return {"text": "", "chunks": []}
+
+        chunks: list[str] = []
+        with get_db_read() as db:
+            from app.services.multimodal_service import MultimodalService
+            svc = MultimodalService(db)
+            for f in files:
+                try:
+                    file_input = _file_object_to_file_input(f)
+                    # Ensure URL is populated for local files
+                    if not file_input.url:
+                        file_input.url = await svc.get_file_url(file_input)
+                    # Reuse cached bytes if already fetched
+                    if f.get_content():
+                        file_input.set_content(f.get_content())
+                    text = await svc.extract_document_text(file_input)
+                    chunks.append(text)
+                except Exception as e:
+                    logger.error(
+                        f"Node {self.node_id}: failed to extract file url={f.url} file_id={f.file_id}: {e}",
+                        exc_info=True,
+                    )
+                    chunks.append("")
+
+        full_text = "\n\n".join(c for c in chunks if c)
+        logger.info(f"Node {self.node_id}: extracted {len(files)} file(s), total chars={len(full_text)}")
+        return {"text": full_text, "chunks": chunks}

api/app/core/workflow/nodes/end/config.py

@@ -1,9 +1,7 @@
 """End 节点配置"""
-
 from pydantic import Field
 
-from app.core.workflow.nodes.base_config import BaseNodeConfig, VariableDefinition
-from app.core.workflow.variable.base_variable import VariableType
+from app.core.workflow.nodes.base_config import BaseNodeConfig
 
 
 class EndNodeConfig(BaseNodeConfig):

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

@@ -36,8 +36,6 @@ class EndNode(BaseNode):
         Returns:
             最终输出字符串
         """
-        logger.info(f"节点 {self.node_id} (End) 开始执行")
-
         # 获取配置的输出模板
         output_template = self.config.get("output")
 
@@ -46,11 +44,4 @@ class EndNode(BaseNode):
             output = self._render_template(output_template, variable_pool, strict=False)
         else:
             output = ""
-
-        # 统计信息（用于日志）
-        node_outputs = state.get("node_outputs", {})
-        total_nodes = len(node_outputs)
-
-        logger.info(f"节点 {self.node_id} (End) 执行完成，共执行 {total_nodes} 个节点")
-
         return output

api/app/core/workflow/nodes/enums.py

@@ -23,12 +23,13 @@ class NodeType(StrEnum):
     BREAK = "break"
     MEMORY_READ = "memory-read"
     MEMORY_WRITE = "memory-write"
+    DOCUMENT_EXTRACTOR = "document-extractor"
 
     UNKNOWN = "unknown"
     NOTES = "notes"
 
 
-BRANCH_NODES = [NodeType.IF_ELSE, NodeType.HTTP_REQUEST, NodeType.QUESTION_CLASSIFIER]
+BRANCH_NODES = frozenset({NodeType.IF_ELSE, NodeType.HTTP_REQUEST, NodeType.QUESTION_CLASSIFIER})
 
 
 class ComparisonOperator(StrEnum):

api/app/core/workflow/nodes/http_request/config.py

@@ -115,7 +115,7 @@ class HttpRetryConfig(BaseModel):
     )
 
 
-class HttpErrorDefaultTamplete(BaseModel):
+class HttpErrorDefaultTemplate(BaseModel):
     body: str = Field(
         default="",
         description="Default body returned on HTTP error",
@@ -143,7 +143,7 @@ class HttpErrorHandleConfig(BaseModel):
         description="Error handling strategy: 'none', 'default', or 'branch'",
     )
 
-    default: HttpErrorDefaultTamplete | None = Field(
+    default: HttpErrorDefaultTemplate | None = Field(
         default=None,
         description="Default response template for error handling",
     )

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

@@ -16,7 +16,7 @@ from app.core.workflow.engine.variable_pool import VariablePool
 from app.core.workflow.nodes.base_node import BaseNode
 from app.core.workflow.nodes.enums import HttpRequestMethod, HttpErrorHandle, HttpAuthType, HttpContentType
 from app.core.workflow.nodes.http_request.config import HttpRequestNodeConfig, HttpRequestNodeOutput
-from app.core.workflow.utils.file_processer import mime_to_file_type
+from app.core.workflow.utils.file_processor import mime_to_file_type
 from app.core.workflow.variable.base_variable import VariableType, FileObject
 from app.core.workflow.variable.variable_objects import FileVariable, ArrayVariable
 from app.schemas import FileType, TransferMethod
@@ -157,8 +157,8 @@ class HttpRequestNode(BaseNode):
     or a branch identifier string when error branching is enabled.
     """
 
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
         self.typed_config: HttpRequestNodeConfig | None = None
 
     def _output_types(self) -> dict[str, VariableType]:

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

@@ -14,8 +14,8 @@ logger = logging.getLogger(__name__)
 
 
 class IfElseNode(BaseNode):
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
         self.typed_config: IfElseNodeConfig | None = None
 
     def _output_types(self) -> dict[str, VariableType]:

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

@@ -12,8 +12,8 @@ logger = logging.getLogger(__name__)
 
 
 class JinjaRenderNode(BaseNode):
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
         self.typed_config: JinjaRenderNodeConfig | None = None
 
     def _output_types(self) -> dict[str, VariableType]:

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

@@ -1,19 +1,23 @@
+import asyncio
 import logging
 import uuid
 from typing import Any
 
