Merge branch 'release/v0.3.2' into feature/rag2

* release/v0.3.2: (245 commits)
  fix(conversation_schema): refine citations field type to Dict[str, Any]
  fix(tool_controller): re-raise HTTPException to preserve original status codes
  fix(workflow): add reasoning content, suggested questions, citations and audio status support
  feat(workflow): augment logging queries and ameliorate error handling
  fix(api_key): bypass publication check for SERVICE type API keys
  fix(multimodal_service): add '文档内容：' prefix to document text and simplify image placeholder text
  fix(api): convert config_id to string in write_router
  fix(api): convert end_user_id to string in write_router
  fix(multimodal_service): refactor image processing to use intermediate list before extending result
  fix(web): node status ui
  fix(api): correct import paths in memory_read and celery task command
  fix(api): correct import paths in memory_read and celery task command
  refactor(tool): flatten request body parameters for model exposure
  fix(api): correct import paths in memory_read and celery task command
  refactor(workflow): streamline node execution handling and log service logic
  feat(web): http request add process
  feat(web): workflow app logs
  fix(app_chat_service,draft_run_service): move system_prompt augmentation before LangChainAgent instantiation
  fix(app_chat_service,draft_run_service): move system_prompt augmentation before LangChainAgent instantiation
  refactor(http_request): simplify request handling and remove unused fields
  ...

# Conflicts:
#	api/app/controllers/file_controller.py
#	api/app/tasks.py

api/app/services/api_key_service.py

@@ -9,7 +9,7 @@ from sqlalchemy.orm import Session
 from sqlalchemy import select
 
 from app.aioRedis import aio_redis
-from app.models.api_key_model import ApiKey
+from app.models.api_key_model import ApiKey, ApiKeyType
 from app.repositories.api_key_repository import ApiKeyRepository, ApiKeyLogRepository
 from app.schemas import api_key_schema
 from app.schemas.response_schema import PageData, PageMeta
@@ -19,6 +19,7 @@ from app.core.exceptions import (
 )
 from app.core.error_codes import BizCode
 from app.core.logging_config import get_business_logger
+from app.models.app_model import App
 
 logger = get_business_logger()
 
@@ -51,6 +52,25 @@ class ApiKeyService:
             if existing:
                 raise BusinessException(f"API Key 名称 {data.name} 已存在", BizCode.API_KEY_DUPLICATE_NAME)
 
+            # 若 rate_limit 超过租户套餐的 api_ops_rate_limit，直接报错
+            from app.models.workspace_model import Workspace
+            from app.core.quota_manager import get_api_ops_rate_limit
+
+            workspace = db.query(Workspace).filter(Workspace.id == workspace_id).first()
+            if workspace:
+                tenant_api_ops_limit = get_api_ops_rate_limit(db, workspace.tenant_id)
+                if tenant_api_ops_limit and data.rate_limit > tenant_api_ops_limit:
+                    raise BusinessException(
+                        f"API Key QPS 不能超过套餐上限 {tenant_api_ops_limit}",
+                        BizCode.BAD_REQUEST
+                    )
+
+            # SERVICE 类型的 resource_id 指向 workspace，非应用，跳过应用发布校验
+            if data.resource_id and data.type != ApiKeyType.SERVICE.value:
+                app = db.get(App, data.resource_id)
+                if not app or not app.current_release_id:
+                    raise BusinessException("该应用未发布", BizCode.APP_NOT_PUBLISHED)
+
             # 生成 API Key
             api_key = generate_api_key(data.type)
 
@@ -152,6 +172,20 @@ class ApiKeyService:
             if existing:
                 raise BusinessException(f"API Key 名称 {data.name} 已存在", BizCode.API_KEY_DUPLICATE_NAME)
 
+        # 若 rate_limit 超过租户套餐的 api_ops_rate_limit，直接报错
+        if data.rate_limit is not None:
+            from app.models.workspace_model import Workspace
+            from app.core.quota_manager import get_api_ops_rate_limit
+
+            workspace = db.query(Workspace).filter(Workspace.id == workspace_id).first()
+            if workspace:
+                tenant_api_ops_limit = get_api_ops_rate_limit(db, workspace.tenant_id)
+                if tenant_api_ops_limit and data.rate_limit > tenant_api_ops_limit:
+                    raise BusinessException(
+                        f"API Key QPS 不能超过套餐上限 {tenant_api_ops_limit}",
+                        BizCode.BAD_REQUEST
+                    )
+
         update_data = data.model_dump(exclude_unset=True)
         ApiKeyRepository.update(db, api_key_id, update_data)
         db.commit()
@@ -249,12 +283,13 @@ class RateLimiterService:
         self.redis = aio_redis
 
     async def check_qps(self, api_key_id: uuid.UUID, limit: int) -> Tuple[bool, dict]:
-        """
-        检查QPS限制
+        """检查QPS限制
+
         Returns:
             (is_allowed, rate_limit_info)
         """
         key = f"rate_limit:qps:{api_key_id}"
+
         async with self.redis.pipeline() as pipe:
             pipe.incr(key)
             pipe.expire(key, 1, nx=True)  # 1 秒过期
@@ -266,8 +301,9 @@ class RateLimiterService:
 
         return current <= limit, {
             "limit": limit,
+            "current": current,
             "remaining": remaining,
-            "reset": reset_time
+            "reset": reset_time,
         }
 
     async def check_daily_requests(
@@ -275,7 +311,9 @@ class RateLimiterService:
             api_key_id: uuid.UUID,
             limit: int
     ) -> Tuple[bool, dict]:
-        """检查日调用量限制"""
+        """检查日调用量限制。
+        使用原子 INCR，先写后判断，极低概率下允许轻微超限（并发场景下可接受）。
+        """
         today = datetime.now().strftime("%Y%m%d")
         key = f"rate_limit:daily:{api_key_id}:{today}"
 
@@ -284,6 +322,7 @@ class RateLimiterService:
             hour=0, minute=0, second=0, microsecond=0
         )
         expire_seconds = int((tomorrow_0 - now).total_seconds())
+        reset_time = int(tomorrow_0.timestamp())
 
         async with self.redis.pipeline() as pipe:
             pipe.incr(key)
@@ -291,36 +330,74 @@ class RateLimiterService:
             results = await pipe.execute()
 
         current = results[0]
-        remaining = max(0, limit - current)
-        reset_time = int(tomorrow_0.timestamp())
 
-        return current <= limit, {
+        if current > limit:
+            return False, {
+                "limit": limit,
+                "remaining": 0,
+                "reset": reset_time,
+            }
+
+        return True, {
             "limit": limit,
-            "remaining": remaining,
-            "reset": reset_time
+            "remaining": max(0, limit - current),
+            "reset": reset_time,
         }
 
     async def check_all_limits(
             self,
-            api_key: ApiKey
+            api_key: ApiKey,
+            db: Optional[Session] = None,
     ) -> Tuple[bool, str, dict]:
         """
-        检查所有限制
-        Returns:
-            (is_allowed, error_message, rate_limit_headers)
+        检查所有限制，按以下顺序：
+        1. API Key QPS：取 api_key.rate_limit 与套餐 api_ops_rate_limit 的最小值作为限额
+        2. API Key 日调用量
         """
-        # Check QPS
-        qps_ok, qps_info = await self.check_qps(
-            api_key.id,
-            api_key.rate_limit
-        )
+        # 1. 取套餐限额与 api_key 自身限额的最小值
+        effective_limit = api_key.rate_limit
+        if db is not None:
+            try:
+                from app.models.workspace_model import Workspace
+                from app.core.quota_manager import get_api_ops_rate_limit
+
+                cache_key = f"tenant_api_ops_limit:{api_key.workspace_id}"
+                cached = await self.redis.get(cache_key)
+                if cached is not None:
+                    try:
+                        tenant_limit = int(cached) if cached != "0" else None
+                    except (ValueError, TypeError):
+                        cached = None
+                        tenant_limit = None
+
+                if cached is None:
+                    workspace = db.query(Workspace).filter(Workspace.id == api_key.workspace_id).first()
+                    if workspace:
+                        tenant_limit = get_api_ops_rate_limit(db, workspace.tenant_id)
+                        await self.redis.set(cache_key, str(tenant_limit) if tenant_limit else "0", ex=60)
+                    else:
+                        tenant_limit = None
+
+                if tenant_limit:
+                    effective_limit = min(api_key.rate_limit, tenant_limit)
+            except Exception as e:
+                logger.warning(f"获取套餐限额失败，使用 api_key 自身限额: {e}")
+
+        # 用最终有效限额做 QPS 检查
+        qps_ok, qps_info = await self.check_qps(api_key.id, effective_limit)
         if not qps_ok:
-            return False, "QPS limit exceeded", {
+            # 判断是套餐限额触发还是 api_key 自身限额触发
+            if tenant_limit and effective_limit == tenant_limit and api_key.rate_limit > tenant_limit:
+                error_msg = "Tenant limit exceeded"
+            else:
+                error_msg = "QPS limit exceeded"
+            return False, error_msg, {
                 "X-RateLimit-Limit-QPS": str(qps_info["limit"]),
                 "X-RateLimit-Remaining-QPS": str(qps_info["remaining"]),
                 "X-RateLimit-Reset": str(qps_info["reset"])
             }
 
+        # 2. 检查日调用量
         daily_ok, daily_info = await self.check_daily_requests(
             api_key.id,
             api_key.daily_request_limit
@@ -332,14 +409,13 @@ class RateLimiterService:
                 "X-RateLimit-Reset": str(daily_info["reset"])
             }
 
-        headers = {
+        return True, "", {
             "X-RateLimit-Limit-QPS": str(qps_info["limit"]),
             "X-RateLimit-Remaining-QPS": str(qps_info["remaining"]),
             "X-RateLimit-Limit-Day": str(daily_info["limit"]),
             "X-RateLimit-Remaining-Day": str(daily_info["remaining"]),
-            "X-RateLimit-Reset": str(daily_info["reset"])
+            "X-RateLimit-Reset": str(daily_info["reset"]),
         }
-        return True, "", headers
 
 
 class ApiKeyAuthService:
@@ -373,6 +449,20 @@ class ApiKeyAuthService:
 
         return api_key_obj
 
+    @staticmethod
+    def check_app_published(db: Session, api_key_obj: ApiKey) -> None:
+        """
+        检查应用是否已发布，未发布则抛出异常
+        SERVICE 类型的 api_key 不绑定应用（resource_id 指向 workspace），跳过校验
+        """
+        if not api_key_obj.resource_id:
+            return
+        if api_key_obj.type == ApiKeyType.SERVICE.value:
+            return
+        app = db.get(App, api_key_obj.resource_id)
+        if not app or not app.current_release_id:
+            raise BusinessException("应用未发布，不可用", BizCode.APP_NOT_PUBLISHED)
+
     @staticmethod
     def check_scope(api_key: ApiKey, required_scope: str) -> bool:
         """检查权限范围"""

api/app/services/app_chat_service.py

@@ -16,7 +16,7 @@ from app.models import MultiAgentConfig, AgentConfig, ModelType
 from app.models import WorkflowConfig
 from app.repositories.tool_repository import ToolRepository
 from app.schemas import DraftRunRequest
-from app.schemas.app_schema import FileInput
+from app.schemas.app_schema import FileInput, FileType
 from app.schemas.model_schema import ModelInfo
 from app.schemas.prompt_schema import render_prompt_message, PromptMessageRole
 from app.services.conversation_service import ConversationService
@@ -26,6 +26,7 @@ from app.services.model_service import ModelApiKeyService
 from app.services.multi_agent_orchestrator import MultiAgentOrchestrator
 from app.services.multimodal_service import MultimodalService
 from app.services.workflow_service import WorkflowService
+from app.models.file_metadata_model import FileMetadata
 
 logger = get_business_logger()
 
@@ -106,22 +107,6 @@ class AppChatService:
         # 获取模型参数
         model_parameters = config.model_parameters
 
-        # 创建 LangChain Agent
-        agent = LangChainAgent(
-            model_name=api_key_obj.model_name,
-            api_key=api_key_obj.api_key,
-            provider=api_key_obj.provider,
-            api_base=api_key_obj.api_base,
-            is_omni=api_key_obj.is_omni,
-            temperature=model_parameters.get("temperature", 0.7),
-            max_tokens=model_parameters.get("max_tokens", 2000),
-            system_prompt=system_prompt,
-            tools=tools,
-            deep_thinking=model_parameters.get("deep_thinking", False),
-            thinking_budget_tokens=model_parameters.get("thinking_budget_tokens"),
-            capability=api_key_obj.capability or [],
-        )
-
         model_info = ModelInfo(
             model_name=api_key_obj.model_name,
             provider=api_key_obj.provider,
@@ -163,8 +148,39 @@ class AppChatService:
         processed_files = None
         if files:
             multimodal_service = MultimodalService(self.db, model_info)
-            processed_files = await multimodal_service.process_files(files)
+            fu_config = features_config.get("file_upload", {})
+            if hasattr(fu_config, "model_dump"):
+                fu_config = fu_config.model_dump()
+            doc_img_recognition = isinstance(fu_config, dict) and fu_config.get("document_image_recognition", False)
+            processed_files = await multimodal_service.process_files(
+                files, document_image_recognition=doc_img_recognition,
+                workspace_id=workspace_id
+            )
             logger.info(f"处理了 {len(processed_files)} 个文件")
+            if doc_img_recognition and "vision" in (api_key_obj.capability or []) and any(
+                f.type == FileType.DOCUMENT for f in files
+            ):
+                system_prompt += (
+                    "\n\n文档文字中包含图片位置标记如 [图片 第2页 第1张]: http://...，请在回答中用 Markdown 格式 ![图片描述](url) 展示对应图片。"
+                )
+
+        # 创建 LangChain Agent
+        agent = LangChainAgent(
+            model_name=api_key_obj.model_name,
+            api_key=api_key_obj.api_key,
+            provider=api_key_obj.provider,
+            api_base=api_key_obj.api_base,
+            is_omni=api_key_obj.is_omni,
+            temperature=model_parameters.get("temperature", 0.7),
+            max_tokens=model_parameters.get("max_tokens", 2000),
+            system_prompt=system_prompt,
+            tools=tools,
+            deep_thinking=model_parameters.get("deep_thinking", False),
+            thinking_budget_tokens=model_parameters.get("thinking_budget_tokens"),
+            json_output=model_parameters.get("json_output", False),
+            capability=api_key_obj.capability or [],
+        )
+
         # 为需要运行时上下文的工具注入上下文
         for t in tools:
             if hasattr(t, 'tool_instance') and hasattr(t.tool_instance, 'set_runtime_context'):
@@ -218,11 +234,29 @@ class AppChatService:
             "reasoning_content": result.get("reasoning_content")
         }
         if files:
+            local_ids = [f.upload_file_id for f in files
+                         if f.transfer_method.value == "local_file" and f.upload_file_id
+                         and (not f.name or not f.size)]
+            meta_map = {}
+            if local_ids:
+                rows = self.db.query(FileMetadata).filter(
+                    FileMetadata.id.in_(local_ids),
+                    FileMetadata.status == "completed"
+                ).all()
+                meta_map = {str(r.id): r for r in rows}
             for f in files:
-                # url = await MultimodalService(self.db).get_file_url(f)
+                name, size = f.name, f.size
+                if f.transfer_method.value == "local_file" and f.upload_file_id and (not name or not size):
+                    meta = meta_map.get(str(f.upload_file_id))
+                    if meta:
+                        name = name or meta.file_name
+                        size = size or meta.file_size
                 human_meta["files"].append({
                     "type": f.type,
-                    "url": f.url
+                    "url": f.url,
+                    "name": name,
+                    "size": size,
+                    "file_type": f.file_type,
                 })
 
         if processed_files:
@@ -283,7 +317,7 @@ class AppChatService:
             "suggested_questions": suggested_questions,
             "citations": filtered_citations,
             "audio_url": audio_url,
-            "audio_status": "pending"
+            "audio_status": "pending" if audio_url else None
         }
 
     async def agnet_chat_stream(
@@ -359,23 +393,6 @@ class AppChatService:
             # 获取模型参数
             model_parameters = config.model_parameters
 
