diff --git a/.github/workflows/sync-to-gitee.yml b/.github/workflows/sync-to-gitee.yml new file mode 100644 index 00000000..71ddf22a --- /dev/null +++ b/.github/workflows/sync-to-gitee.yml @@ -0,0 +1,36 @@ +name: Sync to Gitee + +on: + push: + branches: + - main # Production + - develop # Integration + - 'release/*' # Release preparation + - 'hotfix/*' # Urgent fixes + tags: + - '*' # All version tags (v1.0.0, etc.) + +jobs: + sync: + runs-on: ubuntu-latest + + steps: + - name: Checkout Source Code + uses: actions/checkout@v4 + with: + fetch-depth: 0 + + - name: Sync to Gitee + run: | + GITEE_URL="https://${{ secrets.GITEE_USERNAME }}:${{ secrets.GITEE_TOKEN }}@gitee.com/hangzhou-hongxiong-intelligent_1/MemoryBear.git" + git remote add gitee "$GITEE_URL" + + # 遍历并推送所有分支 + for branch in $(git branch -r | grep -v HEAD | sed 's/origin\///'); do + echo "Syncing branch: $branch" + git push -f gitee "origin/$branch:refs/heads/$branch" + done + + # 推送所有标签 + echo "Syncing tags..." + git push gitee --tags --force diff --git a/.gitignore b/.gitignore index ae3261f0..0ec6822c 100644 --- a/.gitignore +++ b/.gitignore @@ -18,6 +18,7 @@ examples/ .kiro .vscode .idea +.claude # Temporary outputs .DS_Store diff --git a/README.md b/README.md index 95d8d737..cd5be68b 100644 --- a/README.md +++ b/README.md @@ -2,6 +2,10 @@ # MemoryBear empowers AI with human-like memory capabilities +[![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](LICENSE) +[![Python](https://img.shields.io/badge/Python-3.12+-green?logo=python&logoColor=white)](https://www.python.org/) +[![Gitee Sync](https://img.shields.io/github/actions/workflow/status/SuanmoSuanyangTechnology/MemoryBear/sync-to-gitee.yml?label=Gitee%20Sync&logo=gitee&logoColor=white)](https://github.com/SuanmoSuanyangTechnology/MemoryBear/actions/workflows/sync-to-gitee.yml) + [中文](./README_CN.md) | English ### [Installation Guide](#memorybear-installation-guide) diff --git a/README_CN.md b/README_CN.md index 1472acac..31ea718f 100644 --- a/README_CN.md +++ b/README_CN.md @@ -2,6 +2,10 @@ # MemoryBear 让AI拥有如同人类一样的记忆 +[![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](LICENSE) +[![Python](https://img.shields.io/badge/Python-3.12+-green?logo=python&logoColor=white)](https://www.python.org/) +[![Gitee Sync](https://img.shields.io/github/actions/workflow/status/SuanmoSuanyangTechnology/MemoryBear/sync-to-gitee.yml?label=Gitee%20Sync&logo=gitee&logoColor=white)](https://github.com/SuanmoSuanyangTechnology/MemoryBear/actions/workflows/sync-to-gitee.yml) + 中文 | [English](./README.md) ### [安装教程](#memorybear安装教程) diff --git a/api/app/controllers/home_page_controller.py b/api/app/controllers/home_page_controller.py index de4a78a3..400d155a 100644 --- a/api/app/controllers/home_page_controller.py +++ b/api/app/controllers/home_page_controller.py @@ -3,9 +3,10 @@ from sqlalchemy.orm import Session from app.core.config import settings from app.core.response_utils import success -from app.db import get_db +from app.db import get_db, SessionLocal from app.dependencies import get_current_user from app.models.user_model import User +from app.repositories.home_page_repository import HomePageRepository from app.schemas.response_schema import ApiResponse from app.services.home_page_service import HomePageService @@ -31,9 +32,32 @@ def get_workspace_list( @router.get("/version", response_model=ApiResponse) def get_system_version(): - """获取系统版本号+说明""" - current_version = settings.SYSTEM_VERSION - version_info = HomePageService.load_version_introduction(current_version) + """获取系统版本号 + 说明""" + current_version = None + version_info = None + + # 1️⃣ 优先从数据库获取最新已发布的版本 + try: + db = SessionLocal() + try: + current_version, version_info = HomePageRepository.get_latest_version_introduction(db) + finally: + db.close() + except Exception as e: + pass + + # 2️⃣ 降级：使用环境变量中的版本号 + if not current_version: + current_version = settings.SYSTEM_VERSION + version_info = HomePageService.load_version_introduction(current_version) + + # 3️⃣ 如果数据库和 JSON 都没有，返回基本信息 + if not version_info: + version_info = { + "introduction": {"codeName": "", "releaseDate": "", "upgradePosition": "", "coreUpgrades": []}, + "introduction_en": {"codeName": "", "releaseDate": "", "upgradePosition": "", "coreUpgrades": []} + } + return success( data={ "version": current_version, diff --git a/api/app/controllers/knowledge_controller.py b/api/app/controllers/knowledge_controller.py index 74b832cd..afda7cce 100644 --- a/api/app/controllers/knowledge_controller.py +++ b/api/app/controllers/knowledge_controller.py @@ -352,6 +352,7 @@ async def delete_knowledge( # 2. Soft-delete knowledge base api_logger.debug(f"Perform a soft delete: {db_knowledge.name} (ID: {knowledge_id})") db_knowledge.status = 2 + db_knowledge.updated_at = datetime.datetime.now() db.commit() api_logger.info(f"The knowledge base has been successfully deleted: {db_knowledge.name} (ID: {knowledge_id})") return success(msg="The knowledge base has been successfully deleted") diff --git a/api/app/controllers/memory_dashboard_controller.py b/api/app/controllers/memory_dashboard_controller.py index bedee987..525fe1eb 100644 --- a/api/app/controllers/memory_dashboard_controller.py +++ b/api/app/controllers/memory_dashboard_controller.py @@ -591,7 +591,7 @@ async def dashboard_data( "total_api_call": None } - # 1. 获取记忆总量（total_memory） + # 1. 获取记忆总量（total_memory）—— neo4j 独有逻辑：查询 neo4j 存储节点 try: total_memory_data = await memory_dashboard_service.get_workspace_total_memory_count( db=db, @@ -600,49 +600,33 @@ async def dashboard_data( end_user_id=end_user_id ) neo4j_data["total_memory"] = total_memory_data.get("total_memory_count", 0) - # total_app: 统计当前空间下的所有app数量 - # 包含自有app + 被分享给本工作空间的app - from app.services import app_service as _app_svc - _, total_app = _app_svc.AppService(db).list_apps( - workspace_id=workspace_id, include_shared=True, pagesize=1 - ) - neo4j_data["total_app"] = total_app - api_logger.info(f"成功获取记忆总量: {neo4j_data['total_memory']}, 应用数量: {neo4j_data['total_app']}") + api_logger.info(f"成功获取记忆总量: {neo4j_data['total_memory']}") except Exception as e: api_logger.warning(f"获取记忆总量失败: {str(e)}") - # 2. 获取知识库类型统计（total_knowledge） - try: - from app.services.memory_agent_service import MemoryAgentService - memory_agent_service = MemoryAgentService() - knowledge_stats = await memory_agent_service.get_knowledge_type_stats( - end_user_id=end_user_id, - only_active=True, - current_workspace_id=workspace_id, - db=db - ) - neo4j_data["total_knowledge"] = knowledge_stats.get("total", 0) - api_logger.info(f"成功获取知识库类型统计total: {neo4j_data['total_knowledge']}") - except Exception as e: - api_logger.warning(f"获取知识库类型统计失败: {str(e)}") + # 2. 获取共享统计数据（total_app、total_knowledge、total_api_call） + common_stats = memory_dashboard_service.get_dashboard_common_stats(db, workspace_id) + neo4j_data.update(common_stats) + api_logger.info(f"成功获取共享统计: app={common_stats['total_app']}, knowledge={common_stats['total_knowledge']}, api_call={common_stats['total_api_call']}") - # 3. 获取API调用统计（total_api_call） + # 计算昨日对比 try: - # 使用 AppStatisticsService 获取真实的API调用统计 - app_stats_service = AppStatisticsService(db) - api_stats = app_stats_service.get_workspace_api_statistics( + changes = memory_dashboard_service.get_dashboard_yesterday_changes( + db=db, workspace_id=workspace_id, - start_date=start_date, - end_date=end_date + storage_type=storage_type, + today_data=neo4j_data ) - # 计算总调用次数 - total_api_calls = sum(item.get("total_calls", 0) for item in api_stats) - neo4j_data["total_api_call"] = total_api_calls - api_logger.info(f"成功获取API调用统计: {neo4j_data['total_api_call']}") + neo4j_data.update(changes) except Exception as e: - api_logger.error(f"获取API调用统计失败: {str(e)}") - neo4j_data["total_api_call"] = 0 - + api_logger.warning(f"计算neo4j昨日对比失败: {str(e)}") + neo4j_data.update({ + "total_memory_change": None, + "total_app_change": None, + "total_knowledge_change": None, + "total_api_call_change": None, + }) + result["neo4j_data"] = neo4j_data api_logger.info("成功获取neo4j_data") @@ -655,44 +639,37 @@ async def dashboard_data( "total_api_call": None } - # 获取RAG相关数据 + # 1. 获取记忆总量（total_memory）—— rag 独有逻辑：查询 document 表的 chunk_num try: - # total_memory: 只统计用户知识库（permission_id='Memory'）的chunk数 total_chunk = memory_dashboard_service.get_rag_user_kb_total_chunk(db, current_user) rag_data["total_memory"] = total_chunk - - # total_app: 统计当前空间下的所有app数量 - # 包含自有app + 被分享给本工作空间的app - from app.services import app_service as _app_svc - _, total_app = _app_svc.AppService(db).list_apps( - workspace_id=workspace_id, include_shared=True, pagesize=1 - ) - rag_data["total_app"] = total_app - - # total_knowledge: 使用 total_kb（总知识库数） - total_kb = memory_dashboard_service.get_rag_total_kb(db, current_user) - rag_data["total_knowledge"] = total_kb - - # total_api_call: 使用 AppStatisticsService 获取真实的API调用统计 - try: - app_stats_service = AppStatisticsService(db) - api_stats = app_stats_service.get_workspace_api_statistics( - workspace_id=workspace_id, - start_date=start_date, - end_date=end_date - ) - # 计算总调用次数 - total_api_calls = sum(item.get("total_calls", 0) for item in api_stats) - rag_data["total_api_call"] = total_api_calls - api_logger.info(f"成功获取RAG模式API调用统计: {rag_data['total_api_call']}") - except Exception as e: - api_logger.warning(f"获取RAG模式API调用统计失败，使用默认值: {str(e)}") - rag_data["total_api_call"] = 0 - - api_logger.info(f"成功获取RAG相关数据: memory={total_chunk}, app={total_app}, knowledge={total_kb}, api_calls={rag_data['total_api_call']}") + api_logger.info(f"成功获取RAG记忆总量: {total_chunk}") except Exception as e: - api_logger.warning(f"获取RAG相关数据失败: {str(e)}") + api_logger.warning(f"获取RAG记忆总量失败: {str(e)}") + # 2. 获取共享统计数据（total_app、total_knowledge、total_api_call） + common_stats = memory_dashboard_service.get_dashboard_common_stats(db, workspace_id) + rag_data.update(common_stats) + api_logger.info(f"成功获取共享统计: app={common_stats['total_app']}, knowledge={common_stats['total_knowledge']}, api_call={common_stats['total_api_call']}") + + # 计算昨日对比 + try: + changes = memory_dashboard_service.get_dashboard_yesterday_changes( + db=db, + workspace_id=workspace_id, + storage_type=storage_type, + today_data=rag_data + ) + rag_data.update(changes) + except Exception as e: + api_logger.warning(f"计算RAG昨日对比失败: {str(e)}") + rag_data.update({ + "total_memory_change": None, + "total_app_change": None, + "total_knowledge_change": None, + "total_api_call_change": None, + }) + result["rag_data"] = rag_data api_logger.info("成功获取rag_data") diff --git a/api/app/controllers/memory_storage_controller.py b/api/app/controllers/memory_storage_controller.py index d8b39325..76eed50f 100644 --- a/api/app/controllers/memory_storage_controller.py +++ b/api/app/controllers/memory_storage_controller.py @@ -26,7 +26,7 @@ from app.services.memory_storage_service import ( analytics_hot_memory_tags, analytics_recent_activity_stats, kb_type_distribution, - search_all, + search_all_batch, search_chunk, search_detials, search_dialogue, @@ -409,7 +409,10 @@ async def search_all_num( ) -> dict: api_logger.info(f"Search all requested for end_user_id: {end_user_id}") try: - result = await search_all(end_user_id) + if not end_user_id: + return success(data={"total": 0}, msg="查询成功") + batch_result = await search_all_batch([end_user_id]) + result = {"total": batch_result.get(end_user_id, 0)} return success(data=result, msg="查询成功") except Exception as e: api_logger.error(f"Search all failed: {str(e)}") diff --git a/api/app/controllers/ontology_controller.py b/api/app/controllers/ontology_controller.py index 3d2a1bdb..fe6b3598 100644 --- a/api/app/controllers/ontology_controller.py +++ b/api/app/controllers/ontology_controller.py @@ -163,6 +163,7 @@ def _get_ontology_service( 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 []), max_retries=3, timeout=60.0 ) diff --git a/api/app/controllers/public_share_controller.py b/api/app/controllers/public_share_controller.py index c10ad14b..ddd31071 100644 --- a/api/app/controllers/public_share_controller.py +++ b/api/app/controllers/public_share_controller.py @@ -453,6 +453,9 @@ async def chat( # 流式返回 agent_config = agent_config_4_app_release(release) + if not (agent_config.model_parameters.get("deep_thinking", False) and payload.thinking): + agent_config.model_parameters["deep_thinking"] = False + if payload.stream: async def event_generator(): async for event in app_chat_service.agnet_chat_stream( @@ -634,7 +637,8 @@ async def config_query( "app_type": release.app.type, "variables": release.config.get("variables"), "memory": release.config.get("memory", {}).get("enabled"), - "features": release.config.get("features") + "features": release.config.get("features"), + "model_parameters": release.config.get("model_parameters") } elif release.app.type == AppType.MULTI_AGENT: content = { diff --git a/api/app/controllers/service/__init__.py b/api/app/controllers/service/__init__.py index 8c679c1f..96da0949 100644 --- a/api/app/controllers/service/__init__.py +++ b/api/app/controllers/service/__init__.py @@ -4,7 +4,7 @@ 认证方式: API Key """ from fastapi import APIRouter -from . import app_api_controller, rag_api_knowledge_controller, rag_api_document_controller, rag_api_file_controller, rag_api_chunk_controller, memory_api_controller +from . import app_api_controller, rag_api_knowledge_controller, rag_api_document_controller, rag_api_file_controller, rag_api_chunk_controller, memory_api_controller, end_user_api_controller # 创建 V1 API 路由器 service_router = APIRouter() @@ -16,5 +16,6 @@ service_router.include_router(rag_api_document_controller.router) service_router.include_router(rag_api_file_controller.router) service_router.include_router(rag_api_chunk_controller.router) service_router.include_router(memory_api_controller.router) +service_router.include_router(end_user_api_controller.router) __all__ = ["service_router"] diff --git a/api/app/controllers/service/app_api_controller.py b/api/app/controllers/service/app_api_controller.py index d4573464..93caa200 100644 --- a/api/app/controllers/service/app_api_controller.py +++ b/api/app/controllers/service/app_api_controller.py @@ -144,6 +144,11 @@ async def chat( # print(app.current_release.default_model_config_id) agent_config = agent_config_4_app_release(app.current_release) # print(agent_config.default_model_config_id) + + # thinking 开关：仅当 agent 配置了 deep_thinking 且请求 thinking=True 时才启用 + if not (agent_config.model_parameters.get("deep_thinking", False) and payload.thinking): + agent_config.model_parameters["deep_thinking"] = False + # 流式返回 if payload.stream: async def event_generator(): diff --git a/api/app/controllers/service/end_user_api_controller.py b/api/app/controllers/service/end_user_api_controller.py new file mode 100644 index 00000000..df9996c2 --- /dev/null +++ b/api/app/controllers/service/end_user_api_controller.py @@ -0,0 +1,92 @@ +"""End User 服务接口 - 基于 API Key 认证""" + +import uuid + +from fastapi import APIRouter, Body, Depends, Request +from sqlalchemy.orm import Session + +from app.core.api_key_auth import require_api_key +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.response_utils import success +from