+from langchain_core.documents import Document
+
 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, ElasticSearchVector
+from app.core.rag.models.chunk import DocumentChunk
+from app.core.rag.vdb.elasticsearch.elasticsearch_vector import ElasticSearchVectorFactory
 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
 from app.core.workflow.nodes.knowledge import KnowledgeRetrievalNodeConfig
 from app.core.workflow.variable.base_variable import VariableType
 from app.db import get_db_read
-from app.models import knowledge_model, knowledgeshare_model, ModelType
-from app.repositories import knowledge_repository, knowledgeshare_repository
+from app.models import knowledge_model, ModelType
+from app.repositories import knowledge_repository
 from app.schemas.chunk_schema import RetrieveType
 from app.services.model_service import ModelConfigService
 
@@ -21,10 +25,9 @@ logger = logging.getLogger(__name__)
 
 
 class KnowledgeRetrievalNode(BaseNode):
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
         self.typed_config: KnowledgeRetrievalNodeConfig | None = None
-        self.vector_service: ElasticSearchVector | None = None
 
     def _output_types(self) -> dict[str, VariableType]:
         return {
@@ -85,46 +88,54 @@ class KnowledgeRetrievalNode(BaseNode):
                 unique.append(doc)
         return unique
 
-    def _get_existing_kb_ids(self, db, kb_ids):
+    def rerank(self, query: str, docs: list[DocumentChunk], top_k: int) -> list[DocumentChunk]:
         """
-        Resolve all accessible and valid knowledge base IDs for retrieval.
-
-        This includes:
-        - Private knowledge bases owned by the user
-        - Shared knowledge bases
-        - Source knowledge bases mapped via knowledge sharing relationships
-
+        Reorder the list of document blocks and return the top_k results most relevant to the query
         Args:
-            db: Database session.
-            kb_ids (list[UUID]): Knowledge base IDs from node configuration.
+            query: query string
+            docs: List of document chunk to be rearranged
+            top_k: The number of top-level documents returned
 
         Returns:
-            list[UUID]: Final list of valid knowledge base IDs.
+            Rearranged document chunk list (sorted in descending order of relevance)
+
+        Raises:
+            ValueError: If the input document list is empty or top_k is invalid
         """
-        filters = self._build_kb_filter(kb_ids, knowledge_model.PermissionType.Private)
-
-        existing_ids = knowledge_repository.get_chunked_knowledgeids(
-            db=db,
-            filters=filters
-        )
-
-        filters = self._build_kb_filter(kb_ids, knowledge_model.PermissionType.Share)
-
-        share_ids = knowledge_repository.get_chunked_knowledgeids(
-            db=db,
-            filters=filters
-        )
-
-        if share_ids:
-            filters = [
-                knowledgeshare_model.KnowledgeShare.target_kb_id.in_(kb_ids)
+        reranker = self.get_reranker_model()
+        # parameter validation
+        if not docs:
+            raise ValueError("retrieval chunks be empty")
+        if top_k <= 0:
+            raise ValueError("top_k must be a positive integer")
+        try:
+            # Convert to LangChain Document object
+            documents = [
+                Document(
+                    page_content=doc.page_content,  # Ensure that DocumentChunk possesses this attribute
+                    metadata=doc.metadata or {}  # Deal with possible None metadata
+                )
+                for doc in docs
             ]
-            items = knowledgeshare_repository.get_source_kb_ids_by_target_kb_id(
-                db=db,
-                filters=filters
+
+            # Perform reordering (compress_documents will automatically handle relevance scores and indexing)
+            reranked_docs = list(reranker.compress_documents(documents, query))
+
+            # Sort in descending order based on relevance score
+            reranked_docs.sort(
+                key=lambda x: x.metadata.get("relevance_score", 0),
+                reverse=True
             )
-            existing_ids.extend(items)
-        return existing_ids
+            # Convert back to a list of DocumentChunk, and save the relevance_score to metadata["score"]
+            result = []
+            for item in reranked_docs[:top_k]:
+                for doc in docs:
+                    if doc.page_content == item.page_content:
+                        doc.metadata["score"] = item.metadata["relevance_score"]
+                        result.append(doc)
+            return result
+        except Exception as e:
+            raise RuntimeError(f"Failed to rerank documents: {str(e)}") from e
 
     def get_reranker_model(self) -> RedBearRerank:
         """
@@ -164,41 +175,77 @@ class KnowledgeRetrievalNode(BaseNode):
         )
         return reranker
 
-    def knowledge_retrieval(self, db, query, rs, db_knowledge, kb_config):
+    async def knowledge_retrieval(self, db, query, db_knowledge, kb_config):
+        rs = []
         if db_knowledge.type == knowledge_model.KnowledgeType.FOLDER:
             children = knowledge_repository.get_knowledges_by_parent_id(db=db, parent_id=db_knowledge.id)
+            tasks = []
             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)
+                child_kb_config = kb_config.model_copy()
+                child_kb_config.kb_id = child.id
+                tasks.append(self.knowledge_retrieval(db, query, child, child_kb_config))
+            if tasks:
+                result = await asyncio.gather(*tasks)
+                for _ in result:
+                    rs.extend(_)
+            return rs
+        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))
+                rs.extend(
+                    await asyncio.to_thread(
+                        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))
+                rs.extend(
+                    await asyncio.to_thread(
+                        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)
+                rs1_task = asyncio.to_thread(
+                        vector_service.search_by_vector, **{
+                            "query": query,
+                            "top_k": kb_config.top_k,
+                            "indices": indices,
+                            "score_threshold": kb_config.vector_similarity_weight
+                        }
+                    )
+                rs2_task = asyncio.to_thread(
+                        vector_service.search_by_full_text, **{
+                            "query": query,
+                            "top_k": kb_config.top_k,
+                            "indices": indices,
+                            "score_threshold": kb_config.similarity_threshold
+                        }
+                    )
+                rs1, rs2 = await asyncio.gather(rs1_task, rs2_task)
 