-            # 创建 LangChain Agent
-            agent = LangChainAgent(
-                model_name=api_key_obj.model_name,
-                api_key=api_key_obj.api_key,
-                provider=api_key_obj.provider,
-                api_base=api_key_obj.api_base,
-                is_omni=api_key_obj.is_omni,
-                temperature=model_parameters.get("temperature", 0.7),
-                max_tokens=model_parameters.get("max_tokens", 2000),
-                system_prompt=system_prompt,
-                tools=tools,
-                streaming=True,
-                deep_thinking=model_parameters.get("deep_thinking", False),
-                thinking_budget_tokens=model_parameters.get("thinking_budget_tokens"),
-                capability=api_key_obj.capability or [],
-            )
-
             model_info = ModelInfo(
                 model_name=api_key_obj.model_name,
                 provider=api_key_obj.provider,
@@ -417,8 +434,40 @@ class AppChatService:
             processed_files = None
             if files:
                 multimodal_service = MultimodalService(self.db, model_info)
-                processed_files = await multimodal_service.process_files(files)
+                fu_config = features_config.get("file_upload", {})
+                if hasattr(fu_config, "model_dump"):
+                    fu_config = fu_config.model_dump()
+                doc_img_recognition = isinstance(fu_config, dict) and fu_config.get("document_image_recognition", False)
+                processed_files = await multimodal_service.process_files(
+                    files, document_image_recognition=doc_img_recognition,
+                    workspace_id=workspace_id
+                )
                 logger.info(f"处理了 {len(processed_files)} 个文件")
+                if doc_img_recognition and "vision" in (api_key_obj.capability or []) and any(
+                    f.type == FileType.DOCUMENT for f in files
+                ):
+                    from langchain.agents import create_agent
+                    system_prompt += (
+                        "\n\n文档文字中包含图片位置标记如 [图片 第2页 第1张]: http://...，请在回答中用 Markdown 格式 ![图片描述](url) 展示对应图片。"
+                    )
+
+            # 创建 LangChain Agent
+            agent = LangChainAgent(
+                model_name=api_key_obj.model_name,
+                api_key=api_key_obj.api_key,
+                provider=api_key_obj.provider,
+                api_base=api_key_obj.api_base,
+                is_omni=api_key_obj.is_omni,
+                temperature=model_parameters.get("temperature", 0.7),
+                max_tokens=model_parameters.get("max_tokens", 2000),
+                system_prompt=system_prompt,
+                tools=tools,
+                streaming=True,
+                deep_thinking=model_parameters.get("deep_thinking", False),
+                thinking_budget_tokens=model_parameters.get("thinking_budget_tokens"),
+                json_output=model_parameters.get("json_output", False),
+                capability=api_key_obj.capability or [],
+            )
 
             # 为需要运行时上下文的工具注入上下文
             for t in tools:
@@ -509,10 +558,29 @@ class AppChatService:
             }
 
             if files:
+                local_ids = [f.upload_file_id for f in files
+                             if f.transfer_method.value == "local_file" and f.upload_file_id
+                             and (not f.name or not f.size)]
+                meta_map = {}
+                if local_ids:
+                    rows = self.db.query(FileMetadata).filter(
+                        FileMetadata.id.in_(local_ids),
+                        FileMetadata.status == "completed"
+                    ).all()
+                    meta_map = {str(r.id): r for r in rows}
                 for f in files:
+                    name, size = f.name, f.size
+                    if f.transfer_method.value == "local_file" and f.upload_file_id and (not name or not size):
+                        meta = meta_map.get(str(f.upload_file_id))
+                        if meta:
+                            name = name or meta.file_name
+                            size = size or meta.file_size
                     human_meta["files"].append({
                         "type": f.type,
-                        "url": f.url
+                        "url": f.url,
+                        "name": name,
+                        "size": size,
+                        "file_type": f.file_type,
                     })
             if processed_files:
                 human_meta["history_files"] = {

api/app/services/app_dsl_service.py

@@ -14,12 +14,14 @@ from app.models.app_model import App, AppType
 from app.models.appshare_model import AppShare
 from app.models.app_release_model import AppRelease
 from app.models.knowledge_model import Knowledge
+from app.models.knowledgeshare_model import KnowledgeShare
 from app.models.models_model import ModelConfig
 from app.models.tool_model import ToolConfig as ToolConfigModel
 from app.models.skill_model import Skill
 from app.models.workflow_model import WorkflowConfig
 from app.services.workflow_service import WorkflowService
 from app.core.workflow.adapters.memory_bear.memory_bear_adapter import MemoryBearAdapter
+from app.core.workflow.nodes.enums import NodeType
 from app.models.memory_config_model import MemoryConfig as MemoryConfigModel
 
 
@@ -73,15 +75,14 @@ class AppDslService:
             AppType.MULTI_AGENT: "multi_agent_config",
             AppType.WORKFLOW: "workflow"
         }.get(app.type, "config")
-        config_data = self._enrich_release_config(app.type, release.config or {})
+        config_data = self._enrich_release_config(app.type, release.config or {}, release.default_model_config_id)
         dsl = {**meta, "app": app_meta, config_key: config_data}
         return yaml.dump(dsl, default_flow_style=False, allow_unicode=True), f"{release.name}_v{release.version_name}.yaml"
 
-    def _enrich_release_config(self, app_type: str, cfg: dict) -> dict:
+    def _enrich_release_config(self, app_type: str, cfg: dict, default_model_config_id=None) -> dict:
         if app_type == AppType.AGENT:
             enriched = {**cfg}
-            if "default_model_config_id" in cfg:
-                enriched["default_model_config_ref"] = self._model_ref(cfg["default_model_config_id"])
+            enriched["default_model_config_ref"] = self._model_ref(default_model_config_id)
             if "knowledge_retrieval" in cfg:
                 enriched["knowledge_retrieval"] = self._enrich_knowledge_retrieval(cfg["knowledge_retrieval"])
             if "tools" in cfg:
@@ -91,8 +92,7 @@ class AppDslService:
             return enriched
         if app_type == AppType.MULTI_AGENT:
             enriched = {**cfg}
-            if "default_model_config_id" in cfg:
-                enriched["default_model_config_ref"] = self._model_ref(cfg["default_model_config_id"])
+            enriched["default_model_config_ref"] = self._model_ref(default_model_config_id)
             if "master_agent_id" in cfg:
                 enriched["master_agent_ref"] = self._release_ref(cfg["master_agent_id"])
             if "sub_agents" in cfg:
@@ -229,8 +229,11 @@ class AppDslService:
         workspace_id: uuid.UUID,
         tenant_id: uuid.UUID,
         user_id: uuid.UUID,
+        app_id: Optional[uuid.UUID] = None,
     ) -> tuple[App, list[str]]:
-        """解析 DSL，创建应用及配置，返回 (new_app, warnings)"""
+        """解析 DSL，创建或覆盖应用配置，返回 (app, warnings)。
+        app_id 不为空时：校验类型一致后覆盖配置；为空时创建新应用。
+        """
         app_meta = dsl.get("app", {})
         app_type = app_meta.get("type")
         if app_type not in (AppType.AGENT, AppType.MULTI_AGENT, AppType.WORKFLOW):
@@ -239,6 +242,9 @@ class AppDslService:
         warnings: list[str] = []
         now = datetime.datetime.now()
 
+        if app_id is not None:
+            return self._overwrite_dsl(dsl, app_id, app_type, workspace_id, tenant_id, warnings, now)
+
         new_app = App(
             id=uuid.uuid4(),
             workspace_id=workspace_id,
@@ -258,11 +264,57 @@ class AppDslService:
         self.db.add(new_app)
         self.db.flush()
 
+        self._write_config(new_app.id, app_type, dsl, workspace_id, tenant_id, warnings, now, create=True)
+
+        self.db.commit()
+        self.db.refresh(new_app)
+        return new_app, warnings
+
+    def _overwrite_dsl(
+        self,
+        dsl: dict,
+        app_id: uuid.UUID,
+        app_type: str,
+        workspace_id: uuid.UUID,
+        tenant_id: uuid.UUID,
+        warnings: list,
+        now: datetime.datetime,
+    ) -> tuple[App, list[str]]:
+        """覆盖已有应用的配置，类型不一致时抛出异常"""
+        app = self.db.query(App).filter(
+            App.id == app_id,
+            App.workspace_id == workspace_id,
+            App.is_active.is_(True)
+        ).first()
+        if not app:
+            raise ResourceNotFoundException("应用", str(app_id))
+        if app.type != app_type:
+            raise BusinessException(
+                f"YAML 类型 '{app_type}' 与应用类型 '{app.type}' 不一致，无法导入",
+                BizCode.BAD_REQUEST
+            )
+
+        self._write_config(app_id, app_type, dsl, workspace_id, tenant_id, warnings, now, create=False)
+
+        self.db.commit()
+        self.db.refresh(app)
+        return app, warnings
+
+    def _write_config(
+        self,
+        app_id: uuid.UUID,
+        app_type: str,
+        dsl: dict,
+        workspace_id: uuid.UUID,
+        tenant_id: uuid.UUID,
+        warnings: list,
+        now: datetime.datetime,
+        create: bool,
+    ) -> None:
+        """写入（新建或覆盖）应用配置"""
         if app_type == AppType.AGENT:
             cfg = dsl.get("agent_config") or {}
-            self.db.add(AgentConfig(
-                id=uuid.uuid4(),
-                app_id=new_app.id,
+            fields = dict(
                 system_prompt=cfg.get("system_prompt"),
                 model_parameters=cfg.get("model_parameters"),
                 default_model_config_id=self._resolve_model(cfg.get("default_model_config_ref"), tenant_id, warnings),
@@ -272,16 +324,21 @@ class AppDslService:
                 tools=self._resolve_tools(cfg.get("tools", []), tenant_id, warnings),
                 skills=self._resolve_skills(cfg.get("skills", {}), tenant_id, warnings),
                 features=cfg.get("features", {}),
-                is_active=True,
-                created_at=now,
                 updated_at=now,
-            ))
+            )
+            if create:
+                self.db.add(AgentConfig(id=uuid.uuid4(), app_id=app_id, is_active=True, created_at=now, **fields))
+            else:
+                existing = self.db.query(AgentConfig).filter(AgentConfig.app_id == app_id).first()
+                if existing:
+                    for k, v in fields.items():
+                        setattr(existing, k, v)
+                else:
+                    self.db.add(AgentConfig(id=uuid.uuid4(), app_id=app_id, is_active=True, created_at=now, **fields))
 
         elif app_type == AppType.MULTI_AGENT:
             cfg = dsl.get("multi_agent_config") or {}
-            self.db.add(MultiAgentConfig(
-                id=uuid.uuid4(),
-                app_id=new_app.id,
+            fields = dict(
                 orchestration_mode=cfg.get("orchestration_mode", "collaboration"),
                 master_agent_name=cfg.get("master_agent_name"),
                 model_parameters=cfg.get("model_parameters"),
@@ -291,13 +348,24 @@ class AppDslService:
                 routing_rules=self._resolve_routing_rules(cfg.get("routing_rules"), warnings),
                 execution_config=cfg.get("execution_config", {}),
                 aggregation_strategy=cfg.get("aggregation_strategy", "merge"),
-                is_active=True,
-                created_at=now,
                 updated_at=now,
-            ))
+            )
+            if create:
+                self.db.add(MultiAgentConfig(id=uuid.uuid4(), app_id=app_id, is_active=True, created_at=now, **fields))
+            else:
+                existing = self.db.query(MultiAgentConfig).filter(MultiAgentConfig.app_id == app_id).first()
+                if existing:
+                    for k, v in fields.items():
+                        setattr(existing, k, v)
+                else:
+                    self.db.add(MultiAgentConfig(id=uuid.uuid4(), app_id=app_id, is_active=True, created_at=now, **fields))
 
         elif app_type == AppType.WORKFLOW:
-            adapter = MemoryBearAdapter(dsl)
+            raw_wf = dsl.get("workflow") or {}
+            raw_nodes = raw_wf.get("nodes") or []
+            resolved_nodes = self._resolve_workflow_nodes(raw_nodes, tenant_id, workspace_id, warnings)
+            resolved_dsl = {**dsl, "workflow": {**raw_wf, "nodes": resolved_nodes}}
+            adapter = MemoryBearAdapter(resolved_dsl)
             if not adapter.validate_config():
                 raise BusinessException("工作流配置格式无效", BizCode.BAD_REQUEST)
             result = adapter.parse_workflow()
@@ -305,21 +373,39 @@ class AppDslService:
                 warnings.append(f"[节点错误] {e.node_name or e.node_id}: {e.detail}")
             for w in result.warnings:
                 warnings.append(f"[节点警告] {w.node_name or w.node_id}: {w.detail}")
-            wf = dsl.get("workflow") or {}
-            WorkflowService(self.db).create_workflow_config(
-                app_id=new_app.id,
-                nodes=[n.model_dump() for n in result.nodes],
-                edges=[e.model_dump() for e in result.edges],
-                variables=[v.model_dump() for v in result.variables],
-                execution_config=wf.get("execution_config", {}),
-                features=wf.get("features", {}),
-                triggers=wf.get("triggers", []),
-                validate=False,
-            )
-
-        self.db.commit()
-        self.db.refresh(new_app)
-        return new_app, warnings
+            wf_service = WorkflowService(self.db)
+            if create:
+                wf_service.create_workflow_config(
+                    app_id=app_id,
+                    nodes=[n.model_dump() for n in result.nodes],
+                    edges=[e.model_dump() for e in result.edges],
+                    variables=[v.model_dump() for v in result.variables],
+                    execution_config=raw_wf.get("execution_config", {}),
+                    features=raw_wf.get("features", {}),
+                    triggers=raw_wf.get("triggers", []),
+                    validate=False,
+                )
+            else:
+                existing = self.db.query(WorkflowConfig).filter(WorkflowConfig.app_id == app_id).first()
+                if existing:
+                    existing.nodes = [n.model_dump() for n in result.nodes]
+                    existing.edges = [e.model_dump() for e in result.edges]
+                    existing.variables = [v.model_dump() for v in result.variables]
+                    existing.execution_config = raw_wf.get("execution_config", {})
+                    existing.features = raw_wf.get("features", {})
+                    existing.triggers = raw_wf.get("triggers", [])
+                    existing.updated_at = now
+                else:
+                    wf_service.create_workflow_config(
+                        app_id=app_id,
+                        nodes=[n.model_dump() for n in result.nodes],
+                        edges=[e.model_dump() for e in result.edges],
+                        variables=[v.model_dump() for v in result.variables],
+                        execution_config=raw_wf.get("execution_config", {}),
+                        features=raw_wf.get("features", {}),
+                        triggers=raw_wf.get("triggers", []),
+                        validate=False,
+                    )
 
     def _unique_app_name(self, name: str, workspace_id: uuid.UUID, app_type: AppType) -> str:
         """生成唯一应用名称，同时检查本空间自有应用和共享到本空间的应用"""
@@ -348,44 +434,98 @@ class AppDslService:
     def _resolve_model(self, ref: Optional[dict], tenant_id: uuid.UUID, warnings: list) -> Optional[uuid.UUID]:
         if not ref:
             return None
-        q = self.db.query(ModelConfig).filter(
-            ModelConfig.tenant_id == tenant_id,
-            ModelConfig.name == ref.get("name"),
-            ModelConfig.is_active.is_(True)
-        )
-        if ref.get("provider"):
-            q = q.filter(ModelConfig.provider == ref["provider"])
-        if ref.get("type"):
-            q = q.filter(ModelConfig.type == ref["type"])
-        m = q.first()
-        if not m:
-            warnings.append(f"模型 '{ref.get('name')}' 未匹配，已置空，请导入后手动配置")
-        return m.id if m else None
+        model_id = ref.get("id")
+        if model_id:
+            try:
+                model_uuid = uuid.UUID(str(model_id))
+                m = self.db.query(ModelConfig).filter(
+                    ModelConfig.id == model_uuid,
+                    ModelConfig.tenant_id == tenant_id,
+                    ModelConfig.is_active.is_(True)
+                ).first()
+                if m:
+                    return str(m.id)
+            except (ValueError, AttributeError):
+                pass
+        model_name = ref.get("name")
+        if model_name:
+            q = self.db.query(ModelConfig).filter(
+                ModelConfig.tenant_id == tenant_id,
+                ModelConfig.name == model_name,
+                ModelConfig.is_active.is_(True)
+            )
+            if ref.get("provider"):
+                q = q.filter(ModelConfig.provider == ref["provider"])
+            if ref.get("type"):
+                q = q.filter(ModelConfig.type == ref["type"])
+            m = q.first()
+            if m:
+                return str(m.id)
+            warnings.append(f"模型 '{model_name}' 未匹配，已置空，请导入后手动配置")
+        else:
+            warnings.append(f"模型 ID '{model_id}' 未匹配，已置空，请导入后手动配置")
+        return None
 