app.db import get_db +from app.repositories.end_user_repository import EndUserRepository +from app.schemas.api_key_schema import ApiKeyAuth +from app.schemas.memory_api_schema import CreateEndUserRequest, CreateEndUserResponse +from app.services.memory_config_service import MemoryConfigService + +router = APIRouter(prefix="/end_user", tags=["V1 - End User API"]) +logger = get_business_logger() + + +@router.post("/create") +@require_api_key(scopes=["memory"]) +async def create_end_user( + request: Request, + api_key_auth: ApiKeyAuth = None, + db: Session = Depends(get_db), + message: str = Body(..., description="Request body"), +): + """ + Create or retrieve an end user for the workspace. + + Creates a new end user and connects it to a memory configuration. + If an end user with the same other_id already exists in the workspace, + returns the existing one. + + Optionally accepts a memory_config_id to connect the end user to a specific + memory configuration. If not provided, falls back to the workspace default config. + """ + body = await request.json() + payload = CreateEndUserRequest(**body) + workspace_id = api_key_auth.workspace_id + + logger.info("Create end user request - other_id: %s, workspace_id: %s", payload.other_id, workspace_id) + + # Resolve memory_config_id: explicit > workspace default + memory_config_id = None + config_service = MemoryConfigService(db) + + if payload.memory_config_id: + try: + memory_config_id = uuid.UUID(payload.memory_config_id) + except ValueError: + raise BusinessException( + f"Invalid memory_config_id format: {payload.memory_config_id}", + BizCode.INVALID_PARAMETER + ) + config = config_service.get_config_with_fallback(memory_config_id, workspace_id) + if not config: + raise BusinessException( + f"Memory config not found: {payload.memory_config_id}", + BizCode.MEMORY_CONFIG_NOT_FOUND + ) + memory_config_id = config.config_id + else: + default_config = config_service.get_workspace_default_config(workspace_id) + if default_config: + memory_config_id = default_config.config_id + logger.info(f"Using workspace default memory config: {memory_config_id}") + else: + logger.warning(f"No default memory config found for workspace: {workspace_id}") + + end_user_repo = EndUserRepository(db) + end_user = end_user_repo.get_or_create_end_user_with_config( + app_id=api_key_auth.resource_id, + workspace_id=workspace_id, + other_id=payload.other_id, + memory_config_id=memory_config_id, + ) + + logger.info(f"End user ready: {end_user.id}") + + result = { + "id": str(end_user.id), + "other_id": end_user.other_id or "", + "other_name": end_user.other_name or "", + "workspace_id": str(end_user.workspace_id), + "memory_config_id": str(end_user.memory_config_id) if end_user.memory_config_id else None, + } + + return success(data=CreateEndUserResponse(**result).model_dump(), msg="End user created successfully") diff --git a/api/app/core/agent/langchain_agent.py b/api/app/core/agent/langchain_agent.py index 38821313..044e7cc9 100644 --- a/api/app/core/agent/langchain_agent.py +++ b/api/app/core/agent/langchain_agent.py @@ -37,7 +37,10 @@ class LangChainAgent: tools: Optional[Sequence[BaseTool]] = None, streaming: bool = False, max_iterations: Optional[int] = None, # 最大迭代次数（None 表示自动计算） - max_tool_consecutive_calls: int = 3 # 单个工具最大连续调用次数 + max_tool_consecutive_calls: int = 3, # 单个工具最大连续调用次数 + deep_thinking: bool = False, # 是否启用深度思考模式 + thinking_budget_tokens: Optional[int] = None, # 深度思考 token 预算 + capability: Optional[List[str]] = None # 模型能力列表，用于校验是否支持深度思考 ): """初始化 LangChain Agent @@ -60,6 +63,7 @@ class LangChainAgent: self.streaming = streaming self.is_omni = is_omni self.max_tool_consecutive_calls = max_tool_consecutive_calls + self.deep_thinking = deep_thinking and ("thinking" in (capability or [])) # 工具调用计数器：记录每个工具的连续调用次数 self.tool_call_counter: Dict[str, int] = {} @@ -82,6 +86,13 @@ class LangChainAgent: f"auto_calculated={max_iterations is None}" ) + # 根据 capability 校验是否真正支持深度思考 + actual_deep_thinking = self.deep_thinking + if deep_thinking and not actual_deep_thinking: + logger.warning( + f"模型 {model_name} 不支持深度思考（capability 中无 'thinking'），已自动关闭 deep_thinking" + ) + # 创建 RedBearLLM（支持多提供商） model_config = RedBearModelConfig( model_name=model_name, @@ -89,10 +100,13 @@ class LangChainAgent: api_key=api_key, base_url=api_base, is_omni=is_omni, + deep_thinking=actual_deep_thinking, + thinking_budget_tokens=thinking_budget_tokens if actual_deep_thinking else None, + support_thinking="thinking" in (capability or []), extra_params={ "temperature": temperature, "max_tokens": max_tokens, - "streaming": streaming # 使用参数控制流式 + "streaming": streaming } ) @@ -249,6 +263,33 @@ class LangChainAgent: return messages + @staticmethod + def _extract_tokens_from_message(msg) -> int: + """从 AIMessage 或类似对象中提取 total_tokens，兼容多种 provider 格式 + + 支持的格式： + - response_metadata.token_usage.total_tokens (OpenAI/ChatOpenAI) + - response_metadata.usage.total_tokens (部分 provider) + - usage_metadata.total_tokens (LangChain 新版) + """ + total = 0 + # 1. response_metadata + response_meta = getattr(msg, "response_metadata", None) + if response_meta and isinstance(response_meta, dict): + # 尝试 token_usage 路径 + token_usage = response_meta.get("token_usage") or response_meta.get("usage", {}) + if isinstance(token_usage, dict): + total = token_usage.get("total_tokens", 0) + # 2. usage_metadata（LangChain 新版 AIMessage 属性） + if not total: + usage_meta = getattr(msg, "usage_metadata", None) + if usage_meta: + if isinstance(usage_meta, dict): + total = usage_meta.get("total_tokens", 0) + else: + total = getattr(usage_meta, "total_tokens", 0) + return total or 0 + def _build_multimodal_content(self, text: str, files: List[Dict[str, Any]]) -> List[Dict[str, Any]]: """ 构建多模态消息内容 @@ -283,6 +324,17 @@ class LangChainAgent: return content_parts + @staticmethod + def _extract_reasoning_content(msg) -> str: + """从 AIMessage 中提取深度思考内容（reasoning_content） + + 所有 provider 统一通过 additional_kwargs.reasoning_content 传递： + - DeepSeek-R1 / QwQ: 原生字段 + - Volcano (Doubao-thinking): 由 VolcanoChatOpenAI 从 delta.reasoning_content 注入 + """ + additional = getattr(msg, "additional_kwargs", None) or {} + return additional.get("reasoning_content") or additional.get("reasoning", "") + async def chat( self, message: str, @@ -348,6 +400,7 @@ class LangChainAgent: logger.debug(f"输出消息数量: {len(output_messages)}") total_tokens = 0 + reasoning_content = "" for msg in reversed(output_messages): if isinstance(msg, AIMessage): logger.debug(f"找到 AI 消息，content 类型: {type(msg.content)}") @@ -382,8 +435,8 @@ class LangChainAgent: else: content = str(msg.content) logger.debug(f"转换为字符串: {content[:100]}...") - response_meta = msg.response_metadata if hasattr(msg, 'response_metadata') else None - total_tokens = response_meta.get("token_usage", {}).get("total_tokens", 0) if response_meta else 0 + total_tokens = self._extract_tokens_from_message(msg) + reasoning_content = self._extract_reasoning_content(msg) if self.deep_thinking else "" break logger.info(f"最终提取的内容长度: {len(content)}") @@ -399,6 +452,8 @@ class LangChainAgent: "total_tokens": total_tokens } } + if reasoning_content: + response["reasoning_content"] = reasoning_content logger.debug( "Agent 调用完成", @@ -420,7 +475,7 @@ class LangChainAgent: history: Optional[List[Dict[str, str]]] = None, context: Optional[str] = None, files: Optional[List[Dict[str, Any]]] = None - ) -> AsyncGenerator[str, None]: + ) -> AsyncGenerator[str | int, None]: """执行流式对话 Args: @@ -431,6 +486,8 @@ class LangChainAgent: Yields: str: 消息内容块 + int: token 统计 + Dict: 深度思考内容 {"type": "reasoning", "content": "..."} """ logger.info("=" * 80) logger.info(" chat_stream 方法开始执行") @@ -451,6 +508,7 @@ class LangChainAgent: # 统一使用 agent 的 astream_events 实现流式输出 logger.debug("使用 Agent astream_events 实现流式输出") full_content = '' + full_reasoning = '' try: last_event = {} async for event in self.agent.astream_events( @@ -467,6 +525,13 @@ class LangChainAgent: # LLM 流式输出 chunk = event.get("data", {}).get("chunk") if chunk and hasattr(chunk, "content"): + # 提取深度思考内容（仅在启用深度思考时） + if self.deep_thinking: + reasoning_chunk = self._extract_reasoning_content(chunk) + if reasoning_chunk: + full_reasoning += reasoning_chunk + yield {"type": "reasoning", "content": reasoning_chunk} + # 处理多模态响应：content 可能是字符串或列表 chunk_content = chunk.content if isinstance(chunk_content, str) and chunk_content: @@ -497,6 +562,13 @@ class LangChainAgent: chunk = event.get("data", {}).get("chunk") if chunk: if hasattr(chunk, "content"): + # 提取深度思考内容（仅在启用深度思考时） + if self.deep_thinking: + reasoning_chunk = self._extract_reasoning_content(chunk) + if reasoning_chunk: + full_reasoning += reasoning_chunk + yield {"type": "reasoning", "content": reasoning_chunk} + chunk_content = chunk.content if isinstance(chunk_content, str) and chunk_content: full_content += chunk_content @@ -535,12 +607,9 @@ class LangChainAgent: output_messages = last_event.get("data", {}).get("output", {}).get("messages", []) for msg in reversed(output_messages): if isinstance(msg, AIMessage): - response_meta = msg.response_metadata if hasattr(msg, 'response_metadata') else None - total_tokens = response_meta.get("token_usage", {}).get( - "total_tokens", - 0 - ) if response_meta else 0 - yield total_tokens + stream_total_tokens = self._extract_tokens_from_message(msg) + logger.info(f"流式 token 统计: total_tokens={stream_total_tokens}") + yield stream_total_tokens break except Exception as e: diff --git a/api/app/core/memory/src/search.py b/api/app/core/memory/src/search.py index e4f0d4d0..a3c40dcd 100644 --- a/api/app/core/memory/src/search.py +++ b/api/app/core/memory/src/search.py @@ -758,8 +758,7 @@ async def run_hybrid_search( model_name=embedder_config_dict["model_name"], provider=embedder_config_dict["provider"], api_key=embedder_config_dict["api_key"], - base_url=embedder_config_dict["base_url"], - type="llm" + base_url=embedder_config_dict["base_url"] ) config_load_time = time.time() - config_load_start logger.info(f"[PERF] Config loading took {config_load_time:.4f}s") diff --git a/api/app/core/models/base.py b/api/app/core/models/base.py index 80117f27..c7d8cfed 100644 --- a/api/app/core/models/base.py +++ b/api/app/core/models/base.py @@ -14,6 +14,7 @@ from pydantic import BaseModel, Field from app.core.error_codes import BizCode from app.core.exceptions import BusinessException from app.models.models_model import ModelProvider, ModelType +from app.core.models.volcano_chat import VolcanoChatOpenAI T = TypeVar("T") @@ -25,6 +26,9 @@ class RedBearModelConfig(BaseModel): api_key: str base_url: Optional[str] = None is_omni: bool = False # 是否为 Omni 模型 + deep_thinking: bool = False # 是否启用深度思考模式 + thinking_budget_tokens: Optional[int] = None # 深度思考 token 预算 + support_thinking: bool = False # 模型是否支持 enable_thinking 参数（capability 含 thinking） # 请求超时时间（秒）- 默认120秒以支持复杂的LLM调用，可通过环境变量 LLM_TIMEOUT 配置 timeout: float = Field(default_factory=lambda: float(os.getenv("LLM_TIMEOUT", "120.0"))) # 最大重试次数 - 默认2次以避免过长等待，可通过环境变量 LLM_MAX_RETRIES 配置 @@ -44,7 +48,7 @@ class RedBearModelFactory: # 打印供应商信息用于调试 from app.core.logging_config import get_business_logger logger = get_business_logger() - logger.debug(f"获取模型参数 - Provider: {provider}, Model: {config.model_name}, is_omni: {config.is_omni}") + logger.debug(f"获取模型参数 - Provider: {provider}, Model: {config.model_name}, is_omni: {config.is_omni}, deep_thinking: {config.deep_thinking}") # dashscope 的 omni 模型使用 OpenAI 兼容模式 if provider == ModelProvider.DASHSCOPE and config.is_omni: @@ -58,7 +62,7 @@ class RedBearModelFactory: write=60.0, pool=10.0, ) - return { + params: Dict[str, Any] = { "model": config.model_name, "base_url": config.base_url, "api_key": config.api_key, @@ -66,6 +70,21 @@ class RedBearModelFactory: "max_retries": config.max_retries, **config.extra_params } + # 流式模式下启用 stream_usage 以获取 token 统计 + is_streaming = bool(config.extra_params.get("streaming")) + if is_streaming: + params["stream_usage"] = True + # 只有支持 thinking 的模型才传 enable_thinking + if config.support_thinking: + model_kwargs: Dict[str, Any] = config.extra_params.get("model_kwargs", {}) + if is_streaming: + model_kwargs["enable_thinking"] = config.deep_thinking + if config.deep_thinking and config.thinking_budget_tokens: + model_kwargs["thinking_budget"] = config.thinking_budget_tokens + else: + model_kwargs["enable_thinking"] = False + params["model_kwargs"] = model_kwargs + return params if provider in [ModelProvider.OPENAI, ModelProvider.XINFERENCE, ModelProvider.GPUSTACK, ModelProvider.OLLAMA, ModelProvider.VOLCANO]: # 使用 httpx.Timeout 对象来设置详细的超时配置 @@ -78,7 +97,7 @@ class RedBearModelFactory: write=60.0, # 写入超时：60秒 pool=10.0, # 连接池超时：10秒 ) - return { + params: Dict[str, Any] = { "model": config.model_name, "base_url": config.base_url, "api_key": config.api_key, @@ -86,16 +105,47 @@ class RedBearModelFactory: "max_retries": config.max_retries, **config.extra_params } + # 流式模式下启用 stream_usage 以获取 token 统计 + if config.extra_params.get("streaming"): + params["stream_usage"] = True + # 深度思考模式 + is_streaming = bool(config.extra_params.get("streaming")) + if is_streaming: + if provider == ModelProvider.VOLCANO: + # 火山引擎深度思考仅流式调用支持，非流式时不传 thinking 参数 + thinking_config: Dict[str, Any] = { + "type": "enabled" if config.deep_thinking else "disabled" + } + if config.deep_thinking and config.thinking_budget_tokens: + thinking_config["budget_tokens"] = config.thinking_budget_tokens + params["extra_body"] = {"thinking": thinking_config} + else: + # 始终显式传递 enable_thinking，不支持该参数的模型（如 DeepSeek-R1）会直接忽略 + model_kwargs: Dict[str, Any] = config.extra_params.get("model_kwargs", {}) + model_kwargs["enable_thinking"] = config.deep_thinking + if config.deep_thinking and config.thinking_budget_tokens: + model_kwargs["thinking_budget"] = config.thinking_budget_tokens + params["model_kwargs"] = model_kwargs + return params elif provider == ModelProvider.DASHSCOPE: - # DashScope (通义千问) 使用自己的参数格式 - # 注意: DashScopeEmbeddings 不支持 timeout 和 base_url 参数 - # 只支持: model, dashscope_api_key, max_retries, client - return { + params = { "model": config.model_name, "dashscope_api_key": config.api_key, "max_retries": config.max_retries, **config.extra_params } + # 只有支持 thinking 的模型才传 enable_thinking + if config.support_thinking: + is_streaming = bool(config.extra_params.get("streaming")) + model_kwargs: Dict[str, Any] = config.extra_params.get("model_kwargs", {}) + if is_streaming: + model_kwargs["enable_thinking"] = config.deep_thinking + if config.deep_thinking and config.thinking_budget_tokens: + model_kwargs["thinking_budget"] = config.thinking_budget_tokens + else: + model_kwargs["enable_thinking"] = False + params["model_kwargs"] = model_kwargs + return params elif provider == ModelProvider.BEDROCK: # Bedrock 使用 AWS 凭证 # api_key 格式: "access_key_id:secret_access_key" 或只是 access_key_id @@ -134,6 +184,13 @@ class RedBearModelFactory: elif "region_name" not in params: params["region_name"] = "us-east-1" # 默认区域 + # 深度思考模式：Claude 3.7 Sonnet 等支持思考的模型 + # 通过 additional_model_request_fields 传递 thinking 块，关闭时不传（Bedrock 无 disabled 选项） + if config.deep_thinking: + budget = config.thinking_budget_tokens or 10000 + params["additional_model_request_fields"] = { + "thinking": {"type": "enabled", "budget_tokens": budget} + } return params else: raise BusinessException(f"不支持的提供商: {provider}", code=BizCode.PROVIDER_NOT_SUPPORTED) @@ -160,7 +217,9 @@ def get_provider_llm_class(config: RedBearModelConfig, type: ModelType = ModelTy # dashscope 的 omni 模型使用 OpenAI 兼容模式 if provider == ModelProvider.DASHSCOPE and config.is_omni: return ChatOpenAI - if provider in [ModelProvider.OPENAI, ModelProvider.XINFERENCE, ModelProvider.GPUSTACK, ModelProvider.VOLCANO]: + if provider == ModelProvider.VOLCANO: + return VolcanoChatOpenAI + if provider in [ModelProvider.OPENAI, ModelProvider.XINFERENCE, ModelProvider.GPUSTACK]: if type == ModelType.LLM: return OpenAI elif type == ModelType.CHAT: diff --git a/api/app/core/models/scripts/bedrock_models.yaml b/api/app/core/models/scripts/bedrock_models.yaml index 2c0ab757..5b3a2f64 100644 --- a/api/app/core/models/scripts/bedrock_models.yaml +++ b/api/app/core/models/scripts/bedrock_models.yaml @@ -11,6 +11,7 @@ models: tags: - 大语言模型 logo: bedrock + - name: amazon nova type: llm provider: bedrock @@ -27,6 +28,7 @@ models: - stream-tool-call - vision logo: bedrock + - name: anthropic claude type: llm provider: bedrock @@ -35,6 +37,7 @@ models: is_official: true capability: - vision + - thinking is_omni: false tags: - 大语言模型 @@ -44,6 +47,7 @@ models: - stream-tool-call - document logo: bedrock + - name: cohere type: llm provider: bedrock @@ -58,6 +62,7 @@ models: - tool-call - stream-tool-call logo: bedrock + - name: deepseek type: llm provider: bedrock @@ -66,6 +71,7 @@ models: is_official: true capability: - vision + - thinking is_omni: false tags: - 大语言模型 @@ -74,6 +80,7 @@ models: - tool-call - stream-tool-call logo: bedrock + - name: meta type: llm provider: bedrock @@ -87,6 +94,7 @@ models: - agent-thought - tool-call logo: bedrock + - name: mistral type: llm provider: bedrock @@ -100,6 +108,7 @@ models: - agent-thought - tool-call logo: bedrock + - name: openai type: llm provider: bedrock @@ -114,6 +123,7 @@ models: - tool-call - stream-tool-call logo: bedrock + - name: qwen type: llm provider: bedrock @@ -128,6 +138,7 @@ models: - tool-call - stream-tool-call logo: bedrock + - name: amazon.rerank-v1:0 type: rerank provider: bedrock @@ -139,6 +150,7 @@ models: tags: - 重排序模型 logo: bedrock + - name: cohere.rerank-v3-5:0 type: rerank provider: bedrock @@ -150,6 +162,7 @@ models: tags: - 重排序模型 logo: bedrock + - name: amazon.nova-2-multimodal-embeddings-v1:0 type: embedding provider: bedrock @@ -163,6 +176,7 @@ models: - 文本嵌入模型 - vision logo: bedrock + - name: amazon.titan-embed-text-v1 type: embedding provider: bedrock @@ -174,6 +188,7 @@ models: tags: - 文本嵌入模型 logo: bedrock + - name: amazon.titan-embed-text-v2:0 type: embedding provider: bedrock @@ -185,6 +200,7 @@ models: tags: - 文本嵌入模型 logo: bedrock + - name: cohere.embed-english-v3 type: embedding provider: bedrock @@ -196,6 +212,7 @@ models: tags: - 文本嵌入模型 logo: bedrock + - name: cohere.embed-multilingual-v3 type: embedding provider: bedrock diff --git a/api/app/core/models/scripts/dashscope_models.yaml b/api/app/core/models/scripts/dashscope_models.yaml index 89a16966..d1b604e0 100644 --- a/api/app/core/models/scripts/dashscope_models.yaml +++ b/api/app/core/models/scripts/dashscope_models.yaml @@ -6,36 +6,42 @@ models: description: DeepSeek-R1-Distill-Qwen-14B大语言模型，支持智能体思考，32000上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 - agent-thought logo: dashscope + - name: deepseek-r1-distill-qwen-32b type: llm provider: dashscope description: DeepSeek-R1-Distill-Qwen-32B大语言模型，支持智能体思考，32000上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 - agent-thought logo: dashscope + - name: deepseek-r1 type: llm provider: dashscope description: DeepSeek-R1大语言模型，支持智能体思考，131072超大上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 - agent-thought logo: dashscope + - name: deepseek-v3.1 type: llm provider: dashscope @@ -48,6 +54,7 @@ models: - 大语言模型 - agent-thought logo: dashscope + - name: deepseek-v3.2-exp type: llm provider: dashscope @@ -60,6 +67,7 @@ models: - 大语言模型 - agent-thought logo: dashscope + - name: deepseek-v3.2 type: llm provider: dashscope @@ -72,6 +80,7 @@ models: - 大语言模型 - agent-thought logo: dashscope + - name: deepseek-v3 type: llm provider: dashscope @@ -84,6 +93,7 @@ models: - 大语言模型 - agent-thought logo: dashscope + - name: farui-plus type: llm provider: dashscope @@ -98,6 +108,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: glm-4.7 type: llm provider: dashscope @@ -112,6 +123,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qvq-max-latest type: llm provider: dashscope @@ -119,7 +131,8 @@ models: is_deprecated: false is_official: true capability: - - vision + - vision + - thinking is_omni: false tags: - 大语言模型 @@ -127,6 +140,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qvq-max type: llm provider: dashscope @@ -134,7 +148,8 @@ models: is_deprecated: false is_official: true capability: - - vision + - vision + - thinking is_omni: false tags: - 大语言模型 @@ -142,6 +157,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen-coder-turbo-0919 type: llm provider: dashscope @@ -155,13 +171,15 @@ models: - 代码模型 - agent-thought logo: dashscope + - name: qwen-max-latest type: llm provider: dashscope description: qwen-max-latest大语言模型，支持多工具调用、智能体思考、流式工具调用，131072上下文窗口，对话模式，支持联网搜索 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -169,6 +187,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen-max-longcontext type: llm provider: dashscope @@ -183,13 +202,15 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen-max type: llm provider: dashscope description: qwen-max大语言模型，支持多工具调用、智能体思考、流式工具调用，32768上下文窗口，对话模式，支持联网搜索 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -197,6 +218,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen-mt-plus type: llm provider: dashscope @@ -210,6 +232,7 @@ models: - 翻译模型 - agent-thought logo: dashscope + - name: qwen-mt-turbo type: llm provider: dashscope @@ -223,6 +246,7 @@ models: - 翻译模型 - agent-thought logo: dashscope + - name: qwen-plus-0112 type: llm provider: dashscope @@ -237,6 +261,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen-plus-0125 type: llm provider: dashscope @@ -251,6 +276,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen-plus-0723 type: llm provider: dashscope @@ -265,6 +291,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen-plus-0806 type: llm provider: dashscope @@ -279,6 +306,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen-plus-0919 type: llm provider: dashscope @@ -293,6 +321,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen-plus-1125 type: llm provider: dashscope @@ -307,6 +336,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen-plus-1127 type: llm provider: dashscope @@ -321,6 +351,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen-plus-1220 type: llm provider: dashscope @@ -335,6 +366,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen-vl-max type: chat provider: dashscope @@ -342,8 +374,8 @@ models: is_deprecated: false is_official: true capability: - - vision - - video + - vision + - video is_omni: false tags: - 大语言模型 @@ -352,6 +384,7 @@ models: - agent-thought - video logo: dashscope + - name: qwen-vl-plus-0809 type: chat provider: dashscope @@ -359,8 +392,8 @@ models: is_deprecated: true is_official: true capability: - - vision - - video + - vision + - video is_omni: false tags: - 大语言模型 @@ -369,6 +402,7 @@ models: - agent-thought - video logo: dashscope + - name: qwen-vl-plus-2025-01-02 type: chat provider: dashscope @@ -376,8 +410,8 @@ models: is_deprecated: false is_official: true capability: - - vision - - video + - vision + - video is_omni: false tags: - 大语言模型 @@ -386,6 +420,7 @@ models: - agent-thought - video logo: dashscope + - name: qwen-vl-plus-2025-01-25 type: chat provider: dashscope @@ -393,8 +428,8 @@ models: is_deprecated: false is_official: true capability: - - vision - - video + - vision + - video is_omni: false tags: - 大语言模型 @@ -403,6 +438,7 @@ models: - agent-thought - video logo: dashscope + - name: qwen-vl-plus-latest type: chat provider: dashscope @@ -410,8 +446,8 @@ models: is_deprecated: false is_official: true capability: - - vision - - video + - vision + - video is_omni: false tags: - 大语言模型 @@ -420,6 +456,7 @@ models: - agent-thought - video logo: dashscope + - name: qwen-vl-plus type: chat provider: dashscope @@ -427,8 +464,8 @@ models: is_deprecated: false is_official: true capability: - - vision - - video + - vision + - video is_omni: false tags: - 大语言模型 @@ -437,6 +474,7 @@ models: - agent-thought - video logo: dashscope + - name: qwen2.5-0.5b-instruct type: llm provider: dashscope @@ -451,13 +489,15 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen3-14b type: llm provider: dashscope description: qwen3-14b大语言模型，支持多工具调用、智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -465,13 +505,15 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen3-235b-a22b-instruct-2507 type: llm provider: dashscope description: qwen3-235b-a22b-instruct-2507大语言模型，支持多工具调用、智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -479,13 +521,15 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen3-235b-a22b-thinking-2507 type: llm provider: dashscope description: qwen3-235b-a22b-thinking-2507大语言模型，支持多工具调用、智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -493,13 +537,15 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen3-235b-a22b type: llm provider: dashscope description: qwen3-235b-a22b大语言模型，支持多工具调用、智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -507,13 +553,15 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen3-30b-a3b-instruct-2507 type: llm provider: dashscope description: qwen3-30b-a3b-instruct-2507大语言模型，支持多工具调用、智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -521,13 +569,15 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen3-30b-a3b type: llm provider: dashscope description: qwen3-30b-a3b大语言模型，支持多工具调用、智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -535,13 +585,15 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen3-32b type: llm provider: dashscope description: qwen3-32b大语言模型，支持多工具调用、智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -549,13 +601,15 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen3-4b type: llm provider: dashscope description: qwen3-4b大语言模型，支持多工具调用、智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -563,13 +617,15 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen3-8b type: llm provider: dashscope description: qwen3-8b大语言模型，支持多工具调用、智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -577,65 +633,75 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen3-coder-30b-a3b-instruct type: llm provider: dashscope description: qwen3-coder-30b-a3b-instruct大语言模型，支持智能体思考，262144上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 - 代码模型 - agent-thought logo: dashscope + - name: qwen3-coder-480b-a35b-instruct type: llm provider: dashscope description: qwen3-coder-480b-a35b-instruct大语言模型，支持智能体思考，262144上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 - 代码模型 - agent-thought logo: dashscope + - name: qwen3-coder-plus-2025-09-23 type: llm provider: dashscope description: qwen3-coder-plus-2025-09-23大语言模型，支持智能体思考，1000000上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 - 代码模型 - agent-thought logo: dashscope + - name: qwen3-coder-plus type: llm provider: dashscope description: qwen3-coder-plus大语言模型，支持智能体思考，1000000上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 - 代码模型 - agent-thought logo: dashscope + - name: qwen3-max-2025-09-23 type: llm provider: dashscope description: qwen3-max-2025-09-23大语言模型，支持多工具调用、智能体思考、流式工具调用，262144上下文窗口，对话模式，支持联网搜索 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -644,13 +710,15 @@ models: - stream-tool-call - 联网搜索 logo: dashscope + - name: qwen3-max-2026-01-23 type: llm provider: dashscope description: qwen3-max-2026-01-23大语言模型，支持多工具调用、智能体思考、流式工具调用，262144上下文窗口，对话模式，支持联网搜索 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -659,13 +727,15 @@ models: - stream-tool-call - 联网搜索 logo: dashscope + - name: qwen3-max-preview type: llm provider: dashscope description: qwen3-max-preview大语言模型，支持多工具调用、智能体思考、流式工具调用，262144上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -673,13 +743,15 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen3-max type: llm provider: dashscope description: qwen3-max大语言模型，支持多工具调用、智能体思考、流式工具调用，262144上下文窗口，对话模式，支持联网搜索 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -688,13 +760,15 @@ models: - stream-tool-call - 联网搜索 logo: dashscope + - name: qwen3-next-80b-a3b-instruct type: llm provider: dashscope description: qwen3-next-80b-a3b-instruct大语言模型，支持多工具调用、智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -702,13 +776,15 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen3-next-80b-a3b-thinking type: llm provider: dashscope description: qwen3-next-80b-a3b-thinking大语言模型，支持多工具调用、智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -716,6 +792,7 @@ models: - agent-thought - stream-tool-call logo: dashscope + - name: qwen3-omni-flash-2025-12-01 type: llm provider: dashscope @@ -723,9 +800,10 @@ models: is_deprecated: false is_official: true capability: - - vision - - video - - audio + - vision + - video + - audio + - thinking is_omni: true tags: - 大语言模型 @@ -735,6 +813,7 @@ models: - video - audio logo: dashscope + - name: qwen3-vl-235b-a22b-instruct type: chat provider: dashscope @@ -742,8 +821,9 @@ models: is_deprecated: false is_official: true capability: - - vision - - video + - vision + - video + - thinking is_omni: false tags: - 大语言模型 @@ -754,6 +834,7 @@ models: - vision - video logo: dashscope + - name: qwen3-vl-235b-a22b-thinking type: chat provider: dashscope @@ -761,8 +842,9 @@ models: is_deprecated: false is_official: true capability: - - vision - - video + - vision + - video + - thinking is_omni: false tags: - 大语言模型 @@ -773,6 +855,7 @@ models: - vision - video logo: dashscope + - name: qwen3-vl-30b-a3b-instruct type: chat provider: dashscope @@ -780,8 +863,9 @@ models: is_deprecated: false is_official: true capability: - - vision - - video + - vision + - video + - thinking is_omni: false tags: - 大语言模型 @@ -792,6 +876,7 @@ models: - vision - video logo: dashscope + - name: qwen3-vl-30b-a3b-thinking type: chat provider: dashscope @@ -799,8 +884,9 @@ models: is_deprecated: false is_official: true capability: - - vision - - video + - vision + - video + - thinking is_omni: false tags: - 大语言模型 @@ -811,6 +897,7 @@ models: - vision - video logo: dashscope + - name: qwen3-vl-flash type: chat provider: dashscope @@ -818,8 +905,9 @@ models: is_deprecated: false is_official: true capability: - - vision - - video + - vision + - video + - thinking is_omni: false tags: - 大语言模型 @@ -830,6 +918,7 @@ models: - vision - video logo: dashscope + - name: qwen3-vl-plus-2025-09-23 type: chat provider: dashscope @@ -837,8 +926,9 @@ models: is_deprecated: false is_official: true capability: - - vision - - video + - vision + - video + - thinking is_omni: false tags: - 大语言模型 @@ -847,6 +937,7 @@ models: - agent-thought - video logo: dashscope + - name: qwen3-vl-plus