                 # Deduplicate hybrid retrieval results
                 unique_rs = self._deduplicate_docs(rs1, rs2)
                 if not unique_rs:
-                    return
+                    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))
+                    rs.extend(
+                        await asyncio.to_thread(
+                            self.rerank,
+                            **{"query": query, "docs": unique_rs, "top_k": kb_config.top_k}
+                        )
+                    )
                 else:
                     rs.extend(sorted(
                         unique_rs,
@@ -207,6 +254,7 @@ class KnowledgeRetrievalNode(BaseNode):
                     )[:kb_config.top_k])
             case _:
                 raise RuntimeError("Unknown retrieval type")
+        return rs
 
     async def execute(self, state: WorkflowState, variable_pool: VariablePool) -> Any:
         """
@@ -238,17 +286,24 @@ class KnowledgeRetrievalNode(BaseNode):
             knowledge_bases = self.typed_config.knowledge_bases
 
             rs = []
+            tasks = []
             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)
+                tasks.append(self.knowledge_retrieval(db, query, db_knowledge, kb_config))
+            if tasks:
+                result = await asyncio.gather(*tasks)
+                for _ in result:
+                    rs.extend(_)
 
             if not rs:
                 return []
             if self.typed_config.reranker_id:
-                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)
+                final_rs = await asyncio.to_thread(
+                    self.rerank,
+                    **{"query": query, "docs": rs, "top_k": self.typed_config.reranker_top_k}
+                )
             else:
                 final_rs = sorted(
                     rs,

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

@@ -70,8 +70,8 @@ class LLMNode(BaseNode):
     - ai/assistant: AI 消息（AIMessage）
     """
 
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
         self.typed_config: LLMNodeConfig | None = None
         self.messages = []
 
@@ -144,7 +144,6 @@ class LLMNode(BaseNode):
             f"创建 LLM 实例: provider={model_info.provider}, model={model_info.model_name}, streaming={stream}")
 
         messages_config = self.typed_config.messages
-
         if messages_config:
             # 使用 LangChain 消息格式
             messages = []
@@ -153,7 +152,6 @@ class LLMNode(BaseNode):
                 content_template = msg_config.content
                 content_template = self._render_context(content_template, variable_pool)
                 content = self._render_template(content_template, variable_pool)
-                user_id = self.get_variable("sys.user_id", variable_pool)
                 # 根据角色创建对应的消息对象
                 if role == "system":
                     messages.append({
@@ -161,32 +159,31 @@ class LLMNode(BaseNode):
                         "content": await self.process_message(
                             model_info,
                             content,
-                            user_id,
                             self.typed_config.vision,
                         )
                     })
                 elif role in ["user", "human"]:
                     messages.append({
                         "role": "user",
-                        "content": await self.process_message(model_info, content, user_id, self.typed_config.vision)
+                        "content": await self.process_message(model_info, content, self.typed_config.vision)
                     })
                 elif role in ["ai", "assistant"]:
                     messages.append({
                         "role": "assistant",
-                        "content": await self.process_message(model_info, content, user_id, self.typed_config.vision)
+                        "content": await self.process_message(model_info, content, self.typed_config.vision)
                     })
                 else:
                     logger.warning(f"未知的消息角色: {role}，默认使用 user")
                     messages.append({
                         "role": "user",
-                        "content": await self.process_message(model_info, content, user_id, self.typed_config.vision)
+                        "content": await self.process_message(model_info, content, self.typed_config.vision)
                     })
 
             if self.typed_config.vision_input and self.typed_config.vision:
                 file_content = []
                 files = variable_pool.get_instance(self.typed_config.vision_input)
                 for file in files.value:
-                    content = await self.process_message(model_info, file.value, user_id, self.typed_config.vision)
+                    content = await self.process_message(model_info, file.value, self.typed_config.vision)
                     if content:
                         file_content.extend(content)
                 if messages and messages[-1]["role"] == 'user':
@@ -200,7 +197,7 @@ class LLMNode(BaseNode):
                     if isinstance(message["content"], list):
                         file_content = []
                         for file in message["content"]:
-                            content = await self.process_message(model_info, file, user_id, self.typed_config.vision)
+                            content = await self.process_message(model_info, file, self.typed_config.vision)
                             if content:
                                 file_content.extend(content)
                         history_message.append(
@@ -210,7 +207,6 @@ class LLMNode(BaseNode):
                         message["content"] = await self.process_message(
                             model_info,
                             message["content"],
-                            user_id,
                             self.typed_config.vision
                         )
                         history_message.append(message)