     def _resolve_kb(self, ref: Optional[dict], workspace_id: uuid.UUID, warnings: list) -> Optional[str]:
         if not ref:
             return None
-        kb = self.db.query(Knowledge).filter(
-            Knowledge.workspace_id == workspace_id,
-            Knowledge.name == ref.get("name")
-        ).first()
-        if not kb:
-            warnings.append(f"知识库 '{ref.get('name')}' 未匹配，已置空，请导入后手动配置")
-        return str(kb.id) if kb else None
+        kb_id = ref.get("id")
+        if kb_id:
+            try:
+                kb_uuid = uuid.UUID(str(kb_id))
+                kb_share = self.db.query(KnowledgeShare).filter(
+                    KnowledgeShare.target_workspace_id == workspace_id,
+                    KnowledgeShare.source_kb_id == kb_uuid
+                ).first()
+                if kb_share:
+                    kb = self.db.query(Knowledge).filter(
+                        Knowledge.id == kb_share.target_kb_id
+                    ).first()
+                    if kb and kb.status == 1:
+                        return str(kb_share.target_kb_id)
+                kb = self.db.query(Knowledge).filter(
+                    Knowledge.workspace_id == workspace_id,
+                    Knowledge.id == kb_uuid,
+                    Knowledge.status == 1
+                ).first()
+                if kb:
+                    return str(kb.id)
+            except (ValueError, AttributeError):
+                pass
+        warnings.append(f"知识库 '{kb_id}' 未匹配，已置空，请导入后手动配置")
+        return None
 
     def _resolve_tool(self, ref: Optional[dict], tenant_id: uuid.UUID, warnings: list) -> Optional[str]:
         if not ref:
             return None
-        q = self.db.query(ToolConfigModel).filter(
-            ToolConfigModel.tenant_id == tenant_id,
-            ToolConfigModel.name == ref.get("name")
-        )
-        if ref.get("tool_type"):
-            q = q.filter(ToolConfigModel.tool_type == ref["tool_type"])
-        t = q.first()
-        if not t:
-            warnings.append(f"工具 '{ref.get('name')}' 未匹配，已置空，请导入后手动配置")
-        return str(t.id) if t else None
+        tool_id = ref.get("id")
+        tool_name = ref.get("name")
+        if tool_id:
+            try:
+                tool_uuid = uuid.UUID(str(tool_id))
+                t = self.db.query(ToolConfigModel).filter(
+                    ToolConfigModel.id == tool_uuid,
+                    ToolConfigModel.tenant_id == tenant_id,
+                    ToolConfigModel.is_active.is_(True)
+                ).first()
+                if t:
+                    return str(t.id)
+            except (ValueError, AttributeError):
+                pass
+        if tool_name:
+            q = self.db.query(ToolConfigModel).filter(
+                ToolConfigModel.tenant_id == tenant_id,
+                ToolConfigModel.name == tool_name
+            )
+            if ref.get("tool_type"):
+                q = q.filter(ToolConfigModel.tool_type == ref["tool_type"])
+            t = q.first()
+            if t:
+                return str(t.id)
+            warnings.append(f"工具 '{tool_name}' 未匹配，已置空，请导入后手动配置")
+        else:
+            warnings.append(f"工具 '{tool_id}' 未匹配，已置空，请导入后手动配置")
+        return None
 
     def _resolve_release(self, ref: Optional[dict], warnings: list) -> Optional[uuid.UUID]:
         if not ref:
@@ -427,6 +567,88 @@ class AppDslService:
             result.append(entry)
         return result
 
+    def _resolve_workflow_nodes(self, nodes: list, tenant_id: uuid.UUID, workspace_id: uuid.UUID, warnings: list) -> list:
+        """解析工作流节点中的工具ID和知识库ID，匹配不到则清空配置"""
+        resolved_nodes = []
+        for node in nodes:
+            node_type = node.get("type")
+            config = dict(node.get("config") or {})
+            node_label = node.get("name") or node.get("id")
+            if node_type == NodeType.TOOL.value:
+                tool_id = config.get("tool_id")
+                if not tool_id:
+                    # tool_id 本身就是空，直接置空不重复 warning
+                    config["tool_id"] = None
+                    config["tool_parameters"] = {}
+                else:
+                    tool_ref = {}
+                    if isinstance(tool_id, str) and len(tool_id) >= 36:
+                        try:
+                            uuid.UUID(tool_id)
+                            tool_ref["id"] = tool_id
+                        except ValueError:
+                            tool_ref["name"] = tool_id
+                    else:
+                        tool_ref["name"] = tool_id
+                    resolved_tool_id = self._resolve_tool(tool_ref, tenant_id, [])
+                    if resolved_tool_id:
+                        config["tool_id"] = resolved_tool_id
+                    else:
+                        warnings.append(f"[{node_label}] 工具 '{tool_id}' 未匹配，已置空，请导入后手动配置")
+                        config["tool_id"] = None
+                        config["tool_parameters"] = {}
+            elif node_type == NodeType.KNOWLEDGE_RETRIEVAL.value:
+                knowledge_bases = config.get("knowledge_bases") or []
+                resolved_kbs = []
+                for kb in knowledge_bases:
+                    kb_id = kb.get("kb_id")
+                    if not kb_id:
+                        continue
+                    kb_ref = {}
+                    if isinstance(kb_id, str):
+                        try:
+                            uuid.UUID(kb_id)
+                            kb_ref["id"] = kb_id
+                        except ValueError:
+                            kb_ref["name"] = kb_id
+                    else:
+                        kb_ref["name"] = kb_id
+                    resolved_id = self._resolve_kb(kb_ref, workspace_id, [])
+                    if resolved_id:
+                        resolved_kbs.append({**kb, "kb_id": resolved_id})
+                    else:
+                        warnings.append(f"[{node_label}] 知识库 '{kb_id}' 未匹配，已移除，请导入后手动配置")
+                config["knowledge_bases"] = resolved_kbs
+            elif node_type in (NodeType.LLM.value, NodeType.QUESTION_CLASSIFIER.value, NodeType.PARAMETER_EXTRACTOR.value):
+                model_ref = config.get("model_id")
+                if model_ref:
+                    ref_dict = None
+                    if isinstance(model_ref, dict):
+                        ref_id = model_ref.get("id")
+                        ref_name = model_ref.get("name")
+                        if ref_id:
+                            ref_dict = {"id": ref_id}
+                        elif ref_name is not None:
+                            ref_dict = {"name": ref_name, "provider": model_ref.get("provider"), "type": model_ref.get("type")}
+                    elif isinstance(model_ref, str):
+                        try:
+                            uuid.UUID(model_ref)
+                            ref_dict = {"id": model_ref}
+                        except ValueError:
+                            ref_dict = {"name": model_ref}
+                    if ref_dict:
+                        resolved_model_id = self._resolve_model(ref_dict, tenant_id, warnings)
+                        if resolved_model_id:
+                            config["model_id"] = resolved_model_id
+                        else:
+                            warnings.append(f"[{node_label}] 模型未匹配，已置空，请导入后手动配置")
+                            config["model_id"] = None
+                    else:
+                        warnings.append(f"[{node_label}] 模型未匹配，已置空，请导入后手动配置")
+                        config["model_id"] = None
+            resolved_nodes.append({**node, "config": config})
+        return resolved_nodes
+
     def _resolve_knowledge_retrieval(self, kr: Optional[dict], workspace_id: uuid.UUID, warnings: list) -> Optional[dict]:
         if not kr:
             return kr

api/app/services/app_log_service.py

@@ -1,13 +1,17 @@
 """应用日志服务层"""
 import uuid
+import datetime as dt
 from typing import Optional, Tuple
-from datetime import datetime
 
+from sqlalchemy import select
 from sqlalchemy.orm import Session
 
 from app.core.logging_config import get_business_logger
+from app.models.app_model import AppType
 from app.models.conversation_model import Conversation, Message
+from app.models.workflow_model import WorkflowExecution
 from app.repositories.conversation_repository import ConversationRepository, MessageRepository
+from app.schemas.app_log_schema import AppLogMessage, AppLogNodeExecution
 
 logger = get_business_logger()
 
@@ -27,6 +31,8 @@ class AppLogService:
         page: int = 1,
         pagesize: int = 20,
         is_draft: Optional[bool] = None,
+        keyword: Optional[str] = None,
+        app_type: Optional[str] = None,
     ) -> Tuple[list[Conversation], int]:
         """
         查询应用日志会话列表
@@ -36,7 +42,9 @@ class AppLogService:
             workspace_id: 工作空间 ID
             page: 页码（从 1 开始）
             pagesize: 每页数量
-            is_draft: 是否草稿会话（None 表示不过滤）
+            is_draft: 是否草稿会话（None表示返回全部）
+            keyword: 搜索关键词（匹配消息内容）
+            app_type: 应用类型（WORKFLOW 时关键词将从 workflow_executions 搜索）
 
         Returns:
             Tuple[list[Conversation], int]: (会话列表，总数)
@@ -48,7 +56,9 @@ class AppLogService:
                 "workspace_id": str(workspace_id),
                 "page": page,
                 "pagesize": pagesize,
-                "is_draft": is_draft
+                "is_draft": is_draft,
+                "keyword": keyword,
+                "app_type": app_type,
             }
         )
 
@@ -57,8 +67,10 @@ class AppLogService:
             app_id=app_id,
             workspace_id=workspace_id,
             is_draft=is_draft,
+            keyword=keyword,
             page=page,
-            pagesize=pagesize
+            pagesize=pagesize,
+            app_type=app_type,
         )
 
         logger.info(
@@ -76,53 +88,325 @@ class AppLogService:
         self,
         app_id: uuid.UUID,
         conversation_id: uuid.UUID,
-        workspace_id: uuid.UUID
-    ) -> Conversation:
+        workspace_id: uuid.UUID,
+        app_type: str = AppType.AGENT
+    ) -> Tuple[Conversation, list, dict[str, list[AppLogNodeExecution]]]:
         """
-        查询会话详情（包含消息）
-
-        Args:
-            app_id: 应用 ID
-            conversation_id: 会话 ID
-            workspace_id: 工作空间 ID
+        查询会话详情
 
         Returns:
-            Conversation: 包含消息的会话对象
-
-        Raises:
-            ResourceNotFoundException: 当会话不存在时
+            Tuple[Conversation, list[AppLogMessage|Message], dict[str, list[AppLogNodeExecution]]]
         """
         logger.info(
             "查询应用日志会话详情",
             extra={
                 "app_id": str(app_id),
                 "conversation_id": str(conversation_id),
-                "workspace_id": str(workspace_id)
+                "workspace_id": str(workspace_id),
+                "app_type": app_type
             }
         )
 
-        # 查询会话
         conversation = self.conversation_repository.get_conversation_for_app_log(
             conversation_id=conversation_id,
             app_id=app_id,
             workspace_id=workspace_id
         )
 
-        # 查询消息（按时间正序）
-        messages = self.message_repository.get_messages_by_conversation(
-            conversation_id=conversation_id
-        )
-
-        # 将消息附加到会话对象
-        conversation.messages = messages
+        if app_type == AppType.WORKFLOW:
+            messages, node_executions_map = self._get_workflow_messages_and_nodes(conversation_id)
+        else:
+            messages = self.message_repository.get_messages_by_conversation(
+                conversation_id=conversation_id
+            )
+            node_executions_map = self._get_workflow_node_executions_with_map(
+                conversation_id, messages
+            )
 
         logger.info(
             "查询应用日志会话详情成功",
             extra={
                 "app_id": str(app_id),
                 "conversation_id": str(conversation_id),
-                "message_count": len(messages)
+                "message_count": len(messages),
+                "message_with_nodes_count": len(node_executions_map)
             }
         )
 
-        return conversation
+        return conversation, messages, node_executions_map
+
+    def _get_workflow_messages_and_nodes(
+        self,
+        conversation_id: uuid.UUID,
+    ) -> Tuple[list[AppLogMessage], dict[str, list[AppLogNodeExecution]]]:
+        """
+        工作流应用专用：从 workflow_executions 构建 messages 和节点日志。
+
+        每条 WorkflowExecution 对应一轮对话：
+          - user message：来自 execution.input_data（content 取 message 字段，files 放 meta_data）
+          - assistant message：来自 execution.output_data（失败时内容为错误信息）
+        开场白的 suggested_questions 合并到第一条 assistant message 的 meta_data 里。
+
+        Returns:
+            (messages 列表, node_executions_map)
+        """
+        stmt = (
+            select(WorkflowExecution)
+            .where(
+                WorkflowExecution.conversation_id == conversation_id,
+                WorkflowExecution.status.in_(["completed", "failed"])
+            )
+            .order_by(WorkflowExecution.started_at.asc())
+        )
+        executions = list(self.db.scalars(stmt).all())
+
+        # 查开场白：Message 表里 meta_data 含 suggested_questions 的第一条 assistant 消息
+        opening_stmt = (
+            select(Message)
+            .where(
+                Message.conversation_id == conversation_id,
+                Message.role == "assistant",
+            )
+            .order_by(Message.created_at.asc())
+            .limit(10)
+        )
+        early_messages = list(self.db.scalars(opening_stmt).all())
+        suggested_questions: list = []
+        for m in early_messages:
+            if isinstance(m.meta_data, dict) and "suggested_questions" in m.meta_data:
+                suggested_questions = m.meta_data.get("suggested_questions") or []
+                break
+
+        messages: list[AppLogMessage] = []
+        node_executions_map: dict[str, list[AppLogNodeExecution]] = {}
+
+        # 如果有开场白，作为第一条 assistant 消息插入
+        if suggested_questions or early_messages:
+            opening_msg = next(
+                (m for m in early_messages
+                 if isinstance(m.meta_data, dict) and "suggested_questions" in m.meta_data),
+                None
+            )
+            if opening_msg:
+                messages.append(AppLogMessage(
+                    id=opening_msg.id,
+                    conversation_id=conversation_id,
+                    role="assistant",
+                    content=opening_msg.content,
+                    status=None,
+                    meta_data={"suggested_questions": suggested_questions},
+                    created_at=opening_msg.created_at,
+                ))
+
+        for execution in executions:
+            started_at = execution.started_at or dt.datetime.now()
+            completed_at = execution.completed_at or started_at
+
+            # assistant message 的 id，同时作为 node_executions_map 的 key
+            assistant_msg_id = uuid.uuid5(execution.id, "assistant")
+
+            # --- user message（输入）---
+            input_data = execution.input_data or {}
+            input_content = input_data.get("message") or _extract_text(input_data)
+
+            # 跳过没有用户输入的 execution（如开场白触发的记录）
+            if not input_content or not input_content.strip():
+                continue
+
+            files = input_data.get("files") or []
+            user_msg = AppLogMessage(
+                id=uuid.uuid5(execution.id, "user"),
+                conversation_id=conversation_id,
+                role="user",
+                content=input_content,
+                meta_data={"files": files} if files else None,
+                created_at=started_at,
+            )
+            messages.append(user_msg)
+
+            # --- assistant message（输出）---
+            if execution.status == "completed":
+                output_content = _extract_text(execution.output_data)
+                meta = {"usage": execution.token_usage or {}, "elapsed_time": execution.elapsed_time}
+            else:
+                output_content = _extract_text(execution.output_data) or ""
+                meta = {"error": execution.error_message, "error_node_id": execution.error_node_id}
+
+            assistant_msg = AppLogMessage(
+                id=assistant_msg_id,
+                conversation_id=conversation_id,
+                role="assistant",
+                content=output_content,
+                status=execution.status,
+                meta_data=meta,
+                created_at=completed_at,
+            )
+            messages.append(assistant_msg)
+
+            # --- 节点执行记录，从 workflow_executions.output_data["node_outputs"] 读取 ---
+            execution_nodes = _build_nodes_from_output_data(execution.output_data)
+
+            if execution_nodes:
+                node_executions_map[str(assistant_msg_id)] = execution_nodes
+
+        return messages, node_executions_map
+
+    def _get_workflow_node_executions_with_map(
+        self,
+        conversation_id: uuid.UUID,
+        messages: list[Message]
+    ) -> dict[str, list[AppLogNodeExecution]]:
+        """
+        从 workflow_executions 表中提取节点执行记录，并按 assistant message 分组
+
+        Args:
+            conversation_id: 会话 ID
+            messages: 消息列表
+
+        Returns:
+            Tuple[list[AppLogNodeExecution], dict[str, list[AppLogNodeExecution]]]:
+                (所有节点执行记录列表, 按 message_id 分组的节点执行记录字典)
+        """
+        node_executions_map: dict[str, list[AppLogNodeExecution]] = {}
+
+        # 查询该会话关联的所有工作流执行记录（按时间正序）
+        stmt = select(WorkflowExecution).where(
+            WorkflowExecution.conversation_id == conversation_id,
+            WorkflowExecution.status.in_(["completed", "failed"])
+        ).order_by(WorkflowExecution.started_at.asc())
+
+        executions = self.db.scalars(stmt).all()
+
+        logger.info(
+            f"查询到 {len(executions)} 条工作流执行记录",
+            extra={
+                "conversation_id": str(conversation_id),
+                "execution_count": len(executions),
+                "execution_ids": [str(e.id) for e in executions]
+            }
+        )
+
+        # 筛选出 workflow 执行产生的 assistant 消息（排除开场白）
+        # workflow 结果的 meta_data 包含 usage，而开场白包含 suggested_questions
+        assistant_messages = [
+            m for m in messages
+            if m.role == "assistant" and m.meta_data and "usage" in m.meta_data
+        ]
+
+        # 通过时序匹配，将 execution 和 assistant message 关联
+        used_message_ids: set[str] = set()
+
+        for execution in executions:
+            # 构建节点执行记录列表，从 workflow_executions.output_data["node_outputs"] 读取
+            execution_nodes = _build_nodes_from_output_data(execution.output_data)
+
+            if not execution_nodes:
+                continue
+
+            # 失败的执行没有 assistant message，直接用 execution id 作为 key
+            if execution.status == "failed":
+                node_executions_map[f"execution_{str(execution.id)}"] = execution_nodes
+                continue
+
+            # completed：通过时序匹配关联到对应的 assistant message
+            # 逻辑：找 execution.started_at 之后最近的、未使用的 assistant message
+            best_msg = None
+            best_dt = None
+            for msg in assistant_messages:
+                msg_id_str = str(msg.id)
+                if msg_id_str in used_message_ids:
+                    continue
+                if msg.created_at and msg.created_at >= execution.started_at:
+                    delta = (msg.created_at - execution.started_at).total_seconds()
+                    if best_dt is None or delta < best_dt:
+                        best_dt = delta
+                        best_msg = msg
+
+            if not best_msg:
+                continue
+
+            msg_id_str = str(best_msg.id)
+            used_message_ids.add(msg_id_str)
+            node_executions_map[msg_id_str] = execution_nodes
+
+        return node_executions_map
+
+
+def _extract_text(data: Optional[dict]) -> str:
+    """从 workflow execution 的 input_data / output_data 中提取可读文本。
+
+    优先取 'text'、'content'、'output' 字段；若都没有则 JSON 序列化整个 dict。
+    """
+    if not data:
+        return ""
+    for key in ("message", "text", "content", "output", "result", "answer"):
+        if key in data and isinstance(data[key], str):
+            return data[key]
+    import json
+    return json.dumps(data, ensure_ascii=False)
+
+
+def _build_nodes_from_output_data(output_data: Optional[dict]) -> list[AppLogNodeExecution]:
+    """从 workflow_executions.output_data["node_outputs"] 构建节点执行记录列表。
+
+    output_data 结构：
+    {
+        "node_outputs": {
+            "<node_id>": {
+                "node_type": ...,
+                "node_name": ...,
+                "status": ...,
+                "input": ...,
+                "output": ...,
+                "elapsed_time": ...,
+                "token_usage": ...,
+                "error": ...,
+                "cycle_items": [...],
+                ...
+            }
+        },
+        "error": ...,
+        ...
+    }
+    """
+    if not output_data:
+        return []
+    node_outputs: dict = output_data.get("node_outputs") or {}
+    # 按 execution_order（节点执行时写入的单调递增序号）排序。
+    # PostgreSQL JSONB 不保证 key 顺序，不能依赖 dict 插入顺序；
+    # 缺失 execution_order 的历史数据退化到 0，保持在最前。
+    ordered_items = sorted(
+        node_outputs.items(),
+        key=lambda kv: (kv[1] or {}).get("execution_order", 0)
+        if isinstance(kv[1], dict) else 0
+    )
+    result = []
+    for node_id, node_data in ordered_items:
+        if not isinstance(node_data, dict):
+            continue
+        output = dict(node_data)
+        cycle_items = output.pop("cycle_items", None)
+        # 把已知的顶层字段剥离，剩余的作为 output
+        node_type = output.pop("node_type", "unknown")
+        node_name = output.pop("node_name", None)
+        status = output.pop("status", "completed")
+        error = output.pop("error", None)
+        inp = output.pop("input", None)
+        elapsed_time = output.pop("elapsed_time", None)
+        token_usage = output.pop("token_usage", None)
+        # execution_order 仅用于排序，不返回给前端
+        output.pop("execution_order", None)
+        result.append(AppLogNodeExecution(
+            node_id=node_id,
+            node_type=node_type,
+            node_name=node_name,
+            status=status,
+            error=error,
+            input=inp,
+            process=None,
+            output=output if output else None,
+            cycle_items=cycle_items,
+            elapsed_time=elapsed_time,
+            token_usage=token_usage,
+        ))
+    return result