type: chat provider: dashscope @@ -854,8 +945,9 @@ models: is_deprecated: false is_official: true capability: - - vision - - video + - vision + - video + - thinking is_omni: false tags: - 大语言模型 @@ -864,45 +956,52 @@ models: - agent-thought - video logo: dashscope + - name: qwq-32b type: llm provider: dashscope description: qwq-32b大语言模型，支持智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 - agent-thought - stream-tool-call logo: dashscope + - name: qwq-plus-0305 type: llm provider: dashscope description: qwq-plus-0305大语言模型，支持智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 - agent-thought - stream-tool-call logo: dashscope + - name: qwq-plus type: llm provider: dashscope description: qwq-plus大语言模型，支持智能体思考、流式工具调用，131072上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 - agent-thought - stream-tool-call logo: dashscope + - name: gte-rerank-v2 type: rerank provider: dashscope @@ -914,6 +1013,7 @@ models: tags: - 重排序模型 logo: dashscope + - name: gte-rerank type: rerank provider: dashscope @@ -925,6 +1025,7 @@ models: tags: - 重排序模型 logo: dashscope + - name: multimodal-embedding-v1 type: embedding provider: dashscope @@ -932,13 +1033,14 @@ models: is_deprecated: false is_official: true capability: - - vision + - vision is_omni: false tags: - 嵌入模型 - 多模态模型 - vision logo: dashscope + - name: text-embedding-v1 type: embedding provider: dashscope @@ -951,6 +1053,7 @@ models: - 嵌入模型 - 文本嵌入 logo: dashscope + - name: text-embedding-v2 type: embedding provider: dashscope @@ -963,6 +1066,7 @@ models: - 嵌入模型 - 文本嵌入 logo: dashscope + - name: text-embedding-v3 type: embedding provider: dashscope @@ -975,6 +1079,7 @@ models: - 嵌入模型 - 文本嵌入 logo: dashscope + - name: text-embedding-v4 type: embedding provider: dashscope @@ -986,4 +1091,4 @@ models: tags: - 嵌入模型 - 文本嵌入 - logo: dashscope \ No newline at end of file + logo: dashscope diff --git a/api/app/core/models/scripts/openai_models.yaml b/api/app/core/models/scripts/openai_models.yaml index 7f6d3a51..08b81008 100644 --- a/api/app/core/models/scripts/openai_models.yaml +++ b/api/app/core/models/scripts/openai_models.yaml @@ -20,6 +20,7 @@ models: - audio - video logo: openai + - name: gpt-3.5-turbo-0125 type: llm provider: openai @@ -34,6 +35,7 @@ models: - agent-thought - stream-tool-call logo: openai + - name: gpt-3.5-turbo-1106 type: llm provider: openai @@ -48,6 +50,7 @@ models: - agent-thought - stream-tool-call logo: openai + - name: gpt-3.5-turbo-16k type: llm provider: openai @@ -62,6 +65,7 @@ models: - agent-thought - stream-tool-call logo: openai + - name: gpt-3.5-turbo-instruct type: llm provider: openai @@ -73,6 +77,7 @@ models: tags: - 大语言模型 logo: openai + - name: gpt-3.5-turbo type: llm provider: openai @@ -87,6 +92,7 @@ models: - agent-thought - stream-tool-call logo: openai + - name: gpt-4-0125-preview type: llm provider: openai @@ -101,6 +107,7 @@ models: - agent-thought - stream-tool-call logo: openai + - name: gpt-4-1106-preview type: llm provider: openai @@ -115,6 +122,7 @@ models: - agent-thought - stream-tool-call logo: openai + - name: gpt-4-turbo-2024-04-09 type: llm provider: openai @@ -131,6 +139,7 @@ models: - stream-tool-call - vision logo: openai + - name: gpt-4-turbo-preview type: llm provider: openai @@ -145,6 +154,7 @@ models: - agent-thought - stream-tool-call logo: openai + - name: gpt-4-turbo type: llm provider: openai @@ -161,6 +171,7 @@ models: - stream-tool-call - vision logo: openai + - name: o1-preview type: llm provider: openai @@ -173,6 +184,7 @@ models: - 大语言模型 - agent-thought logo: openai + - name: o1 type: llm provider: openai @@ -181,6 +193,7 @@ models: is_official: true capability: - vision + - thinking is_omni: false tags: - 大语言模型 @@ -190,6 +203,7 @@ models: - vision - structured-output logo: openai + - name: o3-2025-04-16 type: llm provider: openai @@ -198,6 +212,7 @@ models: is_official: true capability: - vision + - thinking is_omni: false tags: - 大语言模型 @@ -207,13 +222,15 @@ models: - stream-tool-call - structured-output logo: openai + - name: o3-mini-2025-01-31 type: llm provider: openai description: o3-mini-2025-01-31大语言模型，支持智能体思考、工具调用、流式工具调用、结构化输出，200000上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -222,13 +239,15 @@ models: - stream-tool-call - structured-output logo: openai + - name: o3-mini type: llm provider: openai description: o3-mini大语言模型，支持智能体思考、工具调用、流式工具调用、结构化输出，200000上下文窗口，对话模式 is_deprecated: false is_official: true - capability: [] + capability: + - thinking is_omni: false tags: - 大语言模型 @@ -237,6 +256,7 @@ models: - stream-tool-call - structured-output logo: openai + - name: o3-pro-2025-06-10 type: llm provider: openai @@ -245,6 +265,7 @@ models: is_official: true capability: - vision + - thinking is_omni: false tags: - 大语言模型 @@ -253,6 +274,7 @@ models: - vision - structured-output logo: openai + - name: o3-pro type: llm provider: openai @@ -261,6 +283,7 @@ models: is_official: true capability: - vision + - thinking is_omni: false tags: - 大语言模型 @@ -269,6 +292,7 @@ models: - vision - structured-output logo: openai + - name: o3 type: llm provider: openai @@ -277,6 +301,7 @@ models: is_official: true capability: - vision + - thinking is_omni: false tags: - 大语言模型 @@ -286,6 +311,7 @@ models: - stream-tool-call - structured-output logo: openai + - name: o4-mini-2025-04-16 type: llm provider: openai @@ -294,6 +320,7 @@ models: is_official: true capability: - vision + - thinking is_omni: false tags: - 大语言模型 @@ -303,6 +330,7 @@ models: - stream-tool-call - structured-output logo: openai + - name: o4-mini type: llm provider: openai @@ -311,6 +339,7 @@ models: is_official: true capability: - vision + - thinking is_omni: false tags: - 大语言模型 @@ -320,6 +349,7 @@ models: - stream-tool-call - structured-output logo: openai + - name: text-embedding-3-large type: embedding provider: openai @@ -331,6 +361,7 @@ models: tags: - 文本向量模型 logo: openai + - name: text-embedding-3-small type: embedding provider: openai @@ -342,6 +373,7 @@ models: tags: - 文本向量模型 logo: openai + - name: text-embedding-ada-002 type: embedding provider: openai diff --git a/api/app/core/models/scripts/volcano_models.yaml b/api/app/core/models/scripts/volcano_models.yaml index 24609f5a..c86d41ac 100644 --- a/api/app/core/models/scripts/volcano_models.yaml +++ b/api/app/core/models/scripts/volcano_models.yaml @@ -10,6 +10,7 @@ models: capability: - vision - video + - thinking is_omni: false tags: - 大语言模型 @@ -24,6 +25,7 @@ models: capability: - vision - video + - thinking is_omni: false tags: - 大语言模型 @@ -38,6 +40,7 @@ models: capability: - vision - video + - thinking is_omni: false tags: - 大语言模型 @@ -52,6 +55,7 @@ models: capability: - vision - video + - thinking is_omni: false tags: - 大语言模型 @@ -82,6 +86,7 @@ models: capability: - vision - video + - thinking is_omni: false tags: - 大语言模型 @@ -96,6 +101,7 @@ models: capability: - vision - video + - thinking is_omni: false tags: - 大语言模型 @@ -110,6 +116,7 @@ models: capability: - vision - video + - thinking is_omni: false tags: - 大语言模型 @@ -124,6 +131,7 @@ models: capability: - vision - video + - thinking is_omni: false tags: - 大语言模型 @@ -139,6 +147,7 @@ models: capability: - vision - video + - thinking is_omni: false tags: - 大语言模型 diff --git a/api/app/core/models/volcano_chat.py b/api/app/core/models/volcano_chat.py new file mode 100644 index 00000000..d86484a5 --- /dev/null +++ b/api/app/core/models/volcano_chat.py @@ -0,0 +1,38 @@ +""" +火山引擎 ChatOpenAI 扩展 + +ChatOpenAI 在解析流式 SSE 时只取 delta.content，会丢弃 delta.reasoning_content。 +此类仅重写 _convert_chunk_to_generation_chunk，将 reasoning_content 补入 additional_kwargs。 +""" +from __future__ import annotations + +from typing import Any, Optional + +from langchain_core.outputs import ChatGenerationChunk +from langchain_openai import ChatOpenAI + + +class VolcanoChatOpenAI(ChatOpenAI): + """火山引擎 Chat 模型，支持深度思考内容（reasoning_content）的流式透传。""" + + def _convert_chunk_to_generation_chunk( + self, + chunk: dict, + default_chunk_class: type, + base_generation_info: Optional[dict], + ) -> Optional[ChatGenerationChunk]: + gen_chunk = super()._convert_chunk_to_generation_chunk( + chunk, default_chunk_class, base_generation_info + ) + if gen_chunk is None: + return None + + # 从原始 chunk 中提取 reasoning_content + choices = chunk.get("choices") or chunk.get("chunk", {}).get("choices", []) + if choices: + delta = choices[0].get("delta") or {} + reasoning: Any = delta.get("reasoning_content") + if reasoning: + gen_chunk.message.additional_kwargs["reasoning_content"] = reasoning + + return gen_chunk diff --git a/api/app/core/tools/mcp/client.py b/api/app/core/tools/mcp/client.py index b437d021..3539d33a 100644 --- a/api/app/core/tools/mcp/client.py +++ b/api/app/core/tools/mcp/client.py @@ -99,7 +99,7 @@ class SimpleMCPClient: # 建立 SSE 连接 response = await self._session.get(self.server_url) - if response.status not in (200, 202): + if not (200 <= response.status < 300): error_text = await response.text() raise MCPConnectionError(f"SSE 连接失败 {response.status}: {error_text}") @@ -190,9 +190,7 @@ class SimpleMCPClient: try: async with self._session.post(self._endpoint_url, json=request) as response: - # MCP SSE 协议：POST 请求返回 200 或 202 均为正常 - # 202 Accepted 表示请求已接受，结果通过 SSE 流异步返回 - if response.status not in (200, 202): + if not (200 <= response.status < 300): error_text = await response.text() raise MCPConnectionError(f"请求失败 {response.status}: {error_text}") @@ -207,7 +205,7 @@ class SimpleMCPClient: raise MCPConnectionError("endpoint URL 未初始化") async with self._session.post(self._endpoint_url, json=notification) as response: - if response.status not in (200, 202): + if not (200 <= response.status < 300): logger.warning(f"通知发送失败: {response.status}") async def _initialize_modelscope_session(self): @@ -225,7 +223,7 @@ class SimpleMCPClient: try: async with self._session.post(self.server_url, json=init_request) as response: - if response.status != 200: + if not (200 <= response.status < 300): error_text = await response.text() raise MCPConnectionError(f"初始化失败 {response.status}: {error_text}") diff --git a/api/app/core/workflow/nodes/llm/node.py b/api/app/core/workflow/nodes/llm/node.py index a691001f..e3c68420 100644 --- a/api/app/core/workflow/nodes/llm/node.py +++ b/api/app/core/workflow/nodes/llm/node.py @@ -135,7 +135,8 @@ class LLMNode(BaseNode): api_key=model_info.api_key, base_url=model_info.api_base, extra_params=extra_params, - is_omni=model_info.is_omni + is_omni=model_info.is_omni, + support_thinking="thinking" in (model_info.capability or []), ), type=model_info.model_type ) diff --git a/api/app/core/workflow/nodes/parameter_extractor/node.py b/api/app/core/workflow/nodes/parameter_extractor/node.py index 3dc5fcc3..d7a2a501 100644 --- a/api/app/core/workflow/nodes/parameter_extractor/node.py +++ b/api/app/core/workflow/nodes/parameter_extractor/node.py @@ -109,6 +109,7 @@ class ParameterExtractorNode(BaseNode): api_key = api_config.api_key api_base = api_config.api_base is_omni = api_config.is_omni + capability = api_config.capability model_type = config.type llm = RedBearLLM( @@ -117,7 +118,8 @@ class ParameterExtractorNode(BaseNode): provider=provider, api_key=api_key, base_url=api_base, - is_omni=is_omni + is_omni=is_omni, + support_thinking="thinking" in (capability or []), ), type=ModelType(model_type) ) diff --git a/api/app/core/workflow/nodes/question_classifier/node.py b/api/app/core/workflow/nodes/question_classifier/node.py index 31fadaf6..ddae1ced 100644 --- a/api/app/core/workflow/nodes/question_classifier/node.py +++ b/api/app/core/workflow/nodes/question_classifier/node.py @@ -62,6 +62,7 @@ class QuestionClassifierNode(BaseNode): api_key = api_config.api_key base_url = api_config.api_base is_omni = api_config.is_omni + capability = api_config.capability model_type = config.type return RedBearLLM( @@ -70,7 +71,8 @@ class QuestionClassifierNode(BaseNode): provider=provider, api_key=api_key, base_url=base_url, - is_omni=is_omni + is_omni=is_omni, + support_thinking="thinking" in (capability or []), ), type=ModelType(model_type) ) diff --git a/api/app/models/models_model.py b/api/app/models/models_model.py index 69bedc3d..fab85ea6 100644 --- a/api/app/models/models_model.py +++ b/api/app/models/models_model.py @@ -81,7 +81,7 @@ class ModelConfig(BaseModel): # 模型配置参数 capability = Column(ARRAY(String), default=list, nullable=False, server_default=text("'{}'::varchar[]"), - comment="模型能力列表（如['vision', 'audio', 'video']）") + comment="模型能力列表（如['vision', 'audio', 'video', 'thinking']）") is_omni = Column(Boolean, default=False, nullable=False, server_default="false", comment="是否为Omni模型（使用特殊API调用）") config = Column(JSON, comment="模型配置参数") # - temperature : 控制生成文本的随机性。值越高，输出越随机、越有创造性；值越低，输出越确定、越保守。 diff --git a/api/app/repositories/home_page_repository.py b/api/app/repositories/home_page_repository.py index 6d74bcaf..d4eaddeb 100644 --- a/api/app/repositories/home_page_repository.py +++ b/api/app/repositories/home_page_repository.py @@ -1,6 +1,6 @@ -from datetime import datetime, timedelta +from datetime import datetime, time from sqlalchemy.orm import Session -from sqlalchemy import func +from sqlalchemy import func, Table, MetaData from uuid import UUID from typing import Dict, Optional, Any @@ -192,10 +192,63 @@ class HomePageRepository: return workspaces, app_count_dict, user_count_dict + @staticmethod + def get_latest_version_introduction(db: Session) -> tuple[Optional[str], Optional[Dict[str, Any]]]: + """ + 从数据库获取最新已发布的版本说明 + 使用反射方式读取表结构，不依赖 premium 模型类 + + Args: + db: 数据库会话 + + Returns: + (版本号，版本说明字典) 的元组 + 如果数据库中没有已发布的版本，返回 (None, None) + """ + try: + metadata = MetaData() + + version_notes = Table('version_notes', metadata, autoload_with=db.bind) + + # 获取最新已发布的版本（按发布时间倒序，日期相同时按版本号倒序） + query = db.query(version_notes).filter( + version_notes.c.is_published == True + ).order_by( + version_notes.c.release_date.desc(), + version_notes.c.version.desc() + ) + + note = query.first() + + if not note: + return None, None + + version_info = { + "introduction": { + "codeName": note.code_name or "", + "releaseDate": int(datetime.combine(note.release_date, time()).timestamp() * 1000) if note.release_date else 0, + "upgradePosition": note.upgrade_position or "", + "coreUpgrades": note.core_upgrades or [] + }, + "introduction_en": { + "codeName": note.code_name_en or note.code_name or "", + "releaseDate": int(datetime.combine(note.release_date, time()).timestamp() * 1000) if note.release_date else 0, + "upgradePosition": note.upgrade_position_en or note.upgrade_position or "", + "coreUpgrades": note.core_upgrades_en or [] + } + } + + return note.version, version_info + + except Exception as e: + import traceback + traceback.print_exc() + return None, None + @staticmethod def get_version_introduction(db: Session, version: str) -> Optional[Dict[str, Any]]: """ - 从数据库获取版本说明（优先读取已发布的版本） + 从数据库获取指定版本说明（优先读取已发布的版本） 使用反射方式读取表结构，不依赖 premium 模型类 Args: @@ -207,11 +260,8 @@ class HomePageRepository: 如果数据库中没有该版本，返回 None """ try: - from sqlalchemy import Table, MetaData - metadata = MetaData() version_notes = Table('version_notes', metadata, autoload_with=db.engine) - version_note_items = Table('version_note_items', metadata, autoload_with=db.engine) note = db.query(version_notes).filter( version_notes.c.version == version, @@ -221,31 +271,18 @@ class HomePageRepository: if not note: return None - items = db.query(version_note_items).filter( - version_note_items.c.note_id == note.id - ).order_by(version_note_items.c.sort_order).all() - - core_upgrades = [] - for item in items: - title = item.title - content = item.content - if content: - core_upgrades.append(f"{title}