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

@@ -1,3 +1,4 @@
+import re
 from typing import Any
 
 from app.core.workflow.engine.state_manager import WorkflowState
@@ -5,14 +6,16 @@ from app.core.workflow.engine.variable_pool import VariablePool
 from app.core.workflow.nodes.base_node import BaseNode
 from app.core.workflow.nodes.memory.config import MemoryReadNodeConfig, MemoryWriteNodeConfig
 from app.core.workflow.variable.base_variable import VariableType
+from app.core.workflow.variable.variable_objects import FileVariable, ArrayVariable
 from app.db import get_db_read
+from app.schemas import FileInput
 from app.services.memory_agent_service import MemoryAgentService
 from app.tasks import write_message_task
 
 
 class MemoryReadNode(BaseNode):
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
         self.typed_config: MemoryReadNodeConfig | None = None
 
     def _output_types(self) -> dict[str, VariableType]:
@@ -36,19 +39,32 @@ class MemoryReadNode(BaseNode):
                 search_switch=self.typed_config.search_switch,
                 history=[],
                 db=db,
-                storage_type="neo4j",
-                user_rag_memory_id=""
+                storage_type=state["memory_storage_type"],
+                user_rag_memory_id=state["user_rag_memory_id"]
             )
 
 
 class MemoryWriteNode(BaseNode):
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
         self.typed_config: MemoryWriteNodeConfig | None = None
 
     def _output_types(self) -> dict[str, VariableType]:
         return {"output": VariableType.STRING}
 
+    @staticmethod
+    def _extract_multimodal_memory_variables(content: str, variable_pool: VariablePool) -> tuple[list[str], str]:
+        variable_pattern_string = r'\{\{\s*[a-zA-Z0-9_]+\.[a-zA-Z0-9_]+\s*\}\}'
+        variable_pattern = re.compile(variable_pattern_string)
+        variables = variable_pattern.findall(content)
+        file_variables = []
+        for variable in variables:
+            if variable_pool.is_file_variable(variable):
+                file_variables.append(variable)
+        for var in file_variables:
+            content = content.replace(var, "")
+        return file_variables, content
+
     async def execute(self, state: WorkflowState, variable_pool: VariablePool) -> Any:
         self.typed_config = MemoryWriteNodeConfig(**self.config)
         end_user_id = self.get_variable("sys.user_id", variable_pool)
@@ -63,17 +79,42 @@ class MemoryWriteNode(BaseNode):
             })
 
         for message in self.typed_config.messages:
+            file_variables, content = self._extract_multimodal_memory_variables(
+                message.content,
+                variable_pool
+            )
+            file_info = []
+            for var in file_variables:
+                instence: FileVariable | ArrayVariable[FileVariable] = variable_pool.get_instance(var)
+                if isinstance(instence, FileVariable):
+                    file_info.append(FileInput(
+                        type=instence.value.type,
+                        transfer_method=instence.value.transfer_method,
+                        upload_file_id=instence.value.file_id,
+                        url=instence.value.url,
+                        file_type=instence.value.origin_file_type
+                    ).model_dump())
+                elif isinstance(instence, ArrayVariable) and instence.child_type == FileVariable:
+                    for file_instence in instence.value:
+                        file_info.append(FileInput(
+                            type=file_instence.value.type,
+                            transfer_method=file_instence.value.transfer_method,
+                            upload_file_id=file_instence.value.file_id,
+                            url=file_instence.value.url,
+                            file_type=file_instence.value.origin_file_type
+                        ).model_dump())
             messages.append({
                 "role": message.role,
-                "content": self._render_template(message.content, variable_pool)
+                "content": self._render_template(content, variable_pool),
+                "files": file_info
             })
 
         write_message_task.delay(
-            end_user_id,
-            messages,
-            str(self.typed_config.config_id),
-            "neo4j",
-            ""
+            end_user_id=end_user_id,
+            message=messages,
+            config_id=str(self.typed_config.config_id),
+            storage_type=state["memory_storage_type"],
+            user_rag_memory_id=state["user_rag_memory_id"]
         )
 
         return "success"

api/app/core/workflow/nodes/node_factory.py

@@ -26,6 +26,7 @@ from app.core.workflow.nodes.variable_aggregator import VariableAggregatorNode
 from app.core.workflow.nodes.question_classifier import QuestionClassifierNode
 from app.core.workflow.nodes.breaker import BreakNode
 from app.core.workflow.nodes.tool import ToolNode
+from app.core.workflow.nodes.document_extractor import DocExtractorNode
 
 logger = logging.getLogger(__name__)
 
@@ -49,7 +50,8 @@ WorkflowNode = Union[
     ToolNode,
     MemoryReadNode,
     MemoryWriteNode,
-    CodeNode
+    CodeNode,
+    DocExtractorNode
 ]
 
 
@@ -81,6 +83,7 @@ class NodeFactory:
         NodeType.MEMORY_READ: MemoryReadNode,
         NodeType.MEMORY_WRITE: MemoryWriteNode,
         NodeType.CODE: CodeNode,
+        NodeType.DOCUMENT_EXTRACTOR: DocExtractorNode
     }
 
     @classmethod
@@ -104,13 +107,15 @@ class NodeFactory:
     def create_node(
             cls,
             node_config: dict[str, Any],
-            workflow_config: dict[str, Any]
+            workflow_config: dict[str, Any],
+            down_stream_nodes: list[str]
     ) -> WorkflowNode | None:
         """创建节点实例
 
         Args:
             node_config: 节点配置
             workflow_config: 工作流配置
+            down_stream_nodes: 下游节点
 
         Returns:
             节点实例或 None（对于不支持的节点类型）
@@ -127,7 +132,7 @@ class NodeFactory:
 
         # 创建节点实例
         logger.debug(f"create node instance: {node_config.get('id')} (type={node_type})")
-        return node_class(node_config, workflow_config)
+        return node_class(node_config, workflow_config, down_stream_nodes)
 
     @classmethod
     def get_supported_types(cls) -> list[str]:

