H0-14 P1 | RAG Chat SSE 流式输出 + DeepSeek V4 Pro 思考过程展示 #71

New Issue

wangdl · 2026-06-06T14:41:33+08:00

wangdl commented

2026-06-06 14:41:33 +08:00

背景

当前 AI 对话使用同步 fetch() 一次性取结果，response_format: json_object 模式下 DeepSeek V4 Pro 不输出 reasoning_content（思考过程）。iOS 端全链路阻塞等待，无流式效果。

需求

后端

DeepSeekProvider 新增 generateStream()：stream: true，去掉 response_format，ReadableStream 逐 chunk yield
AiProvider 接口新增 generateStream() 方法
AiGatewayService 新增 generateStream()：透传 provider stream，流结束后汇总 usage
RagChatService 新增 sendMessageStream()：调 gateway stream，保存最终 AI 回复到 DB
POST /api/rag-chat/sessions/:id/stream SSE endpoint

数据流

SSE chunk: { "type": "thinking", "content": "让我分析知识点..." }
SSE chunk: { "type": "thinking", "content": "根据《继承法》..." }
SSE chunk: { "type": "content", "content": "你好！" }
SSE chunk: { "type": "done", "usage": {...} }

关联

iOS: wangdl/ios-projects — SSE 流式接收 + 思考过程 UI
API: deepseek.com/v1/chat/completions (stream=true)

## 背景当前 AI 对话使用同步 `fetch()` 一次性取结果，`response_format: json_object` 模式下 DeepSeek V4 Pro 不输出 `reasoning_content`（思考过程）。iOS 端全链路阻塞等待，无流式效果。 ## 需求 ### 后端 1. `DeepSeekProvider` 新增 `generateStream()`：`stream: true`，去掉 `response_format`，ReadableStream 逐 chunk yield 2. `AiProvider` 接口新增 `generateStream()` 方法 3. `AiGatewayService` 新增 `generateStream()`：透传 provider stream，流结束后汇总 usage 4. `RagChatService` 新增 `sendMessageStream()`：调 gateway stream，保存最终 AI 回复到 DB 5. `POST /api/rag-chat/sessions/:id/stream` SSE endpoint ### 数据流 ``` SSE chunk: { "type": "thinking", "content": "让我分析知识点..." } SSE chunk: { "type": "thinking", "content": "根据《继承法》..." } SSE chunk: { "type": "content", "content": "你好！" } SSE chunk: { "type": "done", "usage": {...} } ``` ## 关联 - iOS: wangdl/ios-projects — SSE 流式接收 + 思考过程 UI - API: deepseek.com/v1/chat/completions (stream=true)

wangdl referenced this issue from a commit

2026-06-06 14:49:22 +08:00

feat: #71 RAG Chat SSE 流式输出 + DeepSeek V4 Pro 思考过程

wangdl commented

2026-06-06 14:49:30 +08:00

后端实现完成 (2026-06-06)

改动文件

文件	变更
`ai-provider.interface.ts`	新增 `StreamChunk` 类型 + `generateStream()` 方法
`deepseek.provider.ts`	实现 `generateStream()`：`stream: true`，ReadableStream 逐 chunk yield，reasoning_content → thinking
`ai-gateway.service.ts`	新增 `generateStream()`：透传 provider stream + 记录用量
`rag-chat.service.ts`	新增 `sendMessageStream()`：流式调用 + 保存最终回复到 DB
`rag-chat.controller.ts`	新增 `POST /api/rag-chat/sessions/:id/stream` SSE endpoint

SSE 数据格式

data: {"type":"thinking","content":"让我分析..."}
data: {"type":"content","content":"你好！"}
data: {"type":"done","usage":{"inputTokens":100,"outputTokens":50}}

待完成

iOS 端 SSE 接收 + 思考过程 UI (ios-projects #38)

## 后端实现完成 (2026-06-06) ### 改动文件 | 文件 | 变更 | |------|------| | `ai-provider.interface.ts` | 新增 `StreamChunk` 类型 + `generateStream()` 方法 | | `deepseek.provider.ts` | 实现 `generateStream()`：`stream: true`，ReadableStream 逐 chunk yield，reasoning_content → thinking | | `ai-gateway.service.ts` | 新增 `generateStream()`：透传 provider stream + 记录用量 | | `rag-chat.service.ts` | 新增 `sendMessageStream()`：流式调用 + 保存最终回复到 DB | | `rag-chat.controller.ts` | 新增 