{content}") - else: - core_upgrades.append(title) - return { "introduction": { - "codeName": "", - "releaseDate": note.release_date.isoformat() if note.release_date else "", - "upgradePosition": "", - "coreUpgrades": core_upgrades + "codeName": note.code_name or "", + "releaseDate": int(datetime.combine(note.release_date, time()).timestamp() * 1000) if note.release_date else 0, + "upgradePosition": note.upgrade_position or "", + "coreUpgrades": note.core_upgrades or [] }, "introduction_en": { - "codeName": "", - "releaseDate": note.release_date.isoformat() if note.release_date else "", - "upgradePosition": "", - "coreUpgrades": core_upgrades + "codeName": note.code_name_en or note.code_name or "", + "releaseDate": int(datetime.combine(note.release_date, time()).timestamp() * 1000) if note.release_date else 0, + "upgradePosition": note.upgrade_position_en or note.upgrade_position or "", + "coreUpgrades": note.core_upgrades_en or [] } } except Exception: diff --git a/api/app/schemas/app_schema.py b/api/app/schemas/app_schema.py index f1e9132f..4ca3e7de 100644 --- a/api/app/schemas/app_schema.py +++ b/api/app/schemas/app_schema.py @@ -241,6 +241,8 @@ class ModelParameters(BaseModel): presence_penalty: float = Field(default=0.0, ge=-2.0, le=2.0, description="存在惩罚") n: int = Field(default=1, ge=1, le=10, description="生成的回复数量") stop: Optional[List[str]] = Field(default=None, description="停止序列") + deep_thinking: bool = Field(default=False, description="是否启用深度思考模式（需模型支持，如 DeepSeek-R1、QwQ 等）") + thinking_budget_tokens: Optional[int] = Field(default=None, ge=1024, le=131072, description="深度思考 token 预算（仅部分模型支持）") class VariableDefinition(BaseModel): @@ -612,6 +614,7 @@ class AppChatRequest(BaseModel): user_id: Optional[str] = Field(default=None, description="用户ID（用于会话管理）") variables: Optional[Dict[str, Any]] = Field(default=None, description="自定义变量参数值") stream: bool = Field(default=False, description="是否流式返回") + thinking: bool = Field(default=False, description="是否启用深度思考（需Agent配置支持）") files: List[FileInput] = Field(default_factory=list, description="附件列表（支持多文件）") diff --git a/api/app/schemas/conversation_schema.py b/api/app/schemas/conversation_schema.py index 98715612..b2f565ef 100644 --- a/api/app/schemas/conversation_schema.py +++ b/api/app/schemas/conversation_schema.py @@ -31,6 +31,7 @@ class ChatRequest(BaseModel): stream: bool = Field(default=False, description="是否流式返回") web_search: bool = Field(default=False, description="是否启用网络搜索") memory: bool = Field(default=True, description="是否启用记忆功能") + thinking: bool = Field(default=False, description="是否启用深度思考（需Agent配置支持）") files: Optional[List[FileInput]] = Field(default=None, description="附件列表（支持多文件）") diff --git a/api/app/schemas/memory_api_schema.py b/api/app/schemas/memory_api_schema.py index 84a34e8a..ff62355f 100644 --- a/api/app/schemas/memory_api_schema.py +++ b/api/app/schemas/memory_api_schema.py @@ -138,21 +138,13 @@ class CreateEndUserRequest(BaseModel): """Request schema for creating an end user. Attributes: - workspace_id: Workspace ID (required) other_id: External user identifier (required) other_name: Display name for the end user + memory_config_id: Optional memory config ID. If not provided, uses workspace default. """ - workspace_id: str = Field(..., description="Workspace ID (required)") other_id: str = Field(..., description="External user identifier (required)") other_name: Optional[str] = Field("", description="Display name") - - @field_validator("workspace_id") - @classmethod - def validate_workspace_id(cls, v: str) -> str: - """Validate that workspace_id is not empty.""" - if not v or not v.strip(): - raise ValueError("workspace_id is required and cannot be empty") - return v.strip() + memory_config_id: Optional[str] = Field(None, description="Memory config ID. Falls back to workspace default if not provided.") @field_validator("other_id") @classmethod @@ -171,11 +163,13 @@ class CreateEndUserResponse(BaseModel): other_id: External user identifier other_name: Display name workspace_id: Workspace the user belongs to + memory_config_id: Connected memory config ID """ id: str = Field(..., description="End user UUID") other_id: str = Field(..., description="External user identifier") other_name: str = Field("", description="Display name") workspace_id: str = Field(..., description="Workspace ID") + memory_config_id: Optional[str] = Field(None, description="Connected memory config ID") class MemoryConfigItem(BaseModel): diff --git a/api/app/services/app_chat_service.py b/api/app/services/app_chat_service.py index df81568f..53ac577a 100644 --- a/api/app/services/app_chat_service.py +++ b/api/app/services/app_chat_service.py @@ -117,7 +117,9 @@ class AppChatService: 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( @@ -205,7 +207,8 @@ class AppChatService: "model": api_key_obj.model_name, "usage": result.get("usage", {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0}), "audio_url": None, - "citations": filtered_citations + "citations": filtered_citations, + "reasoning_content": result.get("reasoning_content") } if files: for f in files: @@ -258,6 +261,7 @@ class AppChatService: "conversation_id": conversation_id, "message_id": str(message_id), "message": result["content"], + "reasoning_content": result.get("reasoning_content"), "usage": result.get("usage", { "prompt_tokens": 0, "completion_tokens": 0, @@ -354,7 +358,10 @@ class AppChatService: max_tokens=model_parameters.get("max_tokens", 2000), system_prompt=system_prompt, tools=tools, - streaming=True + 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( @@ -403,6 +410,7 @@ class AppChatService: # 流式调用 Agent（支持多模态），同时并行启动 TTS full_content = "" + full_reasoning = "" total_tokens = 0 text_queue: asyncio.Queue = asyncio.Queue() @@ -426,6 +434,9 @@ class AppChatService: ): if isinstance(chunk, int): total_tokens = chunk + elif isinstance(chunk, dict) and chunk.get("type") == "reasoning": + full_reasoning += chunk['content'] + yield f"event: reasoning\ndata: {json.dumps({'content': chunk['content']}, ensure_ascii=False)}\n\n" else: full_content += chunk yield f"event: message\ndata: {json.dumps({'content': chunk}, ensure_ascii=False)}\n\n" @@ -472,7 +483,8 @@ class AppChatService: "model": api_key_obj.model_name, "usage": {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": total_tokens}, "audio_url": None, - "citations": filtered_citations + "citations": filtered_citations, + "reasoning_content": full_reasoning or None } if files: @@ -652,13 +664,13 @@ class AppChatService: storage_type=storage_type, user_rag_memory_id=user_rag_memory_id ): - if "sub_usage" in event: + # 拦截 sub_usage 事件，累加 token + if "event: sub_usage" in event: if "data:" in event: try: data_line = event.split("data: ", 1)[1].strip() data = json.loads(data_line) - if "total_tokens" in data: - total_tokens += data["total_tokens"] + total_tokens += data.get("total_tokens", 0) except: pass else: diff --git a/api/app/services/app_service.py b/api/app/services/app_service.py index 377f9479..5af6bf41 100644 --- a/api/app/services/app_service.py +++ b/api/app/services/app_service.py @@ -13,7 +13,7 @@ import uuid from typing import Annotated, Any, Dict, List, Optional, Tuple from fastapi import Depends -from sqlalchemy import and_, delete, func, or_, select +from sqlalchemy import and_, delete, func, or_, select, update as sa_update from sqlalchemy.orm import Session from app.core.error_codes import BizCode @@ -757,6 +757,17 @@ class AppService: # 逻辑删除应用 app.is_active = False + + # 更新 app_shares 表中该应用的所有共享记录为失效状态，并更新 updated_at 时间 + stmt = sa_update(AppShare).where( + AppShare.source_app_id == app_id, + AppShare.is_active.is_(True) + ).values( + is_active=False, + updated_at=datetime.datetime.now() + ) + self.db.execute(stmt) + self.db.commit() logger.info( diff --git a/api/app/services/conversation_service.py b/api/app/services/conversation_service.py index bd7f7496..6e9f3544 100644 --- a/api/app/services/conversation_service.py +++ b/api/app/services/conversation_service.py @@ -534,6 +534,7 @@ class ConversationService: api_key = api_config.api_key api_base = api_config.api_base is_omni = api_config.is_omni + capability = api_config.capability model_type = config.type llm = RedBearLLM( @@ -542,7 +543,8 @@ class ConversationService: provider=provider, api_key=api_key, base_url=api_base, - is_omni=is_omni + is_omni=is_omni, + support_thinking="thinking" in (capability or []), ), type=ModelType(model_type) ) diff --git a/api/app/services/draft_run_service.py b/api/app/services/draft_run_service.py index 4b503f2b..978dfdab 100644 --- a/api/app/services/draft_run_service.py +++ b/api/app/services/draft_run_service.py @@ -458,7 +458,7 @@ class AgentRunService: statement = opening["statement"] suggested_questions = opening["suggested_questions"] - + # 如果有变量，进行替换（仅支持 {{var_name}} 格式） if variables: for var_name, var_value in variables.items(): @@ -595,6 +595,9 @@ class AgentRunService: 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 @@ -689,7 +692,8 @@ class AgentRunService: "prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0 - }) + }), + "reasoning_content": result.get("reasoning_content") }, files=files, processed_files=processed_files, @@ -701,6 +705,7 @@ class AgentRunService: response = { "message": result["content"], + "reasoning_content": result.get("reasoning_content"), "conversation_id": conversation_id, "usage": result.get("usage", { "prompt_tokens": 0, @@ -838,7 +843,10 @@ class AgentRunService: max_tokens=effective_params.get("max_tokens", 2000), system_prompt=system_prompt, tools=tools, - streaming=True + 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 @@ -898,6 +906,7 @@ class AgentRunService: # 9. 流式调用 Agent（支持多模态），同时并行启动 TTS full_content = "" + full_reasoning = "" total_tokens = 0 # 启动流式 TTS（文本边输出边合成） @@ -916,6 +925,9 @@ class AgentRunService: ): if isinstance(chunk, int): total_tokens = chunk + elif isinstance(chunk, dict) and chunk.get("type") == "reasoning": + full_reasoning += chunk["content"] + yield self._format_sse_event("reasoning", {"content": chunk["content"]}) else: full_content += chunk yield self._format_sse_event("message", {"content": chunk}) @@ -944,7 +956,8 @@ class AgentRunService: app_id=agent_config.app_id, user_id=user_id, meta_data={ - "usage": {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": total_tokens} + "usage": {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": total_tokens}, + "reasoning_content": full_reasoning or None }, files=files, processed_files=processed_files, @@ -1665,7 +1678,7 @@ class AgentRunService: """从 text_queue 取文本按句子切分后喂给 synthesizer""" import re buf = "" - sentence_end = re.compile(r'[\u3002\uff01\uff1f\.!?\n]') + sentence_end = re.compile(r'[\u3002\uff01\uff1f.!?\n]') while True: chunk = await text_queue.get() if chunk is None: @@ -1894,6 +1907,7 @@ class AgentRunService: "conversation_id": result['conversation_id'], "parameters_used": model_info["parameters"], "message": result.get("message"), + "reasoning_content": result.get("reasoning_content"), "usage": usage, "elapsed_time": elapsed, "tokens_per_second": ( @@ -2012,7 +2026,7 @@ class AgentRunService: # 需要从 ModelApiKey 获取实际的模型名称，或者在 ModelConfig 中添加 model 字段 return None - def _with_parameters(self, agent_config: AgentConfig, parameters: Dict[str, Any]) -> AgentConfig: + def _with_parameters(self, agent_config: AgentConfig, parameters: Dict[str, Any]) -> tuple[AgentConfig, Any]: """创建一个带有覆盖参数的 agent_config（浅拷贝，只修改 model_parameters） Args: @@ -2110,6 +2124,7 @@ class AgentRunService: start_time = time.time() full_content = "" + full_reasoning = "" returned_conversation_id = model_conversation_id audio_url = None audio_status = None @@ -2168,6 +2183,18 @@ class AgentRunService: "content": chunk })) + # 转发深度思考事件（带模型标识） + if event_type == "reasoning" and event_data: + reasoning_chunk = event_data.get("content", "") + full_reasoning += reasoning_chunk + await event_queue.put(self._format_sse_event("model_reasoning", { + "model_index": idx, + "model_config_id": model_config_id, + "label": model_label, + "conversation_id": returned_conversation_id, + "content": event_data.get("content", "") + })) + # 从 end 事件中提取 features 输出字段 if event_type == "end" and event_data: audio_url = event_data.get("audio_url") @@ -2199,6 +2226,7 @@ class AgentRunService: "conversation_id": returned_conversation_id, "parameters_used": model_info["parameters"], "message": full_content, + "reasoning_content": full_reasoning or None, "elapsed_time": elapsed, "audio_url": audio_url, "audio_status": audio_status, @@ -2351,6 +2379,7 @@ class AgentRunService: "label": r["label"], "conversation_id": r.get("conversation_id"), "message": r.get("message"), + "reasoning_content": r.get("reasoning_content"), "elapsed_time": r.get("elapsed_time", 0), "audio_url": r.get("audio_url"), "audio_status": r.get("audio_status"), diff --git a/api/app/services/llm_router.py b/api/app/services/llm_router.py index 02895d6b..7087415e 100644 --- a/api/app/services/llm_router.py +++ b/api/app/services/llm_router.py @@ -415,6 +415,7 @@ 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 ) diff --git a/api/app/services/master_agent_router.py b/api/app/services/master_agent_router.py index b0f43b51..206443bd 100644 --- a/api/app/services/master_agent_router.py +++ b/api/app/services/master_agent_router.py @@ -393,6 +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 []), extra_params = extra_params ) @@ -403,6 +404,17 @@ class MasterAgentRouter: response = await llm.ainvoke(prompt) ModelApiKeyService.record_api_key_usage(self.db, api_key_config.id) + # 提取 token 消耗 + self._last_routing_tokens = 0 + if hasattr(response, 'usage_metadata') and response.usage_metadata: + um = response.usage_metadata + self._last_routing_tokens = um.get("total_tokens", 0) if isinstance(um, dict) else getattr(um, "total_tokens", 0) + elif hasattr(response, 'response_metadata') and response.response_metadata: + token_usage = response.response_metadata.get("token_usage") or response.response_metadata.get("usage", {}) + if isinstance(token_usage, dict): + self._last_routing_tokens = token_usage.get("total_tokens", 0) + logger.info(f"Master Agent 路由 token 消耗: {self._last_routing_tokens}") + # 提取响应内容 if hasattr(response, 'content'): return response.content diff --git a/api/app/services/memory_dashboard_service.py b/api/app/services/memory_dashboard_service.py index 791a6fe8..b96f4bde 100644 --- a/api/app/services/memory_dashboard_service.py +++ b/api/app/services/memory_dashboard_service.py @@ -353,15 +353,13 @@ async def get_workspace_total_memory_count( "details": [] } - # 2. 对每个 host_id 调用 search_all 获取 total + # 2. 