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

@@ -21,8 +21,8 @@ logger = logging.getLogger(__name__)
 
 
 class ParameterExtractorNode(BaseNode):
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
         self.typed_config: ParameterExtractorNodeConfig | None = None
         self.response_metadata = {}
 

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

@@ -22,8 +22,8 @@ DEFAULT_EMPTY_QUESTION_CASE = f"{DEFAULT_CASE_PREFIX}1"
 class QuestionClassifierNode(BaseNode):
     """问题分类器节点"""
 
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
         self.typed_config: QuestionClassifierNodeConfig | None = None
         self.category_to_case_map = {}
         self.response_metadata = {}

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

@@ -27,14 +27,8 @@ class StartNode(BaseNode):
     注意：变量的验证和默认值处理由 Executor 在初始化时完成。
     """
 
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        """初始化 Start 节点
-        
-        Args:
-            node_config: 节点配置
-            workflow_config: 工作流配置
-        """
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
 
         # 解析并验证配置
         self.typed_config: StartNodeConfig | None = None
@@ -62,7 +56,6 @@ class StartNode(BaseNode):
             包含系统参数、会话变量和自定义变量的字典
         """
         self.typed_config = StartNodeConfig(**self.config)
-        logger.info(f"节点 {self.node_id} (Start) 开始执行")
 
         # 处理自定义变量（传入 pool 避免重复创建）
         custom_vars = self._process_custom_variables(variable_pool)
@@ -77,9 +70,9 @@ class StartNode(BaseNode):
             **custom_vars  # 自定义变量作为节点输出的一部分
         }
 
-        logger.info(
-            f"节点 {self.node_id} (Start) 执行完成，"
-            f"输出了 {len(custom_vars)} 个自定义变量"
+        logger.debug(
+            f"Node {self.node_id} (Start) execution completed, "
+            f"outputting {len(custom_vars)} custom variables"
         )
 
         return result

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

@@ -20,8 +20,8 @@ TEMPLATE_PATTERN = re.compile(r"\{\{.*?}}")
 class ToolNode(BaseNode):
     """工具节点"""
 
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
         self.typed_config: ToolNodeConfig | None = None
 
     def _output_types(self) -> dict[str, VariableType]:

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

@@ -12,8 +12,8 @@ logger = logging.getLogger(__name__)
 
 
 class VariableAggregatorNode(BaseNode):
-    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any]):
-        super().__init__(node_config, workflow_config)
+    def __init__(self, node_config: dict[str, Any], workflow_config: dict[str, Any], down_stream_nodes: list[str]):
+        super().__init__(node_config, workflow_config, down_stream_nodes)
         self.typed_config: VariableAggregatorNodeConfig | None = None
 
     def _output_types(self) -> dict[str, VariableType]:

api/app/core/workflow/utils/expression_evaluator.py

@@ -4,32 +4,33 @@ from typing import Any
 
 from simpleeval import simple_eval, NameNotDefined, InvalidExpression
 
+from app.core.workflow.engine.variable_pool import LazyVariableDict, VARIABLE_PATTERN
+
 logger = logging.getLogger(__name__)
 
+_NORMALIZE_PATTERN = re.compile(r"\{\{\s*(\d+)\.(\w+)\s*}}")
+
 
 class ExpressionEvaluator:
     """Safe expression evaluator for workflow variables and node outputs."""
-    
+
     # Reserved namespaces
     RESERVED_NAMESPACES = {"var", "node", "sys", "nodes"}
 
     @classmethod
     def normalize_template(cls, template: str) -> str:
-        pattern = re.compile(
-            r"\{\{\s*(\d+)\.(\w+)\s*}}"
-        )
-        return pattern.sub(
+        return _NORMALIZE_PATTERN.sub(
             r'{{ node["\1"].\2 }}',
             template
         )
 
     @classmethod
     def evaluate(
-        cls,
-        expression: str,
-        conv_vars: dict[str, Any],
-        node_outputs: dict[str, Any],
-        system_vars: dict[str, Any] | None = None
+            cls,
+            expression: str,
+            conv_vars: dict[str, Any],
+            node_outputs: dict[str, Any],
+            system_vars: dict[str, Any] | None = None
     ) -> Any:
         """
         Safely evaluate an expression using workflow variables.
@@ -49,48 +50,47 @@ class ExpressionEvaluator:
         # Remove Jinja2-style brackets if present
         expression = expression.strip()
         expression = cls.normalize_template(expression)
-        pattern = r"\{\{\s*(.*?)\s*\}\}"
-        expression = re.sub(pattern, r"\1", expression).strip()
+        expression = VARIABLE_PATTERN.sub(r"\1", expression).strip()
 
         # Build context for evaluation
         context = {
-            "conv": conv_vars,                   # conversation variables
-            "node": node_outputs,                # node outputs
-            "sys": system_vars or {},            # system variables
+            "conv": conv_vars,  # conversation variables
+            "node": node_outputs,  # node outputs
+            "sys": system_vars or {},  # system variables
         }
 