api/app/services/app_service.py

@@ -1452,6 +1452,32 @@ class AppService:
         logger.debug("配置不存在，返回默认模板", extra={"app_id": str(app_id)})
         return self._create_default_agent_config(app_id)
 
+    def get_default_model_parameters(
+            self,
+            *,
+            app_id: uuid.UUID,
+    ) -> "ModelParameters":
+        """获取 Agent 默认模型参数（不修改数据库）
+
+        Args:
+            app_id: 应用ID
+
+        Returns:
+            ModelParameters: 默认模型参数
+        """
+        logger.info("获取 Agent 默认模型参数", extra={"app_id": str(app_id)})
+
+        app = self._get_app_or_404(app_id)
+
+        if app.type != "agent":
+            raise BusinessException("只有 Agent 类型应用支持 Agent 配置", BizCode.APP_TYPE_NOT_SUPPORTED)
+
+        from app.schemas.app_schema import ModelParameters
+        default_model_parameters = ModelParameters()
+
+        logger.info("获取 Agent 默认模型参数成功", extra={"app_id": str(app_id)})
+        return default_model_parameters
+
     def _create_default_agent_config(self, app_id: uuid.UUID) -> AgentConfig:
         """创建默认的 Agent 配置模板（不保存到数据库）
 

api/app/services/auth_service.py

@@ -1,3 +1,5 @@
+import uuid
+
 from sqlalchemy.orm import Session
 from typing import Optional, Tuple, Union
 import jwt
@@ -130,7 +132,7 @@ def register_user_with_invite(
     email: str,
     password: str,
     invite_token: str,
-    workspace_id: str,
+    workspace_id: uuid.UUID,
     username: Optional[str] = None,
 ) -> User:
     """
@@ -147,6 +149,7 @@ def register_user_with_invite(
     from app.schemas.user_schema import UserCreate
     from app.schemas.workspace_schema import InviteAcceptRequest
     from app.services import user_service, workspace_service
+    from app.repositories import workspace_repository as ws_repo
     from app.core.logging_config import get_business_logger
     
     logger = get_business_logger()
@@ -159,7 +162,8 @@ def register_user_with_invite(
             password=password,
             username=email.split('@')[0] if not username else username
         )
-        user = user_service.create_user(db=db, user=user_create)
+        workspace = ws_repo.get_workspace_by_id(db=db, workspace_id=workspace_id)
+        user = user_service.create_user(db=db, user=user_create, workspace=workspace)
         logger.info(f"用户创建成功: {user.email} (ID: {user.id})")
         
         # 接受工作空间邀请（此时用户已成为工作空间成员，并且会 commit）

api/app/services/conversation_service.py

@@ -544,7 +544,7 @@ class ConversationService:
                 api_key=api_key,
                 base_url=api_base,
                 is_omni=is_omni,
-                support_thinking="thinking" in (capability or []),
+                capability=capability,
             ),
             type=ModelType(model_type)
         )

api/app/services/draft_run_service.py

@@ -10,29 +10,29 @@ import time
 import uuid
 from typing import Any, AsyncGenerator, Dict, List, Optional
 
+from langchain.agents import create_agent
 from langchain.tools import tool
 from pydantic import BaseModel, Field
 from sqlalchemy import select
 from sqlalchemy.orm import Session
 
-from app.celery_app import celery_app
 from app.core.agent.agent_middleware import AgentMiddleware
 from app.core.agent.langchain_agent import LangChainAgent
 from app.core.config import settings
 from app.core.error_codes import BizCode
 from app.core.exceptions import BusinessException
 from app.core.logging_config import get_business_logger
+from app.core.memory.enums import SearchStrategy
+from app.core.memory.memory_service import MemoryService
 from app.core.rag.nlp.search import knowledge_retrieval
 from app.db import get_db_context
 from app.models import AgentConfig, ModelConfig
 from app.repositories.tool_repository import ToolRepository
-from app.schemas.app_schema import FileInput, Citation
+from app.schemas.app_schema import FileInput, Citation, FileType
 from app.schemas.model_schema import ModelInfo
 from app.schemas.prompt_schema import PromptMessageRole, render_prompt_message
-from app.services import task_service
 from app.services.conversation_service import ConversationService
 from app.services.langchain_tool_server import Search
-from app.services.memory_agent_service import MemoryAgentService
 from app.services.model_parameter_merger import ModelParameterMerger
 from app.services.model_service import ModelApiKeyService
 from app.services.multimodal_service import MultimodalService
@@ -107,38 +107,41 @@ def create_long_term_memory_tool(
         logger.info(f" 长期记忆工具被调用！question={question}, user={end_user_id}")
         try:
             with get_db_context() as db:
-                memory_content = asyncio.run(
-                    MemoryAgentService().read_memory(
-                        end_user_id=end_user_id,
-                        message=question,
-                        history=[],
-                        search_switch="2",
-                        config_id=config_id,
-                        db=db,
-                        storage_type=storage_type,
-                        user_rag_memory_id=user_rag_memory_id
-                    )
-                )
-                task = celery_app.send_task(
-                    "app.core.memory.agent.read_message",
-                    args=[end_user_id, question, [], "1", config_id, storage_type, user_rag_memory_id]
-                )
-                result = task_service.get_task_memory_read_result(task.id)
-                status = result.get("status")
-                logger.info(f"读取任务状态：{status}")
-                if memory_content:
-                    memory_content = memory_content['answer']
-            logger.info(f'用户ID：Agent:{end_user_id}')
-            logger.debug("调用长期记忆 API", extra={"question": question, "end_user_id": end_user_id})
+                memory_service = MemoryService(db, config_id, end_user_id)
+                search_result = asyncio.run(memory_service.read(question, SearchStrategy.QUICK))
 
-            logger.info(
-                "长期记忆检索成功",
-                extra={
-                    "end_user_id": end_user_id,
-                    "content_length": len(str(memory_content))
-                }
-            )
-            return f"检索到以下历史记忆：\n\n{memory_content}"
+            #     memory_content = asyncio.run(
+            #         MemoryAgentService().read_memory(
+            #             end_user_id=end_user_id,
+            #             message=question,
+            #             history=[],
+            #             search_switch="2",
+            #             config_id=config_id,
+            #             db=db,
+            #             storage_type=storage_type,
+            #             user_rag_memory_id=user_rag_memory_id
+            #         )
+            #     )
+            #     task = celery_app.send_task(
+            #         "app.core.memory.agent.read_message",
+            #         args=[end_user_id, question, [], "1", config_id, storage_type, user_rag_memory_id]
+            #     )
+            #     result = task_service.get_task_memory_read_result(task.id)
+            #     status = result.get("status")
+            #     logger.info(f"读取任务状态：{status}")
+            #     if memory_content:
+            #         memory_content = memory_content['answer']
+            # logger.info(f'用户ID：Agent:{end_user_id}')
+            # logger.debug("调用长期记忆 API", extra={"question": question, "end_user_id": end_user_id})
+            #
+            # logger.info(
+            #     "长期记忆检索成功",
+            #     extra={
+            #         "end_user_id": end_user_id,
+            #         "content_length": len(str(memory_content))
+            #     }
+            # )
+            return f"检索到以下历史记忆：\n\n{search_result.content}"
         except Exception as e:
             logger.error("长期记忆检索失败", extra={"error": str(e), "error_type": type(e).__name__})
             return f"记忆检索失败: {str(e)}"
@@ -472,11 +475,19 @@ class AgentRunService:
             features_config: Dict[str, Any],
             citations: List[Citation]
     ) -> List[Any]:
-        """根据 citation 开关决定是否返回引用来源"""
+        """根据 citation 开关决定是否返回引用来源，并根据 allow_download 附加下载链接"""
         citation_cfg = features_config.get("citation", {})
-        if isinstance(citation_cfg, dict) and citation_cfg.get("enabled"):
-            return [cit.model_dump() for cit in citations]
-        return []
+        if not (isinstance(citation_cfg, dict) and citation_cfg.get("enabled")):
+            return []
+        allow_download = citation_cfg.get("allow_download", False)
+        result = []
+        for cit in citations:
+            item = cit.model_dump() if hasattr(cit, "model_dump") else dict(cit)
+            if allow_download and item.get("document_id"):
+                from app.core.config import settings
+                item["download_url"] = f"{settings.FILE_LOCAL_SERVER_URL}/apps/citations/{item['document_id']}/download"
+            result.append(item)
+        return result
 
     async def run(
             self,
@@ -584,22 +595,6 @@ class AgentRunService:
                 )
                 tools.extend(memory_tools)
 
-            # 4. 创建 LangChain Agent
-            agent = LangChainAgent(
-                model_name=api_key_config["model_name"],
-                api_key=api_key_config["api_key"],
-                provider=api_key_config.get("provider", "openai"),
-                api_base=api_key_config.get("api_base"),
-                is_omni=api_key_config.get("is_omni", False),
-                temperature=effective_params.get("temperature", 0.7),
-                max_tokens=effective_params.get("max_tokens", 2000),
-                system_prompt=system_prompt,
-                tools=tools,
-                deep_thinking=effective_params.get("deep_thinking", False),
-                thinking_budget_tokens=effective_params.get("thinking_budget_tokens"),
-                capability=api_key_config.get("capability", []),
-            )
-
             # 5. 处理会话ID（创建或验证），新会话时写入开场白
             is_new_conversation = not conversation_id
             opening, suggested_questions = None, None
@@ -634,12 +629,46 @@ class AgentRunService:
 
             # 6. 处理多模态文件
             processed_files = None
+            has_doc_with_images = False
             if files:
-                # 获取 provider 信息
                 provider = api_key_config.get("provider", "openai")
                 multimodal_service = MultimodalService(self.db, model_info)
-                processed_files = await multimodal_service.process_files(files)
+                fu_config = features_config.get("file_upload", {})
+                if hasattr(fu_config, "model_dump"):
+                    fu_config = fu_config.model_dump()
+                doc_img_recognition = isinstance(fu_config, dict) and fu_config.get("document_image_recognition", False)
+                processed_files = await multimodal_service.process_files(
+                    files, document_image_recognition=doc_img_recognition,
+                    workspace_id=workspace_id
+                )
                 logger.info(f"处理了 {len(processed_files)} 个文件，provider={provider}")
+                capability = api_key_config.get("capability", [])
+                has_doc_with_images = (
+                    doc_img_recognition
+                    and "vision" in capability
+                    and any(f.type == FileType.DOCUMENT for f in files)
+                )
+            if has_doc_with_images:
+                system_prompt += (
+                    "\n\n文档文字中包含图片位置标记如 [图片 第2页 第1张]: http://...，请在回答中用 Markdown 格式 ![图片描述](url) 展示对应图片。"
+                )
+
+            agent = LangChainAgent(
+                model_name=api_key_config["model_name"],
+                api_key=api_key_config["api_key"],
+                provider=api_key_config.get("provider", "openai"),
+                api_base=api_key_config.get("api_base"),
+                is_omni=api_key_config.get("is_omni", False),
+                temperature=effective_params.get("temperature", 0.7),
+                max_tokens=effective_params.get("max_tokens", 2000),
+                system_prompt=system_prompt,
+                tools=tools,
+                deep_thinking=effective_params.get("deep_thinking", False),
+                thinking_budget_tokens=effective_params.get("thinking_budget_tokens"),
+                json_output=effective_params.get("json_output", False),
+                capability=api_key_config.get("capability", []),
+            )
+
             # 为需要运行时上下文的工具注入上下文
             for t in tools:
                 if hasattr(t, 'tool_instance') and hasattr(t.tool_instance, 'set_runtime_context'):
@@ -725,7 +754,7 @@ class AgentRunService:
                 ) if not sub_agent else [],
                 "citations": filtered_citations,
                 "audio_url": audio_url,
-                "audio_status": "pending"
+                "audio_status": "pending" if audio_url else None
             }
 
             logger.info(
@@ -839,23 +868,6 @@ class AgentRunService:
                                                                     user_rag_memory_id)
                 tools.extend(memory_tools)
 
-            # 4. 创建 LangChain Agent
-            agent = LangChainAgent(
-                model_name=api_key_config["model_name"],
-                api_key=api_key_config["api_key"],
-                provider=api_key_config.get("provider", "openai"),
-                api_base=api_key_config.get("api_base"),
-                is_omni=api_key_config.get("is_omni", False),
-                temperature=effective_params.get("temperature", 0.7),
-                max_tokens=effective_params.get("max_tokens", 2000),
-                system_prompt=system_prompt,
-                tools=tools,
-                streaming=True,
-                deep_thinking=effective_params.get("deep_thinking", False),
-                thinking_budget_tokens=effective_params.get("thinking_budget_tokens"),
-                capability=api_key_config.get("capability", []),
-            )
-
             # 5. 处理会话ID（创建或验证），新会话时写入开场白
             is_new_conversation = not conversation_id
             opening, suggested_questions = None, None
@@ -891,12 +903,48 @@ class AgentRunService:
 