使用 search_all_batch 批量查询所有宿主的记忆数量 from app.services import memory_storage_service - total_count = 0 - details = [] - # 如果提供了 end_user_id，只查询该用户 if end_user_id: - search_result = await memory_storage_service.search_all(end_user_id=end_user_id) + batch_result = await memory_storage_service.search_all_batch([end_user_id]) + count = batch_result.get(end_user_id, 0) # 查询用户名称 from app.repositories.end_user_repository import EndUserRepository repo = EndUserRepository(db) @@ -369,42 +367,31 @@ async def get_workspace_total_memory_count( user_name = end_user.other_name if end_user else None return { - "total_memory_count": search_result.get("total", 0), + "total_memory_count": count, "host_count": 1, "details": [{ "end_user_id": end_user_id, - "count": search_result.get("total", 0), + "count": count, "name": user_name }] } - for host in hosts: - try: - end_user_id_str = str(host.id) - - search_result = await memory_storage_service.search_all( - end_user_id=end_user_id_str - ) - - host_total = search_result.get("total", 0) - total_count += host_total - - details.append({ - "end_user_id": end_user_id_str, - "count": host_total, - "name": host.other_name # 使用 other_name 字段 - }) - - business_logger.debug(f"EndUser {end_user_id_str} ({host.other_name}) 记忆数: {host_total}") - - except Exception as e: - business_logger.warning(f"获取 end_user {host.id} 记忆数失败: {str(e)}") - # 失败的 host 记为 0 - details.append({ - "end_user_id": str(host.id), - "count": 0, - "name": host.other_name # 使用 other_name 字段 - }) + # 批量查询所有宿主记忆数量（一次 Neo4j 查询） + end_user_ids = [str(host.id) for host in hosts] + batch_result = await memory_storage_service.search_all_batch(end_user_ids) + + # 构建 host name 映射 + host_name_map = {str(host.id): host.other_name for host in hosts} + + total_count = sum(batch_result.values()) + details = [ + { + "end_user_id": uid, + "count": batch_result.get(uid, 0), + "name": host_name_map.get(uid) + } + for uid in end_user_ids + ] result = { "total_memory_count": total_count, @@ -519,6 +506,206 @@ def get_rag_user_kb_total_chunk( business_logger.error(f"获取用户知识库总chunk数失败: workspace_id={workspace_id} - {str(e)}") raise +def get_dashboard_yesterday_changes( + db: Session, + workspace_id: uuid.UUID, + storage_type: str, + today_data: dict +) -> dict: + """ + 计算各指标相比昨天的变化量。 + + Args: + db: 数据库会话 + workspace_id: 工作空间ID + storage_type: 存储类型 'neo4j' | 'rag' + today_data: 当前数据，包含 total_memory, total_app, total_knowledge, total_api_call + + Returns: + { + "total_memory_change": int | None, + "total_app_change": int | None, + "total_knowledge_change": int | None, + "total_api_call_change": int | None + } + """ + from datetime import datetime, timedelta + from sqlalchemy import func + from app.models.api_key_model import ApiKey, ApiKeyLog + from app.models.knowledge_model import Knowledge + from app.models.app_model import App + from app.models.appshare_model import AppShare + + business_logger.info(f"计算昨日对比: workspace_id={workspace_id}, storage_type={storage_type}") + + now_local = datetime.now() + today_start = now_local.replace(hour=0, minute=0, second=0, microsecond=0) + yesterday_start = today_start - timedelta(days=1) + + changes = { + "total_memory_change": None, + "total_app_change": None, + "total_knowledge_change": None, + "total_api_call_change": None, + } + + # --- total_api_call_change --- + try: + # 获取该workspace下所有api_key的id + api_key_ids = [ + row[0] for row in db.query(ApiKey.id).filter( + ApiKey.workspace_id == workspace_id + ).all() + ] + if api_key_ids: + # 今日累计 + today_api_count = db.query(func.count(ApiKeyLog.id)).filter( + ApiKeyLog.api_key_id.in_(api_key_ids), + ApiKeyLog.created_at >= today_start, + ApiKeyLog.created_at < now_local + ).scalar() or 0 + # 昨日全天 + yesterday_api_count = db.query(func.count(ApiKeyLog.id)).filter( + ApiKeyLog.api_key_id.in_(api_key_ids), + ApiKeyLog.created_at >= yesterday_start, + ApiKeyLog.created_at < today_start + ).scalar() or 0 + changes["total_api_call_change"] = today_api_count - yesterday_api_count + else: + # 没有api_key，如果今日也是0则无对比意义 + changes["total_api_call_change"] = None + except Exception as e: + business_logger.warning(f"计算API调用昨日对比失败: {str(e)}") + + # --- total_knowledge_change --- + try: + # 今天有效总量：当前status=1的知识库总数，排除用户知识库(permission_id='Memory') + today_knowledge = db.query(func.count(Knowledge.id)).filter( + Knowledge.workspace_id == workspace_id, + Knowledge.status == 1, + Knowledge.permission_id != "Memory" + ).scalar() or 0 + # 昨日有效总量：昨天之前创建的、当前仍有效的知识库，排除用户知识库 + yesterday_knowledge = db.query(func.count(Knowledge.id)).filter( + Knowledge.workspace_id == workspace_id, + Knowledge.status == 1, + Knowledge.permission_id != "Memory", + Knowledge.created_at < today_start + ).scalar() or 0 + # 今日软删：今天被软删的知识库(status=2 且 updated_at >= today_start)，排除用户知识库 + today_deleted_knowledge = db.query(func.count(Knowledge.id)).filter( + Knowledge.workspace_id == workspace_id, + Knowledge.status == 2, + Knowledge.permission_id != "Memory", + Knowledge.updated_at >= today_start + ).scalar() or 0 + + if yesterday_knowledge == 0 and today_knowledge == 0 and today_deleted_knowledge == 0: + changes["total_knowledge_change"] = None + else: + # change = 今天有效总量 - 今日软删 - 昨日有效总量 + changes["total_knowledge_change"] = today_knowledge - today_deleted_knowledge - yesterday_knowledge + except Exception as e: + business_logger.warning(f"计算知识库昨日对比失败: {str(e)}") + + # --- total_app_change --- + try: + # === 自有app === + # 今天有效总量 + today_own_apps = db.query(func.count(App.id)).filter( + App.workspace_id == workspace_id, + App.is_active == True + ).scalar() or 0 + # 昨日有效总量 + yesterday_own_apps = db.query(func.count(App.id)).filter( + App.workspace_id == workspace_id, + App.is_active == True, + App.created_at < today_start + ).scalar() or 0 + # 今日软删 + today_deleted_own_apps = db.query(func.count(App.id)).filter( + App.workspace_id == workspace_id, + App.is_active == False, + App.updated_at >= today_start + ).scalar() or 0 + + # === 被分享app === + # 今天有效总量 + today_shared_apps = db.query(func.count(AppShare.id)).filter( + AppShare.target_workspace_id == workspace_id, + AppShare.is_active == True + ).scalar() or 0 + # 昨日有效总量 + yesterday_shared_apps = db.query(func.count(AppShare.id)).filter( + AppShare.target_workspace_id == workspace_id, + AppShare.is_active == True, + AppShare.created_at < today_start + ).scalar() or 0 + # 今日软删 + today_deleted_shared_apps = db.query(func.count(AppShare.id)).filter( + AppShare.target_workspace_id == workspace_id, + AppShare.is_active == False, + AppShare.updated_at >= today_start + ).scalar() or 0 + + today_total_app = today_own_apps + today_shared_apps + yesterday_total_app = yesterday_own_apps + yesterday_shared_apps + total_deleted = today_deleted_own_apps + today_deleted_shared_apps + + if yesterday_total_app == 0 and today_total_app == 0 and total_deleted == 0: + changes["total_app_change"] = None + else: + # change = 今天有效总量 - 今日软删 - 昨日有效总量 + changes["total_app_change"] = today_total_app - total_deleted - yesterday_total_app + except Exception as e: + business_logger.warning(f"计算应用数量昨日对比失败: {str(e)}") + + # --- total_memory_change --- + try: + today_memory = today_data.get("total_memory") + if today_memory is None: + changes["total_memory_change"] = None + elif storage_type == "neo4j": + # 从 memory_increments 取最近一条 created_at < today_start 的记录 + last_record = db.query(MemoryIncrement).filter( + MemoryIncrement.workspace_id == workspace_id, + MemoryIncrement.created_at < today_start + ).order_by(desc(MemoryIncrement.created_at)).first() + if last_record is None: + changes["total_memory_change"] = None + else: + changes["total_memory_change"] = today_memory - last_record.total_num + elif storage_type == "rag": + # RAG: 查 documents 表中 created_at < today_start 的 chunk_num 之和 + from app.models.document_model import Document + from app.models.end_user_model import EndUser as _EndUser + from app.models.app_model import App as _App + + end_user_ids = [ + str(eid) for (eid,) in db.query(_EndUser.id) + .join(_App, _EndUser.app_id == _App.id) + .filter(_App.workspace_id == workspace_id) + .all() + ] + if not end_user_ids: + changes["total_memory_change"] = None + else: + file_names = [f"{uid}.txt" for uid in end_user_ids] + yesterday_chunk = db.query(func.sum(Document.chunk_num)).filter( + Document.file_name.in_(file_names), + Document.created_at < today_start + ).scalar() + if yesterday_chunk is None: + changes["total_memory_change"] = None + else: + changes["total_memory_change"] = today_memory - int(yesterday_chunk) + except Exception as e: + business_logger.warning(f"计算记忆总量昨日对比失败: {str(e)}") + + business_logger.info(f"昨日对比计算完成: {changes}") + return changes + + def get_current_user_total_chunk( end_user_id: str, db: Session, @@ -881,4 +1068,66 @@ async def generate_rag_profile( "tags_count": len(tags), "personas_count": len(personas), "insight_generated": bool(insight_sections.get("memory_insight")), - } \ No newline at end of file + } + + +def get_dashboard_common_stats(db: Session, workspace_id) -> dict: + """ + 获取 dashboard 中 neo4j/rag 分支共享的统计数据： + total_app、total_knowledge、total_api_call + + Returns: + dict: {"total_app": int, "total_knowledge": int, "total_api_call": int} + """ + result = {"total_app": 0, "total_knowledge": 0, "total_api_call": 0} + + # total_app: 统计当前空间下的所有app数量（包含自有 + 被分享给本工作空间的app） + try: + from app.services import app_service as _app_svc + _, total_app = _app_svc.AppService(db).list_apps( + workspace_id=workspace_id, include_shared=True, pagesize=1 + ) + result["total_app"] = total_app + except Exception as e: + business_logger.warning(f"获取应用数量失败: {e}") + + # total_knowledge: 统计顶层知识库（parent_id = workspace_id） + try: + from sqlalchemy import func as _func + from app.models.knowledge_model import Knowledge as _Knowledge + total_knowledge = db.query(_func.count(_Knowledge.id)).filter( + _Knowledge.workspace_id == workspace_id, + _Knowledge.status == 1, + _Knowledge.parent_id == _Knowledge.workspace_id + ).scalar() or 0 + result["total_knowledge"] = total_knowledge + except Exception as e: + business_logger.warning(f"获取知识库数量失败: {e}") + + # total_api_call: 仅统计当天 api_key_log 调用次数 + try: + from datetime import datetime + from sqlalchemy import func as _api_func + from app.models.api_key_model import ApiKey as _ApiKey, ApiKeyLog as _ApiKeyLog + + _now = datetime.now() + _today_start = _now.replace(hour=0, minute=0, second=0, microsecond=0) + + _api_key_ids = [ + row[0] for row in db.query(_ApiKey.id).filter( + _ApiKey.workspace_id == workspace_id + ).all() + ] + if _api_key_ids: + total_api_calls = db.query(_api_func.count(_ApiKeyLog.id)).filter( + _ApiKeyLog.api_key_id.in_(_api_key_ids), + _ApiKeyLog.created_at >= _today_start, + _ApiKeyLog.created_at < _now + ).scalar() or 0 + else: + total_api_calls = 0 + result["total_api_call"] = total_api_calls + except Exception as e: + business_logger.warning(f"获取API调用统计失败: {e}") + + return result diff --git a/api/app/services/memory_perceptual_service.py b/api/app/services/memory_perceptual_service.py index 7cf94a1a..7d6d1092 100644 --- a/api/app/services/memory_perceptual_service.py +++ b/api/app/services/memory_perceptual_service.py @@ -232,7 +232,8 @@ class MemoryPerceptualService: provider=model_config.provider, api_key=model_config.api_key, base_url=model_config.api_base, - is_omni=model_config.is_omni + is_omni=model_config.is_omni, + support_thinking="thinking" in (model_config.capability or []), ) ) return llm, model_config diff --git a/api/app/services/memory_storage_service.py b/api/app/services/memory_storage_service.py index b3a66734..132370b6 100644 --- a/api/app/services/memory_storage_service.py +++ b/api/app/services/memory_storage_service.py @@ -613,37 +613,6 @@ async def search_entity(end_user_id: Optional[str] = None) -> Dict[str, Any]: return data -async def search_all(end_user_id: Optional[str] = None) -> Dict[str, Any]: - result = await _neo4j_connector.execute_query( - MemoryConfigRepository.SEARCH_FOR_ALL, - end_user_id=end_user_id, - ) - - # 检查结果是否为空或长度不足 - if not result or len(result) < 4: - data = { - "total": 0, - "counts": { - "dialogue": 0, - "chunk": 0, - "statement": 0, - "entity": 0, - }, - } - return data - - data = { - "total": result[-1]["Count"], - "counts": { - "dialogue": result[0]["Count"], - "chunk": result[1]["Count"], - "statement": result[2]["Count"], - "entity": result[3]["Count"], - }, - } - return data - - async def kb_type_distribution(end_user_id: Optional[str] = None) -> Dict[str, Any]: """统一知识库类型分布接口。 diff --git a/api/app/services/model_parameter_merger.py b/api/app/services/model_parameter_merger.py index 262e3d49..4be83851 100644 --- a/api/app/services/model_parameter_merger.py +++ b/api/app/services/model_parameter_merger.py @@ -45,12 +45,20 @@ class ModelParameterMerger: "frequency_penalty": 0.0, "presence_penalty": 0.0, "n": 1, - "stop": None + "stop": None, + "deep_thinking": False, + "thinking_budget_tokens": None } # 合并参数：默认值 -> 模型配置 -> Agent 配置 merged = default_params.copy() + # Pydantic 对象转为 dict + if model_config_params and hasattr(model_config_params, 'model_dump'): + model_config_params = model_config_params.model_dump() + if agent_config_params and hasattr(agent_config_params, 'model_dump'): + agent_config_params = agent_config_params.model_dump() + # 应用模型配置参数 if model_config_params: for key in default_params: diff --git a/api/app/services/model_service.py b/api/app/services/model_service.py index c9266667..4cbb3509 100644 --- a/api/app/services/model_service.py +++ b/api/app/services/model_service.py @@ -85,15 +85,16 @@ class ModelConfigService: @staticmethod async def validate_model_config( - db: Session, - *, - model_name: str, - provider: str, - api_key: str, - api_base: Optional[str] = None, - model_type: str = "llm", - test_message: str = "Hello", - is_omni: bool = False + db: Session, + *, + model_name: str, + provider: str, + api_key: str, + api_base: Optional[str] = None, + model_type: str = "llm", + test_message: str = "Hello", + is_omni: bool = False, + capability: Optional[list] = None ) -> Dict[str, Any]: """验证模型配置是否有效 @@ -124,6 +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 ) @@ -320,7 +322,8 @@ class ModelConfigService: api_base=api_key_data.api_base, model_type=model_data.type, test_message="Hello", - is_omni=model_data.is_omni + is_omni=model_data.is_omni, + capability=model_data.capability ) if not validation_result["valid"]: raise BusinessException( @@ -590,7 +593,8 @@ class ModelApiKeyService: api_base=data.api_base, model_type=model_config.type, test_message="Hello", - is_omni=data.is_omni + is_omni=data.is_omni, + capability=model_config.capability ) if not validation_result["valid"]: # 记录验证失败的模型，但不抛出异常 @@ -675,7 +679,8 @@ class ModelApiKeyService: api_base=api_key_data.api_base, model_type=model_config.type, test_message="Hello", - is_omni=api_key_data.is_omni + is_omni=api_key_data.is_omni, + capability=model_config.capability ) if not validation_result["valid"]: raise BusinessException( @@ -707,7 +712,8 @@ class ModelApiKeyService: api_base=api_key_data.api_base or existing_api_key.api_base, model_type=model_config.type, test_message="Hello", - is_omni=model_config.is_omni + is_omni=model_config.is_omni, + capability=model_config.capability ) if not validation_result["valid"]: raise BusinessException( diff --git a/api/app/services/multi_agent_orchestrator.py b/api/app/services/multi_agent_orchestrator.py index 60a3b5b8..216aeb6e 100644 --- a/api/app/services/multi_agent_orchestrator.py +++ b/api/app/services/multi_agent_orchestrator.py @@ -287,6 +287,11 @@ class MultiAgentOrchestrator: sub_conversation_id = None total_tokens = 0 + # 累加 Master Agent 路由决策消耗的 token + total_tokens += task_analysis.get("routing_tokens", 0) + # 累加 Master Agent 整合消耗的 token + total_tokens += getattr(self, '_last_merge_tokens', 0) + if isinstance(results, dict): sub_conversation_id = results.get("conversation_id") or results.get("result", {}).get("conversation_id") # 提取 token 信息 @@ -358,12 +363,16 @@ class MultiAgentOrchestrator: variables=variables ) + # 获取路由决策消耗的 token + routing_tokens = getattr(self.router, '_last_routing_tokens', 0) + logger.info( "Master Agent 分析完成", extra={ "selected_agent": routing_decision.get("selected_agent_id"), "confidence": routing_decision.get("confidence"), - "strategy": routing_decision.get("strategy") + "strategy": routing_decision.get("strategy"), + "routing_tokens": routing_tokens } ) @@ -372,7 +381,8 @@ class MultiAgentOrchestrator: "variables": variables or {}, "sub_agents": self.config.sub_agents, "initial_context": variables or {}, - "routing_decision": routing_decision + "routing_decision": routing_decision, + "routing_tokens": routing_tokens } async def _execute_sequential( @@ -1032,6 +1042,11 @@ class MultiAgentOrchestrator: # 5. 