-        context.update(conv_vars)
-        context["nodes"] = node_outputs
+        # context.update(conv_vars)
+        # context["nodes"] = node_outputs
         context.update(node_outputs)
-        
+
         try:
             # simpleeval supports safe operations:
             # arithmetic, comparisons, logical ops, attribute/dict/list access
             result = simple_eval(expression, names=context)
             return result
-            
+
         except NameNotDefined as e:
             logger.error(f"Undefined variable in expression: {expression}, error: {e}")
             raise ValueError(f"Undefined variable: {e}")
-            
+
         except InvalidExpression as e:
             logger.error(f"Invalid expression syntax: {expression}, error: {e}")
             raise ValueError(f"Invalid expression syntax: {e}")
-            
+
         except SyntaxError as e:
             logger.error(f"Syntax error in expression: {expression}, error: {e}")
             raise ValueError(f"Syntax error: {e}")
-            
+
         except Exception as e:
             logger.error(f"Expression evaluation failed: {expression}, error: {e}")
             raise ValueError(f"Expression evaluation failed: {e}")
-    
+
     @staticmethod
     def evaluate_bool(
-        expression: str,
-        conv_var: dict[str, Any],
-        node_outputs: dict[str, Any],
-        system_vars: dict[str, Any] | None = None
+            expression: str,
+            conv_var: dict[str, Any],
+            node_outputs: dict[str, Any],
+            system_vars: dict[str, Any] | None = None
     ) -> bool:
         """
         Evaluate a boolean expression (for conditions).
@@ -108,7 +108,7 @@ class ExpressionEvaluator:
             expression, conv_var, node_outputs, system_vars
         )
         return bool(result)
-    
+
     @staticmethod
     def validate_variable_names(variables: list[dict]) -> list[str]:
         """
@@ -121,7 +121,7 @@ class ExpressionEvaluator:
             list[str]: List of error messages. Empty if all names are valid.
         """
         errors = []
-        
+
         for var in variables:
             var_name = var.get("name", "")
 
@@ -134,16 +134,16 @@ class ExpressionEvaluator:
                 errors.append(
                     f"Variable name '{var_name}' is not a valid Python identifier"
                 )
-        
+
         return errors
 
 
 # 便捷函数
 def evaluate_expression(
-    expression: str,
-    conv_var: dict[str, Any],
-    node_outputs: dict[str, Any],
-    system_vars: dict[str, Any]
+        expression: str,
+        conv_var: dict[str, Any] | LazyVariableDict,
+        node_outputs: dict[str, dict[str, Any] | LazyVariableDict],
+        system_vars: dict[str, Any] | LazyVariableDict
 ) -> Any:
     """Evaluate an expression (convenience function)."""
     return ExpressionEvaluator.evaluate(
@@ -152,11 +152,11 @@ def evaluate_expression(
 
 
 def evaluate_condition(
-    expression: str,
-    conv_var: dict[str, Any],
-    node_outputs: dict[str, Any],
-    system_vars: dict[str, Any] | None = None
-) -> bool:
+        expression: str,
+        conv_var: dict[str, Any] | LazyVariableDict,
+        node_outputs: dict[str, dict[str, Any] | LazyVariableDict],
+        system_vars: dict[str, Any] | LazyVariableDict
+) -> Any:
     """Evaluate a boolean condition expression (convenience function)."""
     return ExpressionEvaluator.evaluate_bool(
         expression, conv_var, node_outputs, system_vars

api/app/core/workflow/utils/template_renderer.py

@@ -1,7 +1,8 @@
 """
-模板渲染器
+Template Renderer
 
-使用 Jinja2 提供安全的模板渲染功能，支持变量引用和表达式。
+Provides safe template rendering using Jinja2, supporting variable references
+and expressions.
 """
 
 import logging
@@ -10,11 +11,15 @@ from typing import Any
 
 from jinja2 import TemplateSyntaxError, UndefinedError, Environment, StrictUndefined, Undefined
 
+from app.core.workflow.engine.variable_pool import LazyVariableDict
+
 logger = logging.getLogger(__name__)
 
+_NORMALIZE_PATTERN = re.compile(r"\{\{\s*(\d+)\.(\w+)\s*}}")
+
 
 class SafeUndefined(Undefined):
-    """访问未定义属性不会报错，返回空字符串"""
+    """Return empty string instead of raising error when accessing undefined variables"""
     __slots__ = ()
 
     def _fail_with_undefined_error(self, *args, **kwargs):
@@ -26,26 +31,22 @@ class SafeUndefined(Undefined):
 
 
 class TemplateRenderer:
-    """模板渲染器"""
-
     def __init__(self, strict: bool = True):
-        """初始化渲染器
-        
+        """Initialize renderer
+
         Args:
-            strict: 是否使用严格模式（未定义变量会抛出异常）
+            strict: Whether to enable strict mode (raise error on undefined variables)
         """
         self.strict = strict
         self.env = Environment(
             undefined=StrictUndefined if strict else SafeUndefined,
-            autoescape=False  # 不自动转义，因为我们处理的是文本而非 HTML
+            autoescape=False  # Disable auto-escaping since we handle plain text instead of HTML
         )
 
