group_xinghuo_jinrong

207 Commits 9 Branches 1 Tag

3 Commits

Include renames

Author	SHA1	Message	Date
zhanghongyu_0626	3ad244add0	feat(analyst): Enhance dashboard metrics and customer service interactions - Updated the `dashboard` function in `analyst.py` to include additional metrics for different user roles, improving data visibility for analysts, customers, advisors, and risk officers. - Introduced a new `prepare_customer_stream` function in `customer_service.py` to facilitate streaming responses for customer interactions, enhancing the chat experience. - Added new API endpoints in `analyst.ts` for fetching dashboard metrics and managing analyst assets, streamlining data handling and user interactions. - Updated frontend components to support new dashboard features and asset management, ensuring a cohesive user experience across the application. This update significantly improves the functionality and usability of the analyst and customer service features, providing users with enhanced tools for data analysis and interaction.	2026-09-09 21:32:59 +08:00
zhanghongyu_0626	d0da627a5c	feat(chat): Update trace ID handling in chat stream API - Modified `chat_stream_api` to use `new_trace()` instead of an empty string for `trace_id`, enhancing traceability. - Added tests to verify the correct generation and propagation of `trace_id` in responses, ensuring no empty trace IDs are sent to clients. This update improves the tracking of chat sessions and ensures compliance with traceability standards.	2026-09-08 21:46:08 +08:00
GaoYiYuan_0626	01ec5fce32	feat(chat): 新增 SSE 流式对话端点 POST /api/chat/stream（方案 C）契约（OpenAI 兼容 chunk，AI SDK / fetch-event-source 可直接接）：首帧 meta（session_id/trace_id/disclaimer）→ delta → finish_reason=stop → [DONE]。前端侧：免责声明由首帧下发、前端常驻渲染；落库文本仍按原口径拼尾部。设计拍板： 1. 新增独立端点，原 POST /api/chat 契约与既有测试零影响； 2. 鉴权/限流/输入防护全部在返回 StreamingResponse 之前完成（SSE 一开就改不了状态码），401/403/404/409/429/400 仍是普通 JSON； 3. 整轮一次性落库：中途异常/断连不落消息（Tool 留痕已落可审计），不产生半截内容污染历史窗口。实现： - agent_service：抽 needs_disclaimer/_degraded_reply/_base_state；新增 stream_chat 生成器（Tool 节点同步跑完再推 LLM 文本，无 key 走降级整块）； - api/chat：抽 _guard_request（准入→空白→限流→注入拦截）与 _prepare_turn （归属+会话解析/创建），同步与流式共用，守卫零偏差； - session_repository：新增 insert_turn——user+assistant 同事务落库 + 事务内取 seq，修掉评审 P0（落库失败会半截落且前端收不到 [DONE] 挂起）与 P1 （两次写非原子）；同步端点一并改用。测试：新增 9 例（契约/免责/降级/异常不落库/落库失败/鉴权边界/超长/closed/Tool 留痕），全量 pytest 494→503 绿。遗留：无心跳帧（长生成空隙靠反代 timeout 配置），断连留空会话待清理策略。	2026-09-08 14:23:33 +08:00

Author

SHA1

Message

Date

zhanghongyu_0626

3ad244add0

feat(analyst): Enhance dashboard metrics and customer service interactions

- Updated the `dashboard` function in `analyst.py` to include additional metrics for different user roles, improving data visibility for analysts, customers, advisors, and risk officers.
- Introduced a new `prepare_customer_stream` function in `customer_service.py` to facilitate streaming responses for customer interactions, enhancing the chat experience.
- Added new API endpoints in `analyst.ts` for fetching dashboard metrics and managing analyst assets, streamlining data handling and user interactions.
- Updated frontend components to support new dashboard features and asset management, ensuring a cohesive user experience across the application.

This update significantly improves the functionality and usability of the analyst and customer service features, providing users with enhanced tools for data analysis and interaction.

2026-09-09 21:32:59 +08:00

zhanghongyu_0626

d0da627a5c

feat(chat): Update trace ID handling in chat stream API

- Modified `chat_stream_api` to use `new_trace()` instead of an empty string for `trace_id`, enhancing traceability.
- Added tests to verify the correct generation and propagation of `trace_id` in responses, ensuring no empty trace IDs are sent to clients.

This update improves the tracking of chat sessions and ensures compliance with traceability standards.

2026-09-08 21:46:08 +08:00

GaoYiYuan_0626

01ec5fce32

feat(chat): 新增 SSE 流式对话端点 POST /api/chat/stream（方案 C）

契约（OpenAI 兼容 chunk，AI SDK / fetch-event-source 可直接接）：
首帧 meta（session_id/trace_id/disclaimer）→ delta → finish_reason=stop → [DONE]。
前端侧：免责声明由首帧下发、前端常驻渲染；落库文本仍按原口径拼尾部。

设计拍板：
1. 新增独立端点，原 POST /api/chat 契约与既有测试零影响；
2. 鉴权/限流/输入防护全部在返回 StreamingResponse 之前完成（SSE 一开就改不了
   状态码），401/403/404/409/429/400 仍是普通 JSON；
3. 整轮一次性落库：中途异常/断连不落消息（Tool 留痕已落可审计），不产生
   半截内容污染历史窗口。

实现：
- agent_service：抽 needs_disclaimer/_degraded_reply/_base_state；新增 stream_chat
  生成器（Tool 节点同步跑完再推 LLM 文本，无 key 走降级整块）；
- api/chat：抽 _guard_request（准入→空白→限流→注入拦截）与 _prepare_turn
  （归属+会话解析/创建），同步与流式共用，守卫零偏差；
- session_repository：新增 insert_turn——user+assistant 同事务落库 + 事务内
  取 seq，修掉评审 P0（落库失败会半截落且前端收不到 [DONE] 挂起）与 P1
  （两次写非原子）；同步端点一并改用。

测试：新增 9 例（契约/免责/降级/异常不落库/落库失败/鉴权边界/超长/closed/Tool
留痕），全量 pytest 494→503 绿。遗留：无心跳帧（长生成空隙靠反代 timeout 配置），
断连留空会话待清理策略。

2026-09-08 14:23:33 +08:00