             # 6. 处理多模态文件
             processed_files = None
+            has_doc_with_images = False
             if files:
-                # 获取 provider 信息
                 provider = api_key_config.get("provider", "openai")
                 multimodal_service = MultimodalService(self.db, model_info)
-                processed_files = await multimodal_service.process_files(files)
+                fu_config = features_config.get("file_upload", {})
+                if hasattr(fu_config, "model_dump"):
+                    fu_config = fu_config.model_dump()
+                doc_img_recognition = isinstance(fu_config, dict) and fu_config.get("document_image_recognition", False)
+                processed_files = await multimodal_service.process_files(
+                    files, document_image_recognition=doc_img_recognition,
+                    workspace_id=workspace_id
+                )
                 logger.info(f"处理了 {len(processed_files)} 个文件，provider={provider}")
+                capability = api_key_config.get("capability", [])
+                has_doc_with_images = (
+                    doc_img_recognition
+                    and "vision" in capability
+                    and any(f.type == FileType.DOCUMENT for f in files)
+                )
+            if has_doc_with_images:
+                system_prompt += (
+                    "\n\n文档文字中包含图片位置标记如 [图片 第2页 第1张]: http://...，请在回答中用 Markdown 格式 ![图片描述](url) 展示对应图片。"
+                )
+
+            # 创建 LangChain Agent
+            agent = LangChainAgent(
+                model_name=api_key_config["model_name"],
+                api_key=api_key_config["api_key"],
+                provider=api_key_config.get("provider", "openai"),
+                api_base=api_key_config.get("api_base"),
+                is_omni=api_key_config.get("is_omni", False),
+                temperature=effective_params.get("temperature", 0.7),
+                max_tokens=effective_params.get("max_tokens", 2000),
+                system_prompt=system_prompt,
+                tools=tools,
+                streaming=True,
+                deep_thinking=effective_params.get("deep_thinking", False),
+                thinking_budget_tokens=effective_params.get("thinking_budget_tokens"),
+                json_output=effective_params.get("json_output", False),
+                capability=api_key_config.get("capability", []),
+            )
+
             # 为需要运行时上下文的工具注入上下文
             for t in tools:
                 if hasattr(t, 'tool_instance') and hasattr(t.tool_instance, 'set_runtime_context'):
@@ -1299,10 +1347,30 @@ class AgentRunService:
                 "history_files": {}
             }
             if files:
+                from app.models.file_metadata_model import FileMetadata
+                local_ids = [f.upload_file_id for f in files
+                             if f.transfer_method.value == "local_file" and f.upload_file_id
+                             and (not f.name or not f.size)]
+                meta_map = {}
+                if local_ids:
+                    rows = self.db.query(FileMetadata).filter(
+                        FileMetadata.id.in_(local_ids),
+                        FileMetadata.status == "completed"
+                    ).all()
+                    meta_map = {str(r.id): r for r in rows}
                 for f in files:
+                    name, size = f.name, f.size
+                    if f.transfer_method.value == "local_file" and f.upload_file_id and (not name or not size):
+                        meta = meta_map.get(str(f.upload_file_id))
+                        if meta:
+                            name = name or meta.file_name
+                            size = size or meta.file_size
                     human_meta["files"].append({
                         "type": f.type,
-                        "url": f.url
+                        "url": f.url,
+                        "file_type": f.file_type,
+                        "name": name,
+                        "size": size
                     })
 
             # 保存 history_files，包含 provider 和 is_omni 信息

api/app/services/emotion_analytics_service.py

@@ -679,9 +679,9 @@ class EmotionAnalyticsService:
 
             # 查询用户的实体和标签
             query = """
-            MATCH (e:Entity)
+            MATCH (e:ExtractedEntity)
             WHERE e.end_user_id = $end_user_id
-            RETURN e.name as name, e.type as type
+            RETURN e.name as name, e.entity_type as type
             ORDER BY e.created_at DESC
             LIMIT 20
             """

api/app/services/implicit_memory_service.py

@@ -34,6 +34,7 @@ from app.schemas.implicit_memory_schema import (
     UserMemorySummary,
 )
 from app.schemas.memory_config_schema import MemoryConfig
+from app.services.memory_base_service import MIN_MEMORY_SUMMARY_COUNT
 from sqlalchemy.orm import Session
 
 logger = logging.getLogger(__name__)
@@ -379,12 +380,59 @@ class ImplicitMemoryService:
             raise
     
 
+    def _build_empty_profile(self) -> dict:
+        """构建 MemorySummary 不足时返回的固定空白画像数据"""
+        now_ms = int(datetime.utcnow().timestamp() * 1000)
+        insufficient = "Insufficient data for analysis"
+
+        def _empty_dimension(name: str) -> dict:
+            return {
+                "evidence": [insufficient],
+                "reasoning": f"No clear evidence found for {name} dimension",
+                "percentage": 0.0,
+                "dimension_name": name,
+                "confidence_level": 20,
+            }
+
+        def _empty_category(name: str) -> dict:
+            return {
+                "evidence": [insufficient],
+                "percentage": 25.0,
+                "category_name": name,
+                "trending_direction": None,
+            }
+
+        return {
+            "habits": [],
+            "portrait": {
+                "aesthetic": _empty_dimension("aesthetic"),
+                "creativity": _empty_dimension("creativity"),
+                "literature": _empty_dimension("literature"),
+                "technology": _empty_dimension("technology"),
+                "historical_trends": None,
+                "analysis_timestamp": now_ms,
+                "total_summaries_analyzed": 0,
+            },
+            "preferences": [],
+            "interest_areas": {
+                "art": _empty_category("art"),
+                "tech": _empty_category("tech"),
+                "music": _empty_category("music"),
+                "lifestyle": _empty_category("lifestyle"),
+                "analysis_timestamp": now_ms,
+                "total_summaries_analyzed": 0,
+            },
+        }
+
     async def generate_complete_profile(
         self,
         user_id: str
     ) -> dict:
         """生成完整的用户画像（包含所有4个模块）
         
+        需要该用户的 MemorySummary 节点数量 >= 5 才会真正调用 LLM 生成画像，
+        否则返回固定的空白画像数据。
+        
         Args:
             user_id: 用户ID
             
@@ -394,6 +442,16 @@ class ImplicitMemoryService:
         logger.info(f"生成完整用户画像: user={user_id}")
         
         try:
+            # 前置检查：查询该用户有效的 MemorySummary 节点数量（排除孤立节点）
+            from app.services.memory_base_service import MemoryBaseService
+            base_service = MemoryBaseService()
+            memory_summary_count = await base_service.get_valid_memory_summary_count(user_id)
+            logger.info(f"用户 MemorySummary 节点数量: {memory_summary_count} (user={user_id})")
+
+            if memory_summary_count < MIN_MEMORY_SUMMARY_COUNT:
+                logger.info(f"MemorySummary 数量不足 {MIN_MEMORY_SUMMARY_COUNT}（当前 {memory_summary_count}），返回空白画像: user={user_id}")
+                return self._build_empty_profile()
+
             # 并行调用4个分析方法
             preferences, portrait, interest_areas, habits = await asyncio.gather(
                 self.get_preference_tags(user_id=user_id),

api/app/services/knowledge_service.py

@@ -2,11 +2,13 @@ import uuid
 from sqlalchemy.orm import Session
 from app.models.user_model import User
 from app.models.knowledge_model import Knowledge
+from app.models.workspace_model import Workspace
+from app.models.models_model import ModelConfig
 from app.schemas.knowledge_schema import KnowledgeCreate, KnowledgeUpdate
 from app.repositories import knowledge_repository
 from app.core.logging_config import get_business_logger
+from app.models.models_model import ModelType
 
-# Obtain a dedicated logger for business logic
 business_logger = get_business_logger()
 
 
@@ -60,13 +62,47 @@ def create_knowledge(
         db: Session, knowledge: KnowledgeCreate, current_user: User
 ) -> Knowledge:
     business_logger.info(f"Create a knowledge base: {knowledge.name}, creator: {current_user.username}")
-    
+
     try:
         knowledge.created_by = current_user.id
         if knowledge.workspace_id is None:
             knowledge.workspace_id = current_user.current_workspace_id
         if knowledge.parent_id is None:
             knowledge.parent_id = knowledge.workspace_id
+
+        workspace = db.query(Workspace).filter(Workspace.id == knowledge.workspace_id).first()
+        if not workspace:
+            raise Exception(f"Workspace {knowledge.workspace_id} not found")
+
+        tenant_id = workspace.tenant_id
+
+        if not knowledge.embedding_id:
+            if not workspace.embedding:
+                raise Exception("工作空间未配置 Embedding 模型，请先完善工作空间配置后重试")
+            knowledge.embedding_id = workspace.embedding
+
+        if not knowledge.reranker_id:
+            if not workspace.rerank:
+                raise Exception("工作空间未配置 Rerank 模型，请先完善工作空间配置后重试")
+            knowledge.reranker_id = workspace.rerank
+
+        if not knowledge.llm_id:
+            if not workspace.llm:
+                raise Exception("工作空间未配置 LLM 模型，请先完善工作空间配置后重试")
+            knowledge.llm_id = workspace.llm
+
+        if not knowledge.image2text_id:
+            model = db.query(ModelConfig).filter(
+                ModelConfig.tenant_id == tenant_id,
+                ModelConfig.type.in_([ModelType.CHAT.value, ModelType.LLM.value]),
+                ModelConfig.capability.contains(["vision"]),
+                ModelConfig.is_active == True,
+            ).order_by(ModelConfig.created_at.desc()).first()
+            if not model:
+                raise Exception("租户下没有可用的视觉模型，创建知识库失败")
+            knowledge.image2text_id = model.id
+            business_logger.debug(f"Auto-bind image2text model: {model.id}")
+
         business_logger.debug(f"Start creating the knowledge base: {knowledge.name}")
         db_knowledge = knowledge_repository.create_knowledge(
             db=db, knowledge=knowledge

api/app/services/llm_router.py

@@ -415,9 +415,11 @@ class LLMRouter:
                 api_key=api_key_config.api_key,
                 base_url=api_key_config.api_base,
                 is_omni=api_key_config.is_omni,
-                support_thinking="thinking" in (api_key_config.capability or []),
-                temperature=0.3,
-                max_tokens=500
+                capability=api_key_config.capability,
+                extra_params={
+                    "temperature": 0.3,
+                    "max_tokens": 500
+                }
             )
             
             logger.debug(f"创建 LLM 实例 - Provider: {api_key_config.provider}, Model: {api_key_config.model_name}")

api/app/services/master_agent_router.py

@@ -393,7 +393,7 @@ class MasterAgentRouter:
                 api_key=api_key_config.api_key,
                 base_url=api_key_config.api_base,
                 is_omni=api_key_config.is_omni,
-                support_thinking="thinking" in (api_key_config.capability or []),
+                capability=api_key_config.capability,
                 extra_params = extra_params
             )
 

api/app/services/memory_agent_service.py

@@ -405,7 +405,7 @@ class MemoryAgentService:
             self,
             end_user_id: str,
             message: str,
-            history: List[Dict],
+            history: List[Dict],  # FIXME: unused parameter
             search_switch: str,
             config_id: Optional[uuid.UUID] | int,
             db: Session,
@@ -505,8 +505,8 @@ class MemoryAgentService:
                 initial_state = {
                     "messages": [HumanMessage(content=message)],
                     "search_switch": search_switch,
-                    "end_user_id": end_user_id
-                    , "storage_type": storage_type,
+                    "end_user_id": end_user_id,
+                    "storage_type": storage_type,
                     "user_rag_memory_id": user_rag_memory_id,
                     "memory_config": memory_config}
                 # 获取节点更新信息
@@ -642,6 +642,8 @@ class MemoryAgentService:
                     "answer": summary,
                     "intermediate_outputs": result
                 }
+
+            # TODO: redis search -> answer
         except Exception as e:
             # Ensure proper error handling and logging
             error_msg = f"Read operation failed: {str(e)}"
@@ -1280,7 +1282,7 @@ def get_end_user_connected_config(end_user_id: str, db: Session) -> Dict[str, An
     }
 
     logger.info(
-        f"Successfully retrieved connected config: memory_config_id={memory_config_id}, workspace_id={app.workspace_id}")
+        f"Successfully retrieved connected config: memory_config_id={memory_config_id}, workspace_id={end_user.workspace_id}")
     return result
 
 

api/app/services/memory_api_service.py

@@ -8,6 +8,9 @@ This service validates inputs and delegates to MemoryAgentService for core memor
 import uuid
 from typing import Any, Dict, Optional
 
+from sqlalchemy.orm import Session
+
+from app.celery_task_scheduler import scheduler
 from app.core.error_codes import BizCode
 from app.core.exceptions import BusinessException, ResourceNotFoundException
 from app.core.logging_config import get_logger
@@ -15,7 +18,6 @@ from app.models.app_model import App
 from app.models.end_user_model import EndUser
 from app.schemas.memory_config_schema import ConfigurationError
 from app.services.memory_agent_service import MemoryAgentService
-from sqlalchemy.orm import Session
 
 logger = get_logger(__name__)
 
@@ -124,7 +126,7 @@ class MemoryAPIService:
         except Exception as e:
             logger.warning(f"Failed to update memory_config_id for end_user {end_user_id}: {e}")
 
-    async def write_memory(
+    def write_memory(
             self,
             workspace_id: uuid.UUID,
             end_user_id: str,
@@ -133,27 +135,28 @@ class MemoryAPIService:
             storage_type: str = "neo4j",
             user_rag_memory_id: Optional[str] = None,
     ) -> Dict[str, Any]:
-        """Write memory with validation.
-        
+        """Submit a memory write task via Celery.
+
         Validates end_user exists and belongs to workspace, updates the end user's
-        memory_config_id, then delegates to MemoryAgentService.write_memory.
-        
+        memory_config_id, then dispatches write_message_task to Celery for async
+        processing with per-user fair locking.
+
         Args:
             workspace_id: Workspace ID for resource validation
-            end_user_id: End user identifier (used as end_user_id)
+            end_user_id: End user identifier
             message: Message content to store
             config_id: Memory configuration ID (required)
             storage_type: Storage backend (neo4j or rag)
             user_rag_memory_id: Optional RAG memory ID
-            
+
         Returns:
-            Dict with status and end_user_id
-            
+            Dict with task_id, status, and end_user_id
+
         Raises:
             ResourceNotFoundException: If end_user not found
-            BusinessException: If end_user not in authorized workspace or write fails
+            BusinessException: If validation fails
         """
-        logger.info(f"Writing memory for end_user: {end_user_id}, workspace: {workspace_id}")
+        logger.info(f"Submitting memory write for end_user: {end_user_id}, workspace: {workspace_id}")
 
         # Validate end_user exists and belongs to workspace
         self.validate_end_user(end_user_id, workspace_id)
@@ -161,9 +164,131 @@ class MemoryAPIService:
         # Update end user's memory_config_id
         self._update_end_user_config(end_user_id, config_id)
 