     @staticmethod
     def normalize_template(template: str) -> str:
-        pattern = re.compile(
-            r"\{\{\s*(\d+)\.(\w+)\s*}}"
-        )
-        return pattern.sub(
+        """Normalize template syntax (convert numeric node reference to dict access)"""
+        return _NORMALIZE_PATTERN.sub(
             r'{{ node["\1"].\2 }}',
             template
         )
@@ -53,24 +54,24 @@ class TemplateRenderer:
     def render(
             self,
             template: str,
-            conv_vars: dict[str, Any],
-            node_outputs: dict[str, Any],
-            system_vars: dict[str, Any] | None = None
+            conv_vars: dict[str, Any] | LazyVariableDict,
+            node_outputs: dict[str, Any] | dict[str, LazyVariableDict],
+            system_vars: dict[str, Any] | LazyVariableDict | None = None
     ) -> str:
-        """渲染模板
-        
+        """Render template
+
         Args:
-            template: 模板字符串
-            conv_vars: 会话变量
-            node_outputs: 节点输出结果
-            system_vars: 系统变量
-        
+            template: Template string
+            conv_vars: Conversation variables
+            node_outputs: Node outputs
+            system_vars: System variables
+
         Returns:
-            渲染后的字符串
-        
+            Rendered string
+
         Raises:
-            ValueError: 模板语法错误或变量未定义
-        
+            ValueError: If template syntax is invalid or variables are undefined
+
         Examples:
             >>> renderer = TemplateRenderer()
             >>> renderer.render(
@@ -80,121 +81,119 @@ class TemplateRenderer:
             ...     {}
             ... )
             'Hello World!'
-            
+
             >>> renderer.render(
-            ...     "分析结果: {{node.analyze.output}}",
+            ...     "Analysis result: {{node.analyze.output}}",
             ...     {},
-            ...     {"analyze": {"output": "正面情绪"}},
+            ...     {"analyze": {"output": "positive sentiment"}},
             ...     {}
             ... )
-            '分析结果: 正面情绪'
+            'Analysis result: positive sentiment'
         """
-        # 构建命名空间上下文
+        # Build namespace context
         context = {
-            "conv": conv_vars,  # 会话变量：{{conv.user_name}}
-            "node": node_outputs,  # 节点输出：{{node.node_1.output}}
-            "sys": system_vars,  # 系统变量：{{sys.execution_id}}
+            "conv": conv_vars,  # Conversation variables: {{conv.user_name}}
+            "node": node_outputs,  # Node outputs: {{node.node_1.output}}
+            "sys": system_vars,  # System variables: {{sys.execution_id}}
         }
 
-        # 支持直接通过节点ID访问节点输出：{{llm_qa.output}}
-        # 将所有节点输出添加到顶层上下文
+        # Allow direct access to node outputs by node ID: {{llm_qa.output}}
         if node_outputs:
             context.update(node_outputs)
 
-        # 支持直接访问会话变量（不需要 conv. 前缀）：{{user_name}}
-        if conv_vars:
-            context.update(conv_vars)
-
-        context["nodes"] = node_outputs or {}  # 旧语法兼容
+        # # 支持直接访问会话变量（不需要 conv. 前缀）：{{user_name}}
+        # if conv_vars:
+        #     context.update(conv_vars)
+        #
+        # context["nodes"] = node_outputs or {}  # 旧语法兼容
         template = self.normalize_template(template)
         try:
             tmpl = self.env.from_string(template)
             return tmpl.render(**context)
 
         except TemplateSyntaxError as e:
-            logger.error(f"模板语法错误: {template}, 错误: {e}")
-            raise ValueError(f"模板语法错误: {e}")
-
+            logger.error(f"Template syntax error: {template}, error: {e}")
+            raise ValueError(f"Template syntax error: {e}")
         except UndefinedError as e:
-            logger.error(f"模板中引用了未定义的变量: {template}, 错误: {e}")
-            raise ValueError(f"未定义的变量: {e}")
-
+            logger.error(f"Undefined variable in template: {template}, error: {e}")
+            raise ValueError(f"Undefined variable: {e}")
         except Exception as e:
-            logger.error(f"模板渲染异常: {template}, 错误: {e}")
-            raise ValueError(f"模板渲染失败: {e}")
+            logger.error(f"Template rendering error: {template}, error: {e}")
+            raise ValueError(f"Template rendering failed: {e}")
 
     def validate(self, template: str) -> list[str]:
-        """验证模板语法
-        
+        """Validate template syntax
+
         Args:
-            template: 模板字符串
-        
+            template: Template string
+
         Returns:
-            错误列表，如果为空则验证通过
-        
+            List of errors (empty if valid)
+
         Examples:
             >>> renderer = TemplateRenderer()
             >>> renderer.validate("Hello {{var.name}}!")
             []
-            
-            >>> renderer.validate("Hello {{var.name")  # 缺少结束标记
-            ['模板语法错误: ...']
+
+            >>> renderer.validate("Hello {{var.name")  # Missing closing tag
+            ['Template syntax error: ...']
         """
         errors = []
 
         try:
             self.env.from_string(template)
         except TemplateSyntaxError as e:
-            errors.append(f"模板语法错误: {e}")
+            errors.append(f"Template syntax error: {e}")
         except Exception as e:
-            errors.append(f"模板验证失败: {e}")
+            errors.append(f"Template validation failed: {e}")
 
         return errors
 
 
-# 全局渲染器实例（严格模式）
-_default_renderer = TemplateRenderer(strict=True)
+# Global renderer instances (strict / lenient)
+_strict_renderer = TemplateRenderer(strict=True)
+_lenient_renderer = TemplateRenderer(strict=False)
 