流式执行子 Agent sub_conversation_id = None + # Master Agent 路由决策消耗的 token，通过 sub_usage 事件发送给上层 + routing_tokens = task_analysis.get("routing_tokens", 0) + if routing_tokens > 0: + yield self._format_sse_event("sub_usage", {"total_tokens": routing_tokens}) + async for event in self._execute_sub_agent_stream( agent_data["config"], message, @@ -1054,6 +1069,7 @@ class MultiAgentOrchestrator: except: pass + # 直接透传所有事件（包括 sub_usage），累加统一由上层处理 yield event # 6. 如果有会话 ID，发送一个包含它的事件 @@ -2600,6 +2616,7 @@ 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 ) @@ -2612,6 +2629,17 @@ class MultiAgentOrchestrator: ModelApiKeyService.record_api_key_usage(self.db, api_key_config.id) + # 提取整合消耗的 token + merge_tokens = 0 + if hasattr(response, 'usage_metadata') and response.usage_metadata: + um = response.usage_metadata + merge_tokens = um.get("total_tokens", 0) if isinstance(um, dict) else getattr(um, "total_tokens", 0) + elif hasattr(response, 'response_metadata') and response.response_metadata: + token_usage = response.response_metadata.get("token_usage") or response.response_metadata.get("usage", {}) + if isinstance(token_usage, dict): + merge_tokens = token_usage.get("total_tokens", 0) + self._last_merge_tokens = merge_tokens + # 提取响应内容 if hasattr(response, 'content'): merged_response = response.content @@ -2621,7 +2649,8 @@ class MultiAgentOrchestrator: logger.info( "Master Agent 整合完成", extra={ - "merged_length": len(merged_response) + "merged_length": len(merged_response), + "merge_tokens": merge_tokens } ) @@ -2766,6 +2795,7 @@ 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 diff --git a/api/app/services/prompt_optimizer_service.py b/api/app/services/prompt_optimizer_service.py index 184220a8..fde8c4f9 100644 --- a/api/app/services/prompt_optimizer_service.py +++ b/api/app/services/prompt_optimizer_service.py @@ -185,7 +185,8 @@ class PromptOptimizerService: provider=api_config.provider, api_key=api_config.api_key, base_url=api_config.api_base, - is_omni=api_config.is_omni + is_omni=api_config.is_omni, + support_thinking="thinking" in (api_config.capability or []), ), type=ModelType(model_config.type)) try: prompt_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'prompt') diff --git a/api/app/services/shared_chat_service.py b/api/app/services/shared_chat_service.py index c74604a5..b1e40a2d 100644 --- a/api/app/services/shared_chat_service.py +++ b/api/app/services/shared_chat_service.py @@ -248,7 +248,9 @@ class SharedChatService: 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 [], ) # 加载历史消息 @@ -450,7 +452,10 @@ class SharedChatService: max_tokens=model_parameters.get("max_tokens", 2000), system_prompt=system_prompt, tools=tools, - streaming=True + 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 [], ) # 加载历史消息 @@ -479,6 +484,8 @@ class SharedChatService: ): if isinstance(chunk, int): total_tokens = chunk + elif isinstance(chunk, dict) and chunk.get("type") == "reasoning": + yield f"event: reasoning\ndata: {json.dumps({'content': chunk['content']}, ensure_ascii=False)}\n\n" else: full_content += chunk # 发送消息块事件 diff --git a/api/app/tasks.py b/api/app/tasks.py index 72421a5f..f918743c 100644 --- a/api/app/tasks.py +++ b/api/app/tasks.py @@ -61,9 +61,9 @@ def _get_or_create_redis_pool() -> redis.ConnectionPool | None: db=settings.REDIS_DB_CELERY_BACKEND, password=settings.REDIS_PASSWORD, decode_responses=True, - max_connections=10, + max_connections=100, socket_connect_timeout=5, - socket_timeout=5, + socket_timeout=10, retry_on_timeout=True, health_check_interval=30, ) @@ -1100,7 +1100,7 @@ def read_message_task(self, end_user_id: str, message: str, history: List[Dict[s } -@celery_app.task(name="app.core.memory.agent.write_message", bind=True) +@celery_app.task(name="app.core.memory.agent.write_message", bind=True, acks_late=False) def write_message_task( self, end_user_id: str, @@ -1176,6 +1176,7 @@ def write_message_task( redis_client = get_sync_redis_client() lock = None + loop = None if redis_client is not None: lock = RedisFairLock( key=f"memory_write:{end_user_id}", @@ -1196,6 +1197,7 @@ def write_message_task( } try: + task_start_time = int(time.time()) loop = set_asyncio_event_loop() result = loop.run_until_complete(_run()) @@ -1205,7 +1207,7 @@ def write_message_task( f"- elapsed_time={elapsed_time:.2f}s, task_id={self.request.id}") try: - _r = get_sync_redis_client() + _r = redis_client if _r is not None: from datetime import timezone as _tz _now_utc = datetime.now(_tz.utc).isoformat() @@ -1219,6 +1221,7 @@ def write_message_task( return { "status": "SUCCESS", "result": result, + "start_at": task_start_time, "end_user_id": end_user_id, "config_id": config_id, "elapsed_time": elapsed_time, @@ -1252,7 +1255,8 @@ def write_message_task( logger.warning(f"[CELERY WRITE] 释放锁失败: {e}") # Gracefully shutdown the event loop to prevent # 'RuntimeError: Event loop is closed' from httpx.AsyncClient.__del__ - _shutdown_loop_gracefully(loop) + if loop: + _shutdown_loop_gracefully(loop) # unused task @@ -1320,7 +1324,7 @@ def write_total_memory_task(workspace_id: str) -> Dict[str, Any]: from app.models.app_model import App from app.models.end_user_model import EndUser from app.repositories.memory_increment_repository import write_memory_increment - from app.services.memory_storage_service import search_all + from app.services.memory_storage_service import search_all_batch with get_db_context() as db: try: @@ -1354,27 +1358,15 @@ def write_total_memory_task(workspace_id: str) -> Dict[str, Any]: EndUser.workspace_id == workspace_id ).distinct().all() - # 3. 遍历所有end_user，查询每个宿主的记忆总量并累加 - total_num = 0 - end_user_details = [] + # 3. 批量查询所有宿主的记忆总量 + end_user_id_list = [str(eid) for (eid,) in end_users] + batch_result = await search_all_batch(end_user_id_list) - for (end_user_id,) in end_users: - try: - # 调用 search_all 接口查询该宿主的总量 - result = await search_all(str(end_user_id)) - user_total = result.get("total", 0) - total_num += user_total - end_user_details.append({ - "end_user_id": str(end_user_id), - "total": user_total - }) - except Exception as e: - # 记录单个用户查询失败，但继续处理其他用户 - end_user_details.append({ - "end_user_id": str(end_user_id), - "total": 0, - "error": str(e) - }) + total_num = sum(batch_result.values()) + end_user_details = [ + {"end_user_id": uid, "total": batch_result.get(uid, 0)} + for uid in end_user_id_list + ] # 4. 写入数据库 memory_increment = write_memory_increment( @@ -1437,7 +1429,7 @@ def write_all_workspaces_memory_task(self) -> Dict[str, Any]: from app.models.end_user_model import EndUser from app.models.workspace_model import Workspace from app.repositories.memory_increment_repository import write_memory_increment - from app.services.memory_storage_service import search_all + from app.services.memory_storage_service import search_all_batch with get_db_context() as db: try: @@ -1495,28 +1487,15 @@ def write_all_workspaces_memory_task(self) -> Dict[str, Any]: EndUser.workspace_id == workspace_id ).distinct().all() - # 3. 遍历所有end_user，查询每个宿主的记忆总量并累加 - total_num = 0 - end_user_details = [] + # 3. 批量查询所有宿主的记忆总量 + end_user_id_list = [str(eid) for (eid,) in end_users] + batch_result = await search_all_batch(end_user_id_list) - for (end_user_id,) in end_users: - try: - # 调用 search_all 接口查询该宿主的总量 - result = await search_all(str(end_user_id)) - user_total = result.get("total", 0) - total_num += user_total - end_user_details.append({ - "end_user_id": str(end_user_id), - "total": user_total - }) - except Exception as e: - # 记录单个用户查询失败，但继续处理其他用户 - logger.warning(f"查询用户 {end_user_id} 记忆失败: {str(e)}") - end_user_details.append({ - "end_user_id": str(end_user_id), - "total": 0, - "error": str(e) - }) + total_num = sum(batch_result.values()) + end_user_details = [ + {"end_user_id": uid, "total": batch_result.get(uid, 0)} + for uid in end_user_id_list + ] # 4. 写入数据库 memory_increment = write_memory_increment( diff --git a/api/app/utils/performance_timer.py b/api/app/utils/performance_timer.py index 6b0ec5d6..04e52fb1 100644 --- a/api/app/utils/performance_timer.py +++ b/api/app/utils/performance_timer.py @@ -6,13 +6,13 @@ """ import time -from contextlib import contextmanager +from contextlib import asynccontextmanager, contextmanager from app.core.logging_config import get_api_logger # 获取API专用日志器 api_logger = get_api_logger() - +# 同步的上下文管理器，使用@contextmanager修饰 @contextmanager def timer(label: str, user_count: int = 0): """上下文管理器：用于测量代码块执行时间 @@ -35,3 +35,27 @@ def timer(label: str, user_count: int = 0): elapsed = (time.perf_counter() - start) * 1000 # 转换为毫秒 extra_info = f", 用户数: {user_count}" if user_count > 0 else "" api_logger.info(f"[性能统计] {label}: {elapsed:.2f}ms{extra_info}") + +# 异步的上下文管理器，使用@asynccontextmanager装饰 +@asynccontextmanager +async def async_timer(label: str, user_count: int = 0): + """异步上下文管理器：用于测量包含 await 的异步代码块执行时间 + + Args: + label: 统计标签，用于标识被测量的代码块 + user_count: 用户数，可选参数，用于记录处理的用户数量 + + Usage: + async with async_timer("获取用户列表"): + users = await get_users() + + async with async_timer("批量处理", user_count=len(user_ids)): + await process_users(user_ids) + """ + start = time.perf_counter() + try: + yield + finally: + elapsed = (time.perf_counter() - start) * 1000 # 转换为毫秒 + extra_info = f", 用户数: {user_count}" if user_count > 0 else "" + api_logger.info(f"[性能统计] {label}: {elapsed:.2f}ms{extra_info}") diff --git a/api/app/utils/redis_lock.py b/api/app/utils/redis_lock.py index a86ba46e..b192c129 100644 --- a/api/app/utils/redis_lock.py +++ b/api/app/utils/redis_lock.py @@ -1,14 +1,21 @@ -import redis -import uuid -import time +import logging import threading +import time +import uuid + +import redis +from redis.exceptions import ( + ConnectionError, + TimeoutError, + RedisError, +) UNLOCK_SCRIPT = """ if redis.call("get", KEYS[1]) == ARGV[1] then return redis.call("del", KEYS[1]) -else - return 0 end + +return 0 """ RENEW_SCRIPT = """ @@ -19,38 +26,44 @@ else end """ -CLEANUP_DEAD_HEAD_SCRIPT = """ +ACQUIRE_SCRIPT = """ local queue_key = KEYS[1] local lock_key = KEYS[2] -local first = redis.call("lindex", queue_key, 0) -if not first then - return 0 +local client_id = ARGV[1] +local expire = tonumber(ARGV[2]) +local time_out = tonumber(ARGV[3]) + +local now = tonumber(redis.call("time")[1]) + +if redis.call("zscore", queue_key, client_id) == false then + redis.call("zadd", queue_key, now, client_id) end -if redis.call("exists", lock_key) == 1 then - return 0 +local expired = redis.call("zrangebyscore", queue_key, 0, now - time_out) + +for _, v in ipairs(expired) do + redis.call("zrem", queue_key, v) end -redis.call("lpop", queue_key) -return 1 -""" +local first = redis.call("zrange", queue_key, 0, 0)[1] +if first == client_id then -SAFE_RELEASE_QUEUE_SCRIPT = """ -local queue_key = KEYS[1] -local value = ARGV[1] + if redis.call("set", lock_key, client_id, "NX", "EX", expire) then + redis.call("zrem", queue_key, client_id) + return 1 + end -local first = redis.call("lindex", queue_key, 0) -if first == value then - redis.call("lpop", queue_key) - return 1 + if redis.call("get", lock_key) == client_id then + redis.call("expire", lock_key, expire) + return 1 + end end return 0 """ def _ensure_str(val): - """统一将 Redis 返回值转为 str，兼容 decode_responses=True/False""" if val is None: return None if isinstance(val, bytes): @@ -59,18 +72,27 @@ def _ensure_str(val): class RedisFairLock: + # ZOMBIE CLEAN BUFFER + CLEANUP_BUFFER = 30 + # Redis 操作失败时的最大重试次数 + MAX_RETRIES = 3 + # 重试间隔基数（秒），实际间隔 = base * 2^attempt（指数退避） + RETRY_BACKOFF_BASE = 0.1 + + _logger = logging.getLogger(__name__) + def __init__( self, key: str, redis_client: redis.StrictRedis, expire: int = 30, - retry_interval: float = 0.05, + retry_interval: float = 1, timeout: float = 600, auto_renewal: bool = True ): self.key = key - self.queue_key = f"{key}:queue" - self.value = str(uuid.uuid4()) + self.queue_key = f"{key}:zset" + self.value = f"{uuid.uuid4().hex}:{int(time.time())}" self.expire = expire self.retry_interval = retry_interval self.timeout = timeout @@ -80,28 +102,56 @@ class RedisFairLock: self._renew_thread = None self._stop_renew = threading.Event() + def _exec_with_retry(self, func, *args, raise_on_fail=True, **kwargs): + """ + 带指数退避重试的 Redis 操作执行器。 + + 对 ConnectionError / TimeoutError 自动重试，其他异常直接抛出。 + """ + last_err = None + for attempt in range(self.MAX_RETRIES): + try: + return func(*args, **kwargs) + except (ConnectionError, TimeoutError) as e: + last_err = e + wait = self.RETRY_BACKOFF_BASE * (2 ** attempt) + self._logger.warning( + f"[RedisFairLock] Redis error on attempt {attempt + 1}/{self.MAX_RETRIES} " + f"for key={self.key}: {e}, retrying in {wait:.2f}s" + ) + time.sleep(wait) + except RedisError: + raise + if raise_on_fail: + raise last_err + return None + def acquire(self): start = time.time() - self.redis.rpush(self.queue_key, self.value) - while True: - first = _ensure_str(self.redis.lindex(self.queue_key, 0)) + ok = self._exec_with_retry( + self.redis.eval, + ACQUIRE_SCRIPT, + 2, + self.queue_key, + self.key, + self.value, + str(self.expire), + str(self.timeout + self.CLEANUP_BUFFER), + ) - if first == self.value: - ok = self.redis.set(self.key, self.value, nx=True, ex=self.expire) - if ok: - self._locked = True - - if self.auto_renewal: - self._start_renewal() - return True - - if first: - self.redis.eval(CLEANUP_DEAD_HEAD_SCRIPT, 2, self.queue_key, self.key) + if ok == 1: + self._locked = True + if self.auto_renewal: + self._start_renewal() + return True if time.time() - start > self.timeout: - self.redis.lrem(self.queue_key, 0, self.value) + self._exec_with_retry( + self.redis.zrem, self.queue_key, self.value, + raise_on_fail=False, + ) return False time.sleep(self.retry_interval) @@ -112,13 +162,17 @@ class RedisFairLock: if self._stop_renew.is_set(): break - self.redis.eval( + success = self._exec_with_retry( + self.redis.eval, RENEW_SCRIPT, 1, self.key, self.value, - str(self.expire) + str(self.expire), + raise_on_fail=False, ) + if not success: + break def _start_renewal(self): self._stop_renew = threading.Event() @@ -137,9 +191,10 @@ class RedisFairLock: if self.auto_renewal: self._stop_renewal() - self.redis.eval(UNLOCK_SCRIPT, 1, self.key, self.value) - - self.redis.eval(SAFE_RELEASE_QUEUE_SCRIPT, 1, self.queue_key, self.value) + self._exec_with_retry( + self.redis.eval, UNLOCK_SCRIPT, 1, self.key, self.value, + raise_on_fail=False, + ) self._locked = False @@ -151,4 +206,3 @@ class RedisFairLock: def __exit__(self, exc_type, exc_val, exc_tb): self.release() - diff --git a/web/src/api/knowledgeBase.ts b/web/src/api/knowledgeBase.ts index 05200221..52384d06 100644 --- a/web/src/api/knowledgeBase.ts +++ b/web/src/api/knowledgeBase.ts @@ -68,7 +68,7 @@ export const getModelTypeList = async () => { return response as any[]; }; // 获取模型列表 -export const getModelList = async (types: string[], pageInfo: PageRequest) => { +export const getModelList = async (pageInfo: PageRequest, types?: string[]) => { const response = await request.get(`${apiPrefix}/models`, { ...pageInfo, type: types?.join(','), is_active: true }); return response as any; }; diff --git a/web/src/views/KnowledgeBase/components/CreateModal.tsx b/web/src/views/KnowledgeBase/components/CreateModal.tsx index 96cb8c92..97c88f0c 100644 --- a/web/src/views/KnowledgeBase/components/CreateModal.tsx +++ b/web/src/views/KnowledgeBase/components/CreateModal.tsx @@ -162,7 +162,7 @@ const CreateModal = forwardRef(({ // If model data hasn't been fetched yet, fetch it once if (!models) { try { - models = await getModelList({ page: 1, pagesize: 100 }); + models = await getModelList({ page: 1, pagesize: 100 }, ['llm', 'embedding', 'rerank', 'chat']); } catch (error) { console.error('Failed to fetch models:', error); models = { items: [] }; diff --git a/web/src/views/KnowledgeBase/index.tsx b/web/src/views/KnowledgeBase/index.tsx index b996bc16..9e1a63ce 100644 --- a/web/src/views/KnowledgeBase/index.tsx +++ b/web/src/views/KnowledgeBase/index.tsx @@ -207,7 +207,7 @@ const KnowledgeBaseManagement: FC = () => { }; const fetchModelList = async () => { try { - const response = await getModelList(['llm', 'embedding', 'rerank', 'chat'], { page: 1, pagesize: 100 }); + const response = await getModelList({ page: 1, pagesize: 100 }, ['llm', 'embedding', 'rerank', 'chat']); // 缓存模型列表，建立 id -> name 的映射 if (response?.items && Array.isArray(response.items)) { const cache: Record = {}; diff --git a/web/src/views/Workflow/components/Editor/Jinja2Editor.tsx b/web/src/views/Workflow/components/Editor/Jinja2Editor.tsx new file mode 100644 index 00000000..b6657450 --- /dev/null +++ b/web/src/views/Workflow/components/Editor/Jinja2Editor.tsx @@ -0,0 +1,186 @@ +/* + * @Author: ZhaoYing + * @Date: 2026-04-02 15:15:36 + * @Last Modified by: ZhaoYing + * @Last Modified time: 2026-04-02 15:15:36 + */ +import { type FC, useEffect, useMemo } from 'react'; +import { LexicalComposer } from '@lexical/react/LexicalComposer'; +import { RichTextPlugin } from '@lexical/react/LexicalRichTextPlugin'; +import { ContentEditable } from '@lexical/react/LexicalContentEditable'; +import { HistoryPlugin } from '@lexical/react/LexicalHistoryPlugin'; +import { LexicalErrorBoundary } from '@lexical/react/LexicalErrorBoundary'; + +import AutocompletePlugin, { type Suggestion } from './plugin/AutocompletePlugin'; +import CharacterCountPlugin from './plugin/CharacterCountPlugin'; +import InitialValuePlugin from './plugin/InitialValuePlugin'; +import CommandPlugin from './plugin/CommandPlugin'; +import Jinja2HighlightPlugin from './plugin/Jinja2HighlightPlugin'; +import LineNumberPlugin from './plugin/LineNumberPlugin'; +import BlurPlugin from './plugin/BlurPlugin'; + +const jinja2Theme = { + paragraph: 'editor-paragraph', + code: 'jinja2-expression', + text: { + bold: 'editor-text-bold', + italic: 'editor-text-italic', + code: 'jinja2-inline', + }, +}; + +const initialConfig = { + namespace: 'AutocompleteEditor', + theme: jinja2Theme, + nodes: [], + onError: (error: Error) => console.error(error), +}; + +const STYLE_ID = 'code-editor-styles'; +const JINJA2_STYLES = ` + .jinja2-expression { + background-color: #f6f8fa !important; + border: 1px solid #d1d9e0 !important; + border-radius: 3px !important; + padding: 2px 4px !important; + font-family: 'SFMono-Regular', Consolas, 'Liberation Mono', Menlo, monospace !important; + font-size: 13px !important; + color: #0969da !important; + } + .jinja2-inline { + background-color: #f6f8fa !important; + padding: 1px 3px !important; + border-radius: 2px !important; + font-family: 'SFMono-Regular', Consolas, 'Liberation Mono', Menlo, monospace !important; + font-size: 13px !important; + color: #0969da !important; + } + .editor-paragraph { margin: 0; } + .editor-with-line-numbers { display: flex; } + .line-numbers { + font-family: 'SFMono-Regular', Consolas, 'Liberation Mono', Menlo, monospace; + font-size: 12px; + line-height: 16px; + padding: 4px 8px; + text-align: right; + user-select: none; + display: flex; + flex-direction: column; + } + .line-numbers > div { min-height: 20px; display: flex; align-items: flex-start; } + .editor-content-wrapper { flex: 1; } + .editor-content-with-numbers { + white-space: pre-wrap; + font-family: 'SFMono-Regular', Consolas, 'Liberation Mono', Menlo, monospace; + } + .editor-content-with-numbers p { margin: 0; min-height: 20px; } +`; + +export interface Jinja2EditorProps { + placeholder?: string; + value?: string; + onChange?: (value: string) => void; + options?: Suggestion[]; + variant?: 'outlined' | 'borderless'; + height?: number; + size?: 'default' | 'small'; + className?: string; +} + +const Jinja2Editor: FC = ({ + placeholder = '请输入内容...', + value = '', + onChange, + options = [], + variant = 'borderless', + size = 'default', + height, + className, +}) => { + useEffect(() => { + if (!document.getElementById(STYLE_ID)) { + const style = document.createElement('style'); + style.id = STYLE_ID; + style.textContent = JINJA2_STYLES; + document.head.appendChild(style); + } + }, []); + + const minheight = useMemo( + () => `${height ?? (size === 'small' ? 60 : 120)}px`, + [height, size], + ); + + const fontSize = size === 'small' ? '12px' : '14px'; + + const lineHeight = useMemo( + () => `${height ? height - 10 : size === 'small' ? 16 : 20}px`, + [height, size], + ); + + const placeHolderMinheight = `${height ? 16 : size === 'small' ? 16 : 30}px`; + + return ( + +
+ +
+
1
+
+
+ +
+
+ } + placeholder={ +
+ {placeholder} +
+ } + ErrorBoundary={LexicalErrorBoundary} + /> + + + + + + {}} onChange={onChange} /> + + + + + ); +}; + +export default Jinja2Editor; diff --git a/web/src/views/Workflow/components/Editor/index.tsx b/web/src/views/Workflow/components/Editor/index.tsx index 60503065..8e72ca72 100644 --- a/web/src/views/Workflow/components/Editor/index.tsx +++ b/web/src/views/Workflow/components/Editor/index.tsx @@ -4,26 +4,20 @@ * @Last Modified by: ZhaoYing * @Last Modified time: 2026-03-25 10:58:47 */ -import { type FC, useState, useEffect, useMemo } from 'react'; +import { type FC, useState, useMemo } from 'react'; import { LexicalComposer } from '@lexical/react/LexicalComposer'; import { RichTextPlugin } from '@lexical/react/LexicalRichTextPlugin'; import { ContentEditable } from '@lexical/react/LexicalContentEditable'; import { HistoryPlugin } from '@lexical/react/LexicalHistoryPlugin'; -// import { AutoFocusPlugin } from '@lexical/react/LexicalAutoFocusPlugin'; import { LexicalErrorBoundary } from '@lexical/react/LexicalErrorBoundary'; -// import { HeadingNode, QuoteNode } from '@lexical/rich-text'; -// import { ListItemNode, ListNode } from '@lexical/list'; -// import { LinkNode } from '@lexical/link'; -// import { CodeNode } from '@lexical/code'; import AutocompletePlugin, { type Suggestion } from './plugin/AutocompletePlugin' import CharacterCountPlugin from './plugin/CharacterCountPlugin' import InitialValuePlugin from './plugin/InitialValuePlugin'; import CommandPlugin from './plugin/CommandPlugin'; -import Jinja2HighlightPlugin from './plugin/Jinja2HighlightPlugin'; -import LineNumberPlugin from './plugin/LineNumberPlugin'; import BlurPlugin from './plugin/BlurPlugin'; import { VariableNode } from './nodes/VariableNode' +import Jinja2Editor from './Jinja2Editor'; // Props interface for Lexical Editor component export interface LexicalEditorProps { @@ -50,16 +44,6 @@ const theme = { }, }; -// Theme with Jinja2 syntax highlighting -const jinja2Theme = { - ...theme, - code: 'jinja2-expression', - text: { - ...theme.text, - code: 'jinja2-inline', - }, -}; - // Main Lexical Editor component const Editor: FC =({ placeholder = "请输入内容...", @@ -74,97 +58,27 @@ const Editor: FC =({ className }) => { const [_count, setCount] = useState(0); - const [enableJinja2, setEnableJinja2] = useState(false) - const [enableLineNumbers, setEnableLineNumbers] = useState(false) - // Setup Jinja2 mode and inject styles when language changes - useEffect(() => { - const needsLineNumbers = language === 'jinja2'; - setEnableJinja2(language === 'jinja2'); - setEnableLineNumbers(needsLineNumbers); - - if (needsLineNumbers) { - const styleId = 'code-editor-styles'; - let existingStyle = document.getElementById(styleId); - - if (!existingStyle) { - const style = document.createElement('style'); - style.id = styleId; - style.textContent = ` - .jinja2-expression { - background-color: #f6f8fa !important; - border: 1px solid #d1d9e0 !important; - border-radius: 3px !important; - padding: 2px 4px !important; - font-family: 'SFMono-Regular', Consolas, 'Liberation Mono', Menlo, monospace !important; - font-size: 13px !important; - color: #0969da !important; - } - .jinja2-inline { - background-color: #f6f8fa !important; - padding: 1px 3px !important; - border-radius: 2px !important; - font-family: 'SFMono-Regular', Consolas, 'Liberation Mono', Menlo, monospace !important; - font-size: 13px !important; - color: #0969da !important; - } - .editor-paragraph { - margin: 0; - } - .editor-paragraph:has-text('{') .editor-text, - .editor-paragraph:has-text('[') .editor-text { - font-family: 'SFMono-Regular', Consolas, 'Liberation Mono', Menlo, monospace !important; - } - .editor-with-line-numbers { - display: flex; - } - .line-numbers { - font-family: 'SFMono-Regular', Consolas, 'Liberation Mono', Menlo, monospace; - font-size: 12px; - line-height: 16px; - padding: 4px 8px; - text-align: right; - user-select: none; - display: flex; - flex-direction: column; - } - .line-numbers > div { - min-height: 20px; - display: flex; - align-items: flex-start; - } - .editor-content-wrapper { - flex: 1; - } - .editor-content-with-numbers { - white-space: pre-wrap; - font-family: 'SFMono-Regular', Consolas, 'Liberation Mono', Menlo, monospace; - } - .editor-content-with-numbers p { - margin: 0; - min-height: 20px; - } - `; - document.head.appendChild(style); - } - } - }, [language]) + if (language === 'jinja2') { + return ( + + ); + } // Lexical editor configuration const initialConfig = { namespace: 'AutocompleteEditor', - theme: enableJinja2 ? jinja2Theme : theme, - nodes: enableJinja2 ? [ - // When Jinja2 is enabled, use plain text instead of VariableNode - ] : [ - // HeadingNode, - // QuoteNode, - // ListItemNode, - // ListNode, - // LinkNode, - // CodeNode, - VariableNode, - ], + theme, + nodes: [VariableNode], onError: (error: Error) => { console.error(error); }, @@ -198,54 +112,26 @@ const Editor: FC =({
-
-
1
-
-
- -
-
- ) : ( - // Standard editor without line numbers - - ) + padding: height ? '4px 6px' : variant === 'borderless' ? '0' : '6px 8px', + border: variant === 'borderless' ? 'none' : '1px solid #EBEBEB', + borderRadius: '8px', + outline: 'none', + resize: 'none', + fontSize: fontSize, + lineHeight: lineHeight, + }} + /> } placeholder={
=({ } ErrorBoundary={LexicalErrorBoundary} /> - {/* Editor plugins */} - {language === 'jinja2' && } - {enableLineNumbers && } - + { setCount(count) }} onChange={onChange} /> - - + +
); diff --git a/web/src/views/Workflow/components/Editor/plugin/CharacterCountPlugin.tsx b/web/src/views/Workflow/components/Editor/plugin/CharacterCountPlugin.tsx index ed07392d..4579375a 100644 --- a/web/src/views/Workflow/components/Editor/plugin/CharacterCountPlugin.tsx +++ b/web/src/views/Workflow/components/Editor/plugin/CharacterCountPlugin.tsx @@ -1,4 +1,4 @@ -import { useEffect } from 'react'; +import { useEffect, useRef } from 'react'; import { $getRoot, $isParagraphNode } from 'lexical'; import { useLexicalComposerContext } from '@lexical/react/LexicalComposerContext'; @@ -6,9 +6,15 @@ import { $isVariableNode } from '../nodes/VariableNode'; const CharacterCountPlugin = ({ setCount, onChange }: { setCount: (count: number) => void; onChange?: (value: string) => void }) => { const [editor] = useLexicalComposerContext(); + const isReadyRef = useRef(false); useEffect(() => { - return editor.registerUpdateListener(({ editorState }) => { + return editor.registerUpdateListener(({ editorState, tags }) => { + if (tags.has('programmatic')) { + isReadyRef.current = true; + return; + } + if (!isReadyRef.current) return; editorState.read(() => { const root = $getRoot(); let serializedContent = ''; diff --git a/web/src/views/Workflow/hooks/useWorkflowGraph.ts b/web/src/views/Workflow/hooks/useWorkflowGraph.ts index c427788b..fa6a2ec8 100644 --- a/web/src/views/Workflow/hooks/useWorkflowGraph.ts +++ b/web/src/views/Workflow/hooks/useWorkflowGraph.ts @@ -491,32 +491,36 @@ export const useWorkflowGraph = ({ * @param node - Clicked node */ const nodeClick = ({ node }: { node: Node }) => { - // Ignore add-node type node clicks - const nodeData = node.getData() - if (nodeData?.type === 'add-node' || nodeData.type === 'break' || nodeData.type === 'cycle-start') { - setSelectedNode(null) - return; - } + blankClick() - const nodes = graphRef.current?.getNodes(); - - nodes?.forEach(vo => { - const data = vo.getData(); - if (data.isSelected) { - vo.setData({ - ...data, - isSelected: false, - }); + setTimeout(() => { + // Ignore add-node type node clicks + const nodeData = node.getData() + if (nodeData?.type === 'add-node' || nodeData.type === 'break' || nodeData.type === 'cycle-start') { + setSelectedNode(null) + return; } - }); - node.setData({ - ...nodeData, - isSelected: true, - }); - clearEdgeSelect() - if (nodeData.type !== 'notes') { - setSelectedNode(node); - } + + const nodes = graphRef.current?.getNodes(); + + nodes?.forEach(vo => { + const data = vo.getData(); + if (data.isSelected) { + vo.setData({ + ...data, + isSelected: false, + }); + } + }); + node.setData({ + ...nodeData, + isSelected: true, + }); + clearEdgeSelect() + if (nodeData.type !== 'notes') { + setSelectedNode(node); + } + }, 0) }; /** * Handle edge click event