+        # Convert to message list format expected by write_message_task
+        messages = message if isinstance(message, list) else [{"role": "user", "content": message}]
+
+        # from app.tasks import write_message_task
+        # task = write_message_task.delay(
+        #     end_user_id,
+        #     messages,
+        #     config_id,
+        #     storage_type,
+        #     user_rag_memory_id or "",
+        # )
+        task_id = scheduler.push_task(
+            "app.core.memory.agent.write_message",
+            end_user_id,
+            {
+                "end_user_id": end_user_id,
+                "message": messages,
+                "config_id": config_id,
+                "storage_type": storage_type,
+                "user_rag_memory_id": user_rag_memory_id or ""
+            }
+        )
+
+        logger.info(f"Memory write task submitted, task_id={task_id} end_user_id={end_user_id}")
+
+        return {
+            "task_id": task_id,
+            "status": "QUEUED",
+            "end_user_id": end_user_id,
+        }
+
+    def read_memory(
+            self,
+            workspace_id: uuid.UUID,
+            end_user_id: str,
+            message: str,
+            search_switch: str = "0",
+            config_id: str = "",
+            storage_type: str = "neo4j",
+            user_rag_memory_id: Optional[str] = None,
+    ) -> Dict[str, Any]:
+        """Submit a memory read task via Celery.
+
+        Validates end_user exists and belongs to workspace, updates the end user's
+        memory_config_id, then dispatches read_message_task to Celery for async processing.
+
+        Args:
+            workspace_id: Workspace ID for resource validation
+            end_user_id: End user identifier
+            message: Query message
+            search_switch: Search mode (0=deep search with verification, 1=deep search, 2=fast search)
+            config_id: Memory configuration ID (required)
+            storage_type: Storage backend (neo4j or rag)
+            user_rag_memory_id: Optional RAG memory ID
+
+        Returns:
+            Dict with task_id, status, and end_user_id
+
+        Raises:
+            ResourceNotFoundException: If end_user not found
+            BusinessException: If validation fails
+        """
+        logger.info(f"Submitting memory read for end_user: {end_user_id}, workspace: {workspace_id}")
+
+        # Validate end_user exists and belongs to workspace
+        self.validate_end_user(end_user_id, workspace_id)
+
+        # Update end user's memory_config_id
+        self._update_end_user_config(end_user_id, config_id)
+
+        from app.tasks import read_message_task
+        task = read_message_task.delay(
+            end_user_id,
+            message,
+            [],  # history
+            search_switch,
+            config_id,
+            storage_type,
+            user_rag_memory_id or "",
+        )
+
+        logger.info(f"Memory read task submitted: task_id={task.id}, end_user_id={end_user_id}")
+
+        return {
+            "task_id": task.id,
+            "status": "PENDING",
+            "end_user_id": end_user_id,
+        }
+
+    async def write_memory_sync(
+            self,
+            workspace_id: uuid.UUID,
+            end_user_id: str,
+            message: str,
+            config_id: str,
+            storage_type: str = "neo4j",
+            user_rag_memory_id: Optional[str] = None,
+    ) -> Dict[str, Any]:
+        """Write memory synchronously (inline, no Celery).
+
+        Validates end_user, then calls MemoryAgentService.write_memory directly.
+        Blocks until the write completes. Use for cases where the caller needs
+        immediate confirmation.
+
+        Args:
+            workspace_id: Workspace ID for resource validation
+            end_user_id: End user identifier
+            message: Message content to store
+            config_id: Memory configuration ID (required)
+            storage_type: Storage backend (neo4j or rag)
+            user_rag_memory_id: Optional RAG memory ID
+
+        Returns:
+            Dict with status and end_user_id
+
+        Raises:
+            ResourceNotFoundException: If end_user not found
+            BusinessException: If write fails
+        """
+        logger.info(f"Writing memory (sync) for end_user: {end_user_id}, workspace: {workspace_id}")
+
+        self.validate_end_user(end_user_id, workspace_id)
+        self._update_end_user_config(end_user_id, config_id)
+
         try:
-            # Delegate to MemoryAgentService
-            # Convert string message to list[dict] format expected by MemoryAgentService
             messages = message if isinstance(message, list) else [{"role": "user", "content": message}]
             result = await MemoryAgentService().write_memory(
                 end_user_id=end_user_id,
@@ -174,11 +299,8 @@ class MemoryAPIService:
                 user_rag_memory_id=user_rag_memory_id or "",
             )
 
-            logger.info(f"Memory write successful for end_user: {end_user_id}")
+            logger.info(f"Memory write (sync) successful for end_user: {end_user_id}")
 
-            # result may be a string "success" or a dict with a "status" key
-            # Preserve the full dict so callers don't silently lose extra fields
-            # (e.g. error codes, metadata) returned by MemoryAgentService.
             if isinstance(result, dict):
                 return {
                     **result,
@@ -192,20 +314,17 @@ class MemoryAPIService:
 
         except ConfigurationError as e:
             logger.error(f"Memory configuration error for end_user {end_user_id}: {e}")
-            raise BusinessException(
-                message=str(e),
-                code=BizCode.MEMORY_CONFIG_NOT_FOUND
-            )
+            raise BusinessException(message=str(e), code=BizCode.MEMORY_CONFIG_NOT_FOUND)
         except BusinessException:
             raise
         except Exception as e:
-            logger.error(f"Memory write failed for end_user {end_user_id}: {e}")
+            logger.error(f"Memory write (sync) failed for end_user {end_user_id}: {e}")
             raise BusinessException(
                 message=f"Memory write failed: {str(e)}",
                 code=BizCode.MEMORY_WRITE_FAILED
             )
 
-    async def read_memory(
+    async def read_memory_sync(
             self,
             workspace_id: uuid.UUID,
             end_user_id: str,
@@ -215,37 +334,34 @@ class MemoryAPIService:
             storage_type: str = "neo4j",
             user_rag_memory_id: Optional[str] = None,
     ) -> Dict[str, Any]:
-        """Read memory with validation.
-        
-        Validates end_user exists and belongs to workspace, updates the end user's
-        memory_config_id, then delegates to MemoryAgentService.read_memory.
-        
+        """Read memory synchronously (inline, no Celery).
+
+        Validates end_user, then calls MemoryAgentService.read_memory directly.
+        Blocks until the read completes. Use for cases where the caller needs
+        the answer immediately.
+
         Args:
             workspace_id: Workspace ID for resource validation
-            end_user_id: End user identifier (used as end_user_id)
+            end_user_id: End user identifier
             message: Query message
             search_switch: Search mode (0=deep search with verification, 1=deep search, 2=fast search)
             config_id: Memory configuration ID (required)
             storage_type: Storage backend (neo4j or rag)
             user_rag_memory_id: Optional RAG memory ID
-            
+
         Returns:
             Dict with answer, intermediate_outputs, and end_user_id
-            
+
         Raises:
             ResourceNotFoundException: If end_user not found
-            BusinessException: If end_user not in authorized workspace or read fails
+            BusinessException: If read fails
         """
-        logger.info(f"Reading memory for end_user: {end_user_id}, workspace: {workspace_id}")
+        logger.info(f"Reading memory (sync) for end_user: {end_user_id}, workspace: {workspace_id}")
 
-        # Validate end_user exists and belongs to workspace
         self.validate_end_user(end_user_id, workspace_id)
-
-        # Update end user's memory_config_id
         self._update_end_user_config(end_user_id, config_id)
 
         try:
-            # Delegate to MemoryAgentService
             result = await MemoryAgentService().read_memory(
                 end_user_id=end_user_id,
                 message=message,
@@ -257,7 +373,7 @@ class MemoryAPIService:
                 user_rag_memory_id=user_rag_memory_id or ""
             )
 
-            logger.info(f"Memory read successful for end_user: {end_user_id}")
+            logger.info(f"Memory read (sync) successful for end_user: {end_user_id}")
 
             return {
                 "answer": result.get("answer", ""),
@@ -267,14 +383,11 @@ class MemoryAPIService:
 
         except ConfigurationError as e:
             logger.error(f"Memory configuration error for end_user {end_user_id}: {e}")
-            raise BusinessException(
-                message=str(e),
-                code=BizCode.MEMORY_CONFIG_NOT_FOUND
-            )
+            raise BusinessException(message=str(e), code=BizCode.MEMORY_CONFIG_NOT_FOUND)
         except BusinessException:
             raise
         except Exception as e:
-            logger.error(f"Memory read failed for end_user {end_user_id}: {e}")
+            logger.error(f"Memory read (sync) failed for end_user {end_user_id}: {e}")
             raise BusinessException(
                 message=f"Memory read failed: {str(e)}",
                 code=BizCode.MEMORY_READ_FAILED

api/app/services/memory_base_service.py

@@ -265,12 +265,50 @@ async def Translation_English(modid, text, fields=None):
     # 其他类型（数字、布尔值、None等）：原样返回
     else:
         return text
+# 隐性记忆画像生成所需的最低 MemorySummary 节点数量
+MIN_MEMORY_SUMMARY_COUNT = 5
+
+
 class MemoryBaseService:
     """记忆服务基类，提供共享的辅助方法"""
     
     def __init__(self):
         self.neo4j_connector = Neo4jConnector()
     
+    async def get_valid_memory_summary_count(
+        self,
+        end_user_id: str
+    ) -> int:
+        """获取用户有效的 MemorySummary 节点数量（排除孤立节点）。
+
+        只统计存在 DERIVED_FROM_STATEMENT 关系的 MemorySummary 节点。
+
+        Args:
+            end_user_id: 终端用户ID
+
+        Returns:
+            有效 MemorySummary 节点数量
+        """
+        try:
+            query = """
+            MATCH (n:MemorySummary)-[:DERIVED_FROM_STATEMENT]->(:Statement)
+            WHERE n.end_user_id = $end_user_id
+            RETURN count(DISTINCT n) as count
+            """
+            result = await self.neo4j_connector.execute_query(
+                query, end_user_id=end_user_id
+            )
+            count = result[0]["count"] if result and len(result) > 0 else 0
+            logger.debug(
+                f"有效 MemorySummary 节点数量: {count} (end_user_id={end_user_id})"
+            )
+            return count
+        except Exception as e:
+            logger.error(
+                f"获取有效 MemorySummary 数量失败: {str(e)}", exc_info=True
+            )
+            return 0
+    
     @staticmethod
     def parse_timestamp(timestamp_value) -> Optional[int]:
         """

api/app/services/memory_config_service.py

@@ -163,7 +163,7 @@ class MemoryConfigService:
 
     def load_memory_config(
             self,
-            config_id: Optional[UUID] = None,
+            config_id: UUID | str | int | None = None,
             workspace_id: Optional[UUID] = None,
             service_name: str = "MemoryConfigService",
     ) -> MemoryConfig:
@@ -187,16 +187,6 @@ class MemoryConfigService:
         """
         start_time = time.time()
 
-        config_logger.info(
-            "Starting memory configuration loading",
-            extra={
-                "operation": "load_memory_config",
-                "service": service_name,
-                "config_id": str(config_id) if config_id else None,
-                "workspace_id": str(workspace_id) if workspace_id else None,
-            },
-        )
-
         logger.info(f"Loading memory configuration from database: config_id={config_id}, workspace_id={workspace_id}")
 
         try:
@@ -236,11 +226,7 @@ class MemoryConfigService:
                     f"Configuration not found: config_id={config_id}, workspace_id={workspace_id}"
                 )
 
-            # Get workspace for the config
-            db_query_start = time.time()
             result = MemoryConfigRepository.get_config_with_workspace(self.db, memory_config.config_id)
-            db_query_time = time.time() - db_query_start
-            logger.info(f"[PERF] Config+Workspace query: {db_query_time:.4f}s")
 
             if not result:
                 raise ConfigurationError(

api/app/services/memory_dashboard_service.py

@@ -821,7 +821,7 @@ def get_rag_content(
         for document in documents:
             try:
                 kb = knowledge_repository.get_knowledge_by_id(db, document.kb_id)
-                if not kb:
+                if not (kb and kb.status == 1):
                     business_logger.warning(f"知识库不存在: kb_id={document.kb_id}")
                     continue
                 

api/app/services/memory_explicit_service.py

@@ -4,7 +4,7 @@
 处理显性记忆相关的业务逻辑，包括情景记忆和语义记忆的查询。
 """
 
-from typing import Any, Dict
+from typing import Any, Dict, Optional
 
 from app.core.logging_config import get_logger
 from app.services.memory_base_service import MemoryBaseService
@@ -104,7 +104,7 @@ class MemoryExplicitService(MemoryBaseService):
                    e.description AS core_definition
             ORDER BY e.name ASC
             """
-            
+
             semantic_result = await self.neo4j_connector.execute_query(
                 semantic_query, 
                 end_user_id=end_user_id
@@ -146,6 +146,209 @@ class MemoryExplicitService(MemoryBaseService):
             logger.error(f"获取显性记忆总览时出错: {str(e)}", exc_info=True)
             raise
 
+
+    async def get_episodic_memory_list(
+        self,
+        end_user_id: str,
+        page: int,
+        pagesize: int,
+        start_date: Optional[int] = None,
+        end_date: Optional[int] = None,
+        episodic_type: str = "all",
+    ) -> Dict[str, Any]:
+        """
+        获取情景记忆分页列表
+
+        Args:
+            end_user_id: 终端用户ID
+            page: 页码
+            pagesize: 每页数量
+            start_date: 开始时间戳（毫秒），可选
+            end_date: 结束时间戳（毫秒），可选
+            episodic_type: 情景类型筛选
+
+        Returns:
+            {
+                "total": int,          # 该用户情景记忆总数（不受筛选影响）
+                "items": [...],        # 当前页数据
+                "page": {
+                    "page": int,
+                    "pagesize": int,
+                    "total": int,      # 筛选后总数
+                    "hasnext": bool
+                }
+            }
+        """
+        try:
+            logger.info(
+                f"情景记忆分页查询: end_user_id={end_user_id}, "
+                f"start_date={start_date}, end_date={end_date}, "
+                f"episodic_type={episodic_type}, page={page}, pagesize={pagesize}"
+            )
+
+            # 1. 查询情景记忆总数（不受筛选条件限制）
+            total_all_query = """
+            MATCH (s:MemorySummary)
+            WHERE s.end_user_id = $end_user_id
+            RETURN count(s) AS total
+            """
+            total_all_result = await self.neo4j_connector.execute_query(
+                total_all_query, end_user_id=end_user_id
+            )
+            total_all = total_all_result[0]["total"] if total_all_result else 0
+
+            # 2. 构建筛选条件
+            where_clauses = ["s.end_user_id = $end_user_id"]
+            params = {"end_user_id": end_user_id}
+
+            # 时间戳筛选（毫秒时间戳转为 UTC ISO 字符串，使用 Neo4j datetime() 精确比较）
+            if start_date is not None and end_date is not None:
+                from datetime import datetime, timezone
+                start_dt = datetime.fromtimestamp(start_date / 1000, tz=timezone.utc)
+                end_dt = datetime.fromtimestamp(end_date / 1000, tz=timezone.utc)
+                # 开始时间取当天 UTC 00:00:00，结束时间取当天 UTC 23:59:59.999999
+                start_iso = start_dt.strftime("%Y-%m-%dT") + "00:00:00.000000"
+                end_iso = end_dt.strftime("%Y-%m-%dT") + "23:59:59.999999"
+            
+                where_clauses.append("datetime(s.created_at) >= datetime($start_iso) AND datetime(s.created_at) <= datetime($end_iso)")
+                params["start_iso"] = start_iso
+                params["end_iso"] = end_iso
+
+            # 类型筛选下推到 Cypher（兼容中英文）
+            if episodic_type != "all":
+                type_mapping = {
+                    "conversation": "对话",
+                    "project_work": "项目/工作",
+                    "learning": "学习",
+                    "decision": "决策",
+                    "important_event": "重要事件"
+                }
+                chinese_type = type_mapping.get(episodic_type)
+                if chinese_type:
+                    where_clauses.append(
+                        "(s.memory_type = $episodic_type OR s.memory_type = $chinese_type)"
+                    )
+                    params["episodic_type"] = episodic_type
+                    params["chinese_type"] = chinese_type
+                else:
+                    where_clauses.append("s.memory_type = $episodic_type")
+                    params["episodic_type"] = episodic_type
+
+            where_str = " AND ".join(where_clauses)
+
+            # 3. 查询筛选后的总数
+            count_query = f"""
+            MATCH (s:MemorySummary)
+            WHERE {where_str}
+            RETURN count(s) AS total
+            """
+            count_result = await self.neo4j_connector.execute_query(count_query, **params)
+            filtered_total = count_result[0]["total"] if count_result else 0
+
+            # 4. 查询分页数据
+            skip = (page - 1) * pagesize
+            data_query = f"""
+            MATCH (s:MemorySummary)
+            WHERE {where_str}
+            RETURN elementId(s) AS id,
+                s.name AS title,
+                s.memory_type AS memory_type,
+                s.content AS content,
+                s.created_at AS created_at
+            ORDER BY s.created_at DESC
+            SKIP $skip LIMIT $limit
+            """
+            params["skip"] = skip
+            params["limit"] = pagesize
+
+            result = await self.neo4j_connector.execute_query(data_query, **params)
+
+            # 5. 处理结果
+            items = []
+            if result:
+                for record in result:
+                    raw_created_at = record.get("created_at")
+                    created_at_timestamp = self.parse_timestamp(raw_created_at)
+                    items.append({
+                        "id": record["id"],
+                        "title": record.get("title") or "未命名",
+                        "memory_type": record.get("memory_type") or "其他",
+                        "content": record.get("content") or "",
+                        "created_at": created_at_timestamp
+                    })
+
+            # 6. 构建返回结果
+            return {
+                "total": total_all,
+                "items": items,
+                "page": {
+                    "page": page,
+                    "pagesize": pagesize,
+                    "total": filtered_total,
+                    "hasnext": (page * pagesize) < filtered_total
+                }
+            }
+
+        except Exception as e:
+            logger.error(f"情景记忆分页查询出错: {str(e)}", exc_info=True)
+            raise
+
+    async def get_semantic_memory_list(
+        self,
+        end_user_id: str
+    ) -> list:
+        """
+        获取语义记忆全量列表
+
+        Args:
+            end_user_id: 终端用户ID
+
+        Returns:
+            [
+                {
+                    "id": str,
+                    "name": str,
+                    "entity_type": str,
+                    "core_definition": str
+                }
+            ]
+        """
+        try:
+            logger.info(f"语义记忆列表查询: end_user_id={end_user_id}")
+
+            semantic_query = """
+            MATCH (e:ExtractedEntity)
+            WHERE e.end_user_id = $end_user_id
+            AND e.is_explicit_memory = true
+            RETURN elementId(e) AS id,
+                e.name AS name,
+                e.entity_type AS entity_type,
+                e.description AS core_definition
+            ORDER BY e.name ASC
+            """
+
+            result = await self.neo4j_connector.execute_query(
+                semantic_query, end_user_id=end_user_id
+            )
+
+            items = []
+            if result:
+                for record in result:
+                    items.append({
+                        "id": record["id"],
+                        "name": record.get("name") or "未命名",
+                        "entity_type": record.get("entity_type") or "未分类",
+                        "core_definition": record.get("core_definition") or ""
+                    })
+
+            logger.info(f"语义记忆列表查询成功: end_user_id={end_user_id}, total={len(items)}")
+
+            return items
+
+        except Exception as e:
+            logger.error(f"语义记忆列表查询出错: {str(e)}", exc_info=True)
+            raise
+
     async def get_explicit_memory_details(
         self,
         end_user_id: str,

api/app/services/memory_perceptual_service.py

@@ -233,7 +233,7 @@ class MemoryPerceptualService:
                     api_key=model_config.api_key,
                     base_url=model_config.api_base,
                     is_omni=model_config.is_omni,
-                    support_thinking="thinking" in (model_config.capability or []),
+                    capability=model_config.capability,
                 )
             )
         return llm, model_config

api/app/services/model_parameter_merger.py

@@ -47,7 +47,8 @@ class ModelParameterMerger:
             "n": 1,
             "stop": None,
             "deep_thinking": False,
-            "thinking_budget_tokens": None
+            "thinking_budget_tokens": None,
+            "json_output": False
         }
         