 
 def render_template(
         template: str,
-        conv_vars: dict[str, Any],
-        node_outputs: dict[str, Any],
-        system_vars: dict[str, Any],
+        conv_vars: dict[str, Any] | LazyVariableDict,
+        node_outputs: dict[str, Any] | dict[str, LazyVariableDict],
+        system_vars: dict[str, Any] | LazyVariableDict,
         strict: bool = True
 ) -> str:
-    """渲染模板（便捷函数）
-    
+    """Render template (convenience function)
+
     Args:
-        strict: 严格模式
-        template: 模板字符串
-        conv_vars: 会话变量
-        node_outputs: 节点输出
-        system_vars: 系统变量
-    
+        strict: Whether to use strict mode
+        template: Template string
+        conv_vars: Conversation variables
+        node_outputs: Node outputs
+        system_vars: System variables
+
     Returns:
-        渲染后的字符串
-    
+        Rendered string
+
     Examples:
         >>> render_template(
-        ...     "请分析: {{var.text}}",
-        ...     {"text": "这是一段文本"},
+        ...     "Analyze: {{var.text}}",
+        ...     {"text": "This is a text"},
         ...     {},
         ...     {}
         ... )
-        '请分析: 这是一段文本'
+        'Analyze: This is a text'
     """
-    renderer = TemplateRenderer(strict=strict)
+    renderer = _strict_renderer if strict else _lenient_renderer
     return renderer.render(template, conv_vars, node_outputs, system_vars)
 
 
 def validate_template(template: str) -> list[str]:
-    """验证模板语法（便捷函数）
-    
+    """Validate template syntax (convenience function)
+
     Args:
-        template: 模板字符串
-    
+        template: Template string
+
     Returns:
-        错误列表
+        List of errors
     """
-    return _default_renderer.validate(template)
+    return _strict_renderer.validate(template)

api/app/core/workflow/validator.py

@@ -6,6 +6,7 @@
 
 import copy
 import logging
+from collections import defaultdict, deque
 from typing import Any, Union, TYPE_CHECKING
 
 from app.core.workflow.nodes.enums import NodeType
@@ -119,7 +120,6 @@ class WorkflowValidator:
         errors = []
 
         graphs = cls.get_subgraph(workflow_config)
-        logger.info(graphs)
         for index, graph in enumerate(graphs):
             nodes = graph.get("nodes", [])
             edges = graph.get("edges", [])
@@ -183,7 +183,7 @@ class WorkflowValidator:
                 has_cycle, cycle_path = WorkflowValidator._has_cycle(nodes, edges)
                 if has_cycle:
                     errors.append(
-                        f"工作流存在循环依赖（请使用 loop 节点实现循环）: {' -> '.join(cycle_path)}"
+                        f"工作流存在循环依赖（请使用 loop/iteration 节点实现循环）: {' -> '.join(cycle_path)}"
                     )
 
             # 8. 验证变量名
@@ -204,18 +204,18 @@ class WorkflowValidator:
         Returns:
             可达节点 ID 集合
         """
+        adj = defaultdict(list)
+        for edge in edges:
+            adj[edge["source"]].append(edge["target"])
+
         reachable = {start_id}
-        queue = [start_id]
-
+        queue = deque([start_id])
         while queue:
-            current = queue.pop(0)
-            for edge in edges:
-                if edge.get("source") == current:
-                    target = edge.get("target")
-                    if target and target not in reachable:
-                        reachable.add(target)
-                        queue.append(target)
-
+            current = queue.popleft()
+            for target in adj[current]:
+                if target not in reachable:
+                    reachable.add(target)
+                    queue.append(target)
         return reachable
 
     @staticmethod
@@ -229,10 +229,6 @@ class WorkflowValidator:
         Returns:
             (has_cycle, cycle_path): 是否有循环和循环路径
         """
-        # 排除 loop 类型的节点
-        loop_nodes = {n["id"] for n in nodes if n.get("type") == "loop"}
-
-        # 构建邻接表（排除 loop 节点的边和错误边）
         graph: dict[str, list[str]] = {}
         for edge in edges:
             source = edge.get("source")
@@ -243,10 +239,6 @@ class WorkflowValidator:
             if edge_type == "error":
                 continue
 
-            # 如果涉及 loop 节点，跳过
-            if source in loop_nodes or target in loop_nodes:
-                continue
-
             if source and target:
                 if source not in graph:
                     graph[source] = []

api/app/core/workflow/variable/variable_objects.py

@@ -54,7 +54,7 @@ class DictVariable(BaseVariable):
 
     def valid_value(self, value) -> dict:
         if not isinstance(value, dict):
-            raise TypeError(f"Value must be a dict  - {type(value)}:{value}")
+            raise TypeError(f"Value must be a dict - {type(value)}:{value}")
         return value
 
     def to_literal(self) -> str:

api/app/main.py

@@ -1,5 +1,6 @@
 import os
 import subprocess
+from app.repositories.neo4j.create_indexes import create_all_indexes
 from contextlib import asynccontextmanager
 
 from fastapi import FastAPI, APIRouter
@@ -60,8 +61,10 @@ async def lifespan(app: FastAPI):
             logger.warning(f"加载预定义模型时出错: {str(e)}")
     else:
         logger.info("预定义模型加载已禁用 (LOAD_MODEL=false)")
-
+    await create_all_indexes()
     logger.info("应用程序启动完成")
+
+
     yield
     # 应用关闭事件
     logger.info("应用程序正在关闭")

api/app/models/__init__.py

@@ -16,6 +16,7 @@ from .agent_app_config_model import AgentConfig
 from .app_release_model import AppRelease
 from .memory_increment_model import MemoryIncrement
 from .end_user_model import EndUser
+from .end_user_info_model import EndUserInfo
 from .appshare_model import AppShare
 from .release_share_model import ReleaseShare
 from .conversation_model import Conversation, Message
@@ -60,6 +61,7 @@ __all__ = [
     "AppRelease",
     "MemoryIncrement",
     "EndUser",
+    "EndUserInfo",
     "AppShare",
     "ReleaseShare",
     "Conversation",