         # 合并参数：默认值 -> 模型配置 -> Agent 配置

api/app/services/model_service.py

@@ -125,9 +125,7 @@ class ModelConfigService:
                 api_key=api_key,
                 base_url=api_base,
                 is_omni=is_omni,
-                support_thinking="thinking" in (capability or []),
-                temperature=0.7,
-                max_tokens=100
+                capability=capability
             )
 
             # 根据模型类型选择不同的验证方式
@@ -371,6 +369,15 @@ class ModelConfigService:
                 raise BusinessException("模型名称已存在", BizCode.DUPLICATE_NAME)
 
         model = ModelConfigRepository.update(db, model_id, model_data, tenant_id=tenant_id)
+
+        # 同步更新关联 api_keys 的 capability 和 is_omni
+        if model_data.capability is not None or model_data.is_omni is not None:
+            for api_key in model.api_keys:
+                if model_data.capability is not None:
+                    api_key.capability = model_data.capability
+                if model_data.is_omni is not None:
+                    api_key.is_omni = model_data.is_omni
+
         db.commit()
         db.refresh(model)
         return model
@@ -729,10 +736,21 @@ class ModelApiKeyService:
     @staticmethod
     def delete_api_key(db: Session, api_key_id: uuid.UUID) -> bool:
         """删除API Key"""
-        if not ModelApiKeyRepository.get_by_id(db, api_key_id):
+        api_key = ModelApiKeyRepository.get_by_id(db, api_key_id)
+        if not api_key:
             raise BusinessException("API Key不存在", BizCode.NOT_FOUND)
 
+        model_config_ids = [mc.id for mc in api_key.model_configs]
+
         success = ModelApiKeyRepository.delete(db, api_key_id)
+
+        for model_config_id in model_config_ids:
+            model_config = ModelConfigRepository.get_by_id(db, model_config_id)
+            if model_config:
+                has_active_key = any(key.is_active for key in model_config.api_keys)
+                if not has_active_key and model_config.is_active:
+                    model_config.is_active = False
+
         db.commit()
         return success
 

api/app/services/multi_agent_orchestrator.py

@@ -2616,9 +2616,11 @@ class MultiAgentOrchestrator:
                 api_key=api_key_config.api_key,
                 base_url=api_key_config.api_base,
                 is_omni=api_key_config.is_omni,
-                support_thinking="thinking" in (api_key_config.capability or []),
-                temperature=0.7,  # 整合任务使用中等温度
-                max_tokens=2000
+                capability=api_key_config.capability,
+                extra_params={
+                    "temperature": 0.7,   # 整合任务使用中等温度
+                    "max_tokens": 2000
+                }
             )
 
             # 创建 LLM 实例
@@ -2795,10 +2797,12 @@ class MultiAgentOrchestrator:
                 api_key=api_key_config.api_key,
                 base_url=api_key_config.api_base,
                 is_omni=api_key_config.is_omni,
-                support_thinking="thinking" in (api_key_config.capability or []),
-                temperature=0.7,
-                max_tokens=2000,
-                extra_params={"streaming": True}  # 启用流式输出
+                capability=api_key_config.capability,
+                extra_params={
+                    "temperature": 0.7,
+                    "max_tokens": 2000,
+                    "streaming": True   # 启用流式输出
+                }
             )
 
             # 创建 LLM 实例

api/app/services/multimodal_service.py

@@ -24,6 +24,7 @@ import chardet
 import httpx
 import magic
 import openpyxl
+import uuid
 from docx import Document
 from sqlalchemy.orm import Session
 
@@ -94,7 +95,7 @@ class DashScopeFormatStrategy(MultimodalFormatStrategy):
         """通义千问文档格式"""
         return True, {
             "type": "text",
-            "text": f"<document name=\"{file_name}\">\n{text}\n</document>"
+            "text": f"<document name=\"{file_name}\">\n文档内容：\n{text}\n</document>"
         }
 
     async def format_audio(
@@ -166,6 +167,7 @@ class BedrockFormatStrategy(MultimodalFormatStrategy):
     async def format_document(self, file_name: str, text: str) -> tuple[bool, Dict[str, Any]]:
         """Bedrock/Anthropic 文档格式（需要 base64 编码）"""
         # Bedrock 文档需要 base64 编码
+        text = f"文档内容：\n{text}\n"
         text_bytes = text.encode('utf-8')
         base64_text = base64.b64encode(text_bytes).decode('utf-8')
 
@@ -222,7 +224,7 @@ class OpenAIFormatStrategy(MultimodalFormatStrategy):
         """OpenAI 文档格式"""
         return True, {
             "type": "text",
-            "text": f"<document name=\"{file_name}\">\n{text}\n</document>"
+            "text": f"<document name=\"{file_name}\">\n文档内容：\n{text}\n</document>"
         }
 
     async def format_audio(
@@ -344,6 +346,8 @@ class MultimodalService:
     async def process_files(
             self,
             files: Optional[List[FileInput]],
+            workspace_id: uuid.UUID = None,
+            document_image_recognition: bool = False,
     ) -> List[Dict[str, Any]]:
         """
         处理文件列表，返回 LLM 可用的格式
@@ -379,6 +383,36 @@ class MultimodalService:
                 elif file.type == FileType.DOCUMENT:
                     is_support, content = await self._process_document(file, strategy)
                     result.append(content)
+                    # 仅当开关开启且模型支持视觉时，才提取文档内嵌图片
+                    if document_image_recognition and "vision" in self.capability:
+                        img_infos = await self.extract_document_images(file)
+                        from app.models.workspace_model import Workspace as WorkspaceModel
+                        ws = self.db.query(WorkspaceModel).filter(WorkspaceModel.id == workspace_id).first()
+                        tenant_id = ws.tenant_id if ws else None
+                        img_result = []
+                        for img_info in img_infos:
+                            page = img_info["page"]
+                            index = img_info["index"]
+                            ext = img_info.get("ext", "png")
+                            try:
+                                _, img_url = await self._save_doc_image_to_storage(img_info["bytes"], ext, tenant_id, workspace_id)
+                                placeholder = f"第{page}页 第{index + 1}张" if page > 0 else f"第{index + 1}张"
+                                # 在文本内容中追加图片位置标记
+                                if result and result[-1].get("type") in ("text", "document"):
+                                    key = "text" if "text" in result[-1] else list(result[-1].keys())[-1]
+                                    result[-1][key] = result[-1].get(key, "") + f"\n[图片 {placeholder}]: {img_url}"
+                                # 将图片以视觉格式追加到消息内容中
+                                img_file = FileInput(
+                                    type=FileType.IMAGE,
+                                    transfer_method=TransferMethod.REMOTE_URL,
+                                    url=img_url,
+                                    file_type="image/png",
+                                )
+                                _, img_content = await self._process_image(img_file, strategy_class(img_file))
+                                img_result.append(img_content)
+                            except Exception as img_err:
+                                logger.warning(f"文档图片处理失败: {img_err}")
+                        result.extend(img_result)
                 elif file.type == FileType.AUDIO and "audio" in self.capability:
                     is_support, content = await self._process_audio(file, strategy)
                     result.append(content)
@@ -431,12 +465,8 @@ class MultimodalService:
         """
         处理文档文件（PDF、Word 等）
         
-        Args:
-            file: 文档文件输入
-            strategy: 格式化策略
-            
         Returns:
-            Dict: 根据 provider 返回不同格式的文档内容
+            仅返回文本内容（图片通过 process_files 中的额外步骤追加）
         """
         if file.transfer_method == TransferMethod.REMOTE_URL:
             return True, {
@@ -444,19 +474,57 @@ class MultimodalService:
                 "text": f"<document url=\"{file.url}\">\n{await self.extract_document_text(file)}\n</document>"
             }
         else:
-            # 本地文件，提取文本内容
             server_url = settings.FILE_LOCAL_SERVER_URL
             file.url = f"{server_url}/storage/permanent/{file.upload_file_id}"
             text = await self.extract_document_text(file)
             file_metadata = self.db.query(FileMetadata).filter(
                 FileMetadata.id == file.upload_file_id
             ).first()
-
             file_name = file_metadata.file_name if file_metadata else "unknown"
-
-            # 使用策略格式化文档
             return await strategy.format_document(file_name, text)
 
+    @staticmethod
+    async def _save_doc_image_to_storage(
+            img_bytes: bytes,
+            ext: str,
+            tenant_id: uuid.UUID,
+            workspace_id: uuid.UUID,
+    ) -> tuple[str, str]:
+        """
+        将文档内嵌图片保存到存储后端，写入 FileMetadata。
+
+        Returns:
+            (file_id_str, permanent_url)
+        """
+        from app.services.file_storage_service import FileStorageService, generate_file_key
+        from app.db import get_db_context
+
+        file_id = uuid.uuid4()
+        file_ext = f".{ext}" if not ext.startswith(".") else ext
+        content_type = f"image/{ext}"
+
+        file_key = generate_file_key(tenant_id, workspace_id, file_id, file_ext)
+        storage_svc = FileStorageService()
+        await storage_svc.storage.upload(file_key, img_bytes, content_type)
+
+        with get_db_context() as db:
+            meta = FileMetadata(
+                id=file_id,
+                tenant_id=tenant_id,
+                workspace_id=workspace_id,
+                file_key=file_key,
+                file_name=f"doc_image_{file_id}{file_ext}",
+                file_ext=file_ext,
+                file_size=len(img_bytes),
+                content_type=content_type,
+                status="completed",
+            )
+            db.add(meta)
+            db.commit()
+
+        url = f"{settings.FILE_LOCAL_SERVER_URL}/storage/permanent/{file_id}"
+        return str(file_id), url
+
     async def _process_audio(self, file: FileInput, strategy) -> tuple[bool, Dict[str, Any]]:
         """
         处理音频文件
@@ -582,6 +650,84 @@ class MultimodalService:
             logger.error(f"Failed to load file. - {e}")
             return "[Failed to load file.]"
 
+    async def extract_document_images(self, file: FileInput) -> list[dict]:
+        """
+        提取文档中的内嵌图片（支持 PDF 和 DOCX），附带位置信息。
+
+        Returns:
+            list[dict]: 每项包含:
+                - bytes: 图片二进制
+                - page: 所在页码（PDF 从 1 开始，DOCX 为 0）
+                - index: 该页/文档内的图片序号（从 0 开始）
+                - ext: 图片扩展名（如 png、jpeg）
+        """
+        try:
+            file_content = file.get_content()
+            if not file_content:
+                async with httpx.AsyncClient(timeout=30.0) as client:
+                    response = await client.get(file.url, follow_redirects=True)
+                    response.raise_for_status()
+                    file_content = response.content
+                    file.set_content(file_content)
+
+            file_mime_type = magic.from_buffer(file_content, mime=True)
+            if file_mime_type in PDF_MIME:
+                return self._extract_pdf_images(file_content)
+            elif self._is_word_file(file_content, file_mime_type):
+                return self._extract_docx_images(file_content)
+            return []
+        except Exception as e:
+            logger.error(f"提取文档图片失败: {e}")
+            return []
+
+    @staticmethod
+    def _extract_pdf_images(file_content: bytes) -> list[dict]:
+        """从 PDF 提取内嵌图片，附带页码和序号"""
+        images = []
+        try:
+            import fitz  # PyMuPDF
+            doc = fitz.open(stream=file_content, filetype="pdf")
+            for page_num, page in enumerate(doc, start=1):
+                for idx, img in enumerate(page.get_images(full=True)):
+                    xref = img[0]
+                    base_image = doc.extract_image(xref)
+                    images.append({
+                        "bytes": base_image["image"],
+                        "ext": base_image.get("ext", "png"),
+                        "page": page_num,
+                        "index": idx,
+                    })
+            doc.close()
+        except ImportError:
+            logger.warning("PyMuPDF 未安装，无法提取 PDF 图片，请执行: uv add pymupdf")
+        except Exception as e:
+            logger.error(f"提取 PDF 图片失败: {e}")
+        return images
+
+    @staticmethod
+    def _extract_docx_images(file_content: bytes) -> list[dict]:
+        """从 DOCX 提取内嵌图片，附带序号（DOCX 无页码概念，page 固定为 0）"""
+        images = []
+        try:
+            if file_content[:2] != b'PK':
+                return []
+            with zipfile.ZipFile(io.BytesIO(file_content)) as zf:
+                media_files = sorted(
+                    name for name in zf.namelist()
+                    if name.startswith("word/media/") and not name.endswith("/")
+                )
+                for idx, name in enumerate(media_files):
+                    ext = name.rsplit(".", 1)[-1].lower() if "." in name else "png"
+                    images.append({
+                        "bytes": zf.read(name),
+                        "ext": ext,
+                        "page": 0,
+                        "index": idx,
+                    })
+        except Exception as e:
+            logger.error(f"提取 DOCX 图片失败: {e}")
+        return images
+
     @staticmethod
     async def _extract_pdf_text(file_content: bytes) -> str:
         """提取 PDF 文本"""

api/app/services/prompt/prompt_optimizer_system.jinja2

@@ -34,7 +34,7 @@ Readability Guideline: Ensure optimized prompts have good readability and logica
 Constraint Handling Guideline: Do not mention variable-related limitations under the [Constraints] label.{% endraw %}{% endif %}
 
 Constraints
-Output Constraint: Must output in JSON format including the fields "prompt" and "desc".
+Output Constraint: Must output in JSON format including the string fields "prompt" and "desc".
 Content Constraint: Must not include any explanations, analyses, or additional comments.
 Language Constraint: Must use clear and concise language.
 {% if skill != true %}Completeness Constraint: Must fully define all missing elements (input details, output format, constraints, etc.).{% endif %}

api/app/services/prompt_optimizer_service.py

@@ -186,7 +186,7 @@ class PromptOptimizerService:
             api_key=api_config.api_key,
             base_url=api_config.api_base,
             is_omni=api_config.is_omni,
-            support_thinking="thinking" in (api_config.capability or []),
+            capability=api_config.capability,
         ), type=ModelType(model_config.type))
         try:
             prompt_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'prompt')
@@ -227,10 +227,20 @@ class PromptOptimizerService:
             content = getattr(chunk, "content", chunk)
             if not content:
                 continue
-            buffer += content
+            if isinstance(content, str):
+                buffer += content
+            elif isinstance(content, list):
+                for _ in content:
+                    buffer += _["text"]
+            else:
+                logger.error(f"Unsupported content type - {content}")
+                raise Exception("Unsupported content type")
             cache = buffer[:-20]
+            last_idx = 19
+            while cache and cache[-1] == '\\' and last_idx > 0:
+                cache = buffer[:-last_idx]
+                last_idx -= 1
 
-            # 尝试找到 "prompt": " 开始位置
             if prompt_finished:
                 continue
 
@@ -272,7 +282,7 @@ class PromptOptimizerService:
     def parser_prompt_variables(prompt: str):
         try:
             pattern = r'\{\{\s*([a-zA-Z_][a-zA-Z0-9_]*)\s*\}\}'
-            matches = re.findall(pattern, prompt)
+            matches = re.findall(pattern, str(prompt))
             variables = list(set(matches))
             return variables
         except Exception as e:

api/app/services/shared_chat_service.py

@@ -250,7 +250,8 @@ class SharedChatService:
             tools=tools,
             deep_thinking=model_parameters.get("deep_thinking", False),
             thinking_budget_tokens=model_parameters.get("thinking_budget_tokens"),
-            capability=api_key_obj.capability or [],
+            json_output=model_parameters.get("json_output", False),
+            capability=api_key_obj.capability,
         )
 
         # 加载历史消息
@@ -455,6 +456,7 @@ class SharedChatService:
                 streaming=True,
                 deep_thinking=model_parameters.get("deep_thinking", False),
                 thinking_budget_tokens=model_parameters.get("thinking_budget_tokens"),
+                json_output=model_parameters.get("json_output", False),
                 capability=api_key_obj.capability or [],
             )
 

api/app/services/tool_service.py

@@ -815,11 +815,12 @@ class ToolService:
                         "default": param_info.get("default")
                     })
                 
-                # 请求体参数
+                # 请求体参数 — _extract_request_body 返回 {"schema": {...}, "required": bool, ...}
                 request_body = operation.get("request_body")
                 if request_body:
-                    schema_props = request_body.get("schema", {}).get("properties", {})
-                    required_props = request_body.get("schema", {}).get("required", [])
+                    body_schema = request_body.get("schema", {})
+                    schema_props = body_schema.get("properties", {})
+                    required_props = body_schema.get("required", [])
                     
                     for prop_name, prop_schema in schema_props.items():
                         parameters.append({

api/app/services/user_memory_service.py

@@ -14,6 +14,7 @@ from pydantic import BaseModel, Field
 from sqlalchemy.orm import Session
 
 from app.core.logging_config import get_logger
+from app.core.memory.storage_services.extraction_engine.deduplication.deduped_and_disamb import _USER_PLACEHOLDER_NAMES
 from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
 from app.db import get_db_context
 from app.repositories.conversation_repository import ConversationRepository
@@ -21,7 +22,7 @@ from app.repositories.end_user_repository import EndUserRepository
 from app.repositories.neo4j.cypher_queries import Graph_Node_query
 from app.repositories.neo4j.neo4j_connector import Neo4jConnector
 from app.schemas.memory_episodic_schema import EmotionSubject, EmotionType, type_mapping
-from app.services.memory_base_service import MemoryBaseService
+from app.services.memory_base_service import MemoryBaseService, MIN_MEMORY_SUMMARY_COUNT
 from app.services.memory_config_service import MemoryConfigService
 from app.services.memory_perceptual_service import MemoryPerceptualService
 from app.services.memory_short_service import ShortService
@@ -398,12 +399,25 @@ class UserMemoryService:
                 }
             
             # 构建响应数据（转换时间为毫秒时间戳）
+            # 将 meta_data 中的 profile、knowledge_tags、behavioral_hints 平铺到顶层
+            meta = end_user_info_record.meta_data or {}
+
+            # profile 列表字段截断：只返回前 MAX_PROFILE_LIST_SIZE 条（按时间从新到旧）
+            MAX_PROFILE_LIST_SIZE = 5
+            profile = meta.get("profile")
+            if isinstance(profile, dict):
+                for key in ("role", "domain", "expertise", "interests"):
+                    if isinstance(profile.get(key), list):
+                        profile[key] = profile[key][:MAX_PROFILE_LIST_SIZE]
+
             response_data = {
                 "end_user_info_id": str(end_user_info_record.id),
                 "end_user_id": str(end_user_info_record.end_user_id),
                 "other_name": end_user_info_record.other_name,
                 "aliases": end_user_info_record.aliases,
-                "meta_data": end_user_info_record.meta_data,
+                "profile": profile,
+                "knowledge_tags": meta.get("knowledge_tags"),
+                "behavioral_hints": meta.get("behavioral_hints"),
                 "created_at": datetime_to_timestamp(end_user_info_record.created_at),
                 "updated_at": datetime_to_timestamp(end_user_info_record.updated_at)
             }
@@ -473,7 +487,7 @@ class UserMemoryService:
             allowed_fields = {'other_name', 'aliases', 'meta_data'}
             
             # 用户占位名称黑名单，不允许作为 other_name 或出现在 aliases 中
-            _user_placeholder_names = {'用户', '我', 'User', 'I'}
+            _user_placeholder_names = _USER_PLACEHOLDER_NAMES
             
             # 过滤 other_name：不允许设置为占位名称
             if 'other_name' in update_data and update_data['other_name'] and update_data['other_name'].strip() in _user_placeholder_names:
@@ -1500,7 +1514,7 @@ async def analytics_memory_types(
     2. 工作记忆 (WORKING_MEMORY) = 会话数量（通过 ConversationRepository.get_conversation_by_user_id 获取）
     3. 短期记忆 (SHORT_TERM_MEMORY) = /short_term 接口返回的问答对数量
     4. 显性记忆 (EXPLICIT_MEMORY) = 情景记忆 + 语义记忆（通过 MemoryBaseService.get_explicit_memory_count 获取）
-    5. 隐性记忆 (IMPLICIT_MEMORY) = Statement 节点数量的三分之一
+    5. 隐性记忆 (IMPLICIT_MEMORY) = MemorySummary 节点数量（需 >= MIN_MEMORY_SUMMARY_COUNT 才显示，否则为 0）
     6. 情绪记忆 (EMOTIONAL_MEMORY) = 情绪标签统计总数（通过 MemoryBaseService.get_emotional_memory_count 获取）
     7. 情景记忆 (EPISODIC_MEMORY) = memory_summary（通过 MemoryBaseService.get_episodic_memory_count 获取）
     8. 遗忘记忆 (FORGET_MEMORY) = 激活值低于阈值的节点数（通过 MemoryBaseService.get_forget_memory_count 获取）
@@ -1557,23 +1571,15 @@ async def analytics_memory_types(
             logger.warning(f"获取会话数量失败，工作记忆数量设为0: {str(e)}")
             work_count = 0
     
-    # 获取隐性记忆数量（基于 Statement 节点数量的三分之一）
+    # 获取隐性记忆数量（基于有关联关系的 MemorySummary 节点数量，需 >= MIN_MEMORY_SUMMARY_COUNT 才计入）
     implicit_count = 0
     if end_user_id:
         try:
-            # 查询 Statement 节点数量
-            query = """
-            MATCH (n:Statement)
-            WHERE n.end_user_id = $end_user_id
-            RETURN count(n) as count
-            """
-            result = await _neo4j_connector.execute_query(query, end_user_id=end_user_id)
-            statement_count = result[0]["count"] if result and len(result) > 0 else 0
-            # 取三分之一作为隐性记忆数量
-            implicit_count = round(statement_count / 3)
-            logger.debug(f"隐性记忆数量（Statement数量的1/3）: {implicit_count} (Statement总数={statement_count}, end_user_id={end_user_id})")
+            memory_summary_count = await base_service.get_valid_memory_summary_count(end_user_id)
+            implicit_count = memory_summary_count if memory_summary_count >= MIN_MEMORY_SUMMARY_COUNT else 0
+            logger.debug(f"隐性记忆数量（有效MemorySummary节点数）: {implicit_count} (有效MemorySummary总数={memory_summary_count}, end_user_id={end_user_id})")
         except Exception as e:
-            logger.warning(f"获取Statement数量失败，隐性记忆数量设为0: {str(e)}")
+            logger.warning(f"获取MemorySummary数量失败，隐性记忆数量设为0: {str(e)}")
             implicit_count = 0
     
     # 原有的基于行为习惯的统计方式（已注释）
@@ -1639,7 +1645,7 @@ async def analytics_memory_types(
         "WORKING_MEMORY": work_count,                             # 工作记忆（基于会话数量）
         "SHORT_TERM_MEMORY": short_term_count,                    # 短期记忆（基于问答对数量）
         "EXPLICIT_MEMORY": explicit_count,                        # 显性记忆（情景记忆 + 语义记忆）
-        "IMPLICIT_MEMORY": implicit_count,                        # 隐性记忆（Statement数量的1/3）
+        "IMPLICIT_MEMORY": implicit_count,                        # 隐性记忆（MemorySummary节点数，需>=MIN_MEMORY_SUMMARY_COUNT）
         "EMOTIONAL_MEMORY": emotion_count,                        # 情绪记忆（使用情绪标签统计）
         "EPISODIC_MEMORY": episodic_count,                        # 情景记忆
         "FORGET_MEMORY": forget_count                             # 遗忘记忆（激活值低于阈值）

api/app/services/user_service.py

@@ -8,6 +8,7 @@ from sqlalchemy.orm import Session
 import uuid
 
 from app.aioRedis import aio_redis_set, aio_redis_get, aio_redis_delete
+from app.models import Workspace
 from app.models.user_model import User
 from app.repositories import user_repository
 from app.schemas.user_schema import UserCreate
@@ -74,7 +75,7 @@ def create_initial_superuser(db: Session):
         )
 
 
-def create_user(db: Session, user: UserCreate) -> User:
+def create_user(db: Session, user: UserCreate, workspace: Workspace) -> User:
     business_logger.info(f"创建用户: {user.username}, email: {user.email}")
     
     try:
@@ -93,24 +94,9 @@ def create_user(db: Session, user: UserCreate) -> User:
         business_logger.debug(f"开始创建用户: {user.username}")
         hashed_password = get_password_hash(user.password)
         
-        # 获取默认租户（第一个活跃租户）
-        from app.repositories.tenant_repository import TenantRepository
-        tenant_repo = TenantRepository(db)
-        tenants = tenant_repo.get_tenants(skip=0, limit=1, is_active=True)
-        
-        if not tenants:
-            business_logger.error("系统中没有可用的租户")
-            raise BusinessException(
-                "系统配置错误：没有可用的租户", 
-                code=BizCode.TENANT_NOT_FOUND,
-                context={"username": user.username, "email": user.email}
-            )
-        
-        default_tenant = tenants[0]
-        
         new_user = user_repository.create_user(
             db=db, user=user, hashed_password=hashed_password, 
-            tenant_id=default_tenant.id, is_superuser=False
+            tenant_id=workspace.tenant_id, is_superuser=False
         )
 
         db.commit()
@@ -285,7 +271,7 @@ def activate_user(db: Session, user_id_to_activate: uuid.UUID, current_user: Use
     try:
         # 查找用户
         business_logger.debug(f"查找待激活用户: {user_id_to_activate}")
-        db_user = user_repository.get_user_by_id(db, user_id=user_id_to_activate)
+        db_user = user_repository.get_user_by_id_regardless_active(db, user_id=user_id_to_activate)
         if not db_user:
             business_logger.warning(f"用户不存在: {user_id_to_activate}")
             raise BusinessException("用户不存在", code=BizCode.USER_NOT_FOUND)

api/app/services/workflow_import_service.py

@@ -14,6 +14,7 @@ from app.core.exceptions import BusinessException
 from app.core.workflow.adapters.base_adapter import WorkflowImportResult, WorkflowParserResult
 from app.core.workflow.adapters.errors import UnsupportedPlatform, InvalidConfiguration
 from app.core.workflow.adapters.registry import PlatformAdapterRegistry
+from app.models.app_model import AppType
 from app.schemas import AppCreate
 from app.schemas.workflow_schema import WorkflowConfigCreate
 from app.services.app_service import AppService
@@ -86,11 +87,12 @@ class WorkflowImportService:
         if config is None:
             raise BusinessException("Configuration import timed out. Please try again.")
         config = json.loads(config)
+        unique_name = self.app_service._unique_app_name(name, workspace_id, AppType.WORKFLOW)
         app = self.app_service.create_app(
             user_id=user_id,
             workspace_id=workspace_id,
             data=AppCreate(
-                name=name,
+                name=unique_name,
                 description=description,
                 type="workflow",
                 workflow_config=WorkflowConfigCreate(

api/app/services/workflow_service.py

@@ -17,8 +17,9 @@ from app.core.workflow.executor import execute_workflow, execute_workflow_stream
 from app.core.workflow.nodes.enums import NodeType
 from app.core.workflow.validator import validate_workflow_config
 from app.db import get_db
+from sqlalchemy import select
 from app.models import App
-from app.models.workflow_model import WorkflowConfig, WorkflowExecution
+from app.models.workflow_model import WorkflowConfig, WorkflowExecution, WorkflowNodeExecution
 from app.repositories import knowledge_repository
 from app.repositories.workflow_repository import (
     WorkflowConfigRepository,
@@ -694,7 +695,8 @@ class WorkflowService:
             "nodes": config.nodes,
             "edges": config.edges,
             "variables": config.variables,
-            "execution_config": config.execution_config
+            "execution_config": config.execution_config,
+            "features": feature_configs
         }
 
         try:
@@ -772,9 +774,16 @@ class WorkflowService:
                 # 过滤 citations
                 citations = result.get("citations", [])
                 citation_cfg = feature_configs.get("citation", {})
-                filtered_citations = (
-                    citations if isinstance(citation_cfg, dict) and citation_cfg.get("enabled") else []
-                )
+                if isinstance(citation_cfg, dict) and citation_cfg.get("enabled"):
+                    allow_download = citation_cfg.get("allow_download", False)
+                    if allow_download:
+                        from app.core.config import settings
+                        for c in citations:
+                            if c.get("document_id"):
+                                c["download_url"] = f"{settings.FILE_LOCAL_SERVER_URL}/apps/citations/{c['document_id']}/download"
+                    filtered_citations = citations
+                else:
+                    filtered_citations = []
                 assistant_meta = {"usage": token_usage, "audio_url": None}
                 if filtered_citations:
                     assistant_meta["citations"] = filtered_citations
@@ -894,7 +903,8 @@ class WorkflowService:
             "nodes": config.nodes,
             "edges": config.edges,
             "variables": config.variables,
-            "execution_config": config.execution_config
+            "execution_config": config.execution_config,
+            "features": feature_configs
         }
 
         try:
@@ -909,6 +919,7 @@ class WorkflowService:
                 input_data["conv_messages"] = conv_messages
             init_message_length = len(input_data.get("conv_messages", []))
             message_id = uuid.uuid4()
+            _cycle_items: dict[str, list] = {}
 
             # 新会话时写入开场白
             is_new_conversation = init_message_length == 0
@@ -939,6 +950,15 @@ class WorkflowService:
                     memory_storage_type=storage_type,
                     user_rag_memory_id=user_rag_memory_id
             ):
+                event_type = event.get("event")
+                event_data = event.get("data", {})
+
+                if event_type == "cycle_item":
+                    cycle_id = event_data.get("cycle_id")
+                    if cycle_id not in _cycle_items:
+                        _cycle_items[cycle_id] = []
+                    _cycle_items[cycle_id].append(event_data)
+
                 if event.get("event") == "workflow_end":
                     status = event.get("data", {}).get("status")
                     token_usage = event.get("data", {}).get("token_usage", {}) or {}
@@ -957,7 +977,10 @@ class WorkflowService:
                                     for file in message["content"]:
                                         human_meta["files"].append({
                                             "type": file.get("type"),
-                                            "url": file.get("url")
+                                            "url": file.get("url"),
+                                            "file_type": file.get("origin_file_type"),
+                                            "name": file.get("name"),
+                                            "size": file.get("size")
                                         })
                             if message["role"] == "assistant":
                                 assistant_message = message["content"]
@@ -970,9 +993,16 @@ class WorkflowService:
                         # 过滤 citations
                         citations = event.get("data", {}).get("citations", [])
                         citation_cfg = feature_configs.get("citation", {})
-                        filtered_citations = (
-                            citations if isinstance(citation_cfg, dict) and citation_cfg.get("enabled") else []
-                        )
+                        if isinstance(citation_cfg, dict) and citation_cfg.get("enabled"):
+                            allow_download = citation_cfg.get("allow_download", False)
+                            if allow_download:
+                                from app.core.config import settings
+                                for c in citations:
+                                    if c.get("document_id"):
+                                        c["download_url"] = f"{settings.FILE_LOCAL_SERVER_URL}/apps/citations/{c['document_id']}/download"
+                            filtered_citations = citations
+                        else:
+                            filtered_citations = []
                         assistant_meta = {"usage": token_usage, "audio_url": None}
                         if filtered_citations:
                             assistant_meta["citations"] = filtered_citations
@@ -1000,6 +1030,18 @@ class WorkflowService:
                         )
                     else:
                         logger.error(f"unexpect workflow run status, status: {status}")
+                    # 把积累的 cycle_item 写入 workflow_executions.output_data["node_outputs"]
+                    if _cycle_items and execution.output_data:
+                        import copy
+                        new_output_data = copy.deepcopy(execution.output_data)
+                        node_outputs = new_output_data.setdefault("node_outputs", {})
+                        for cycle_node_id, items in _cycle_items.items():
+                            if cycle_node_id in node_outputs:
+                                node_outputs[cycle_node_id]["cycle_items"] = items
+                            else:
+                                node_outputs[cycle_node_id] = {"cycle_items": items}
+                        execution.output_data = new_output_data
+                        self.db.commit()
                 elif event.get("event") == "workflow_start":
                     event["data"]["message_id"] = str(message_id)
                 event = self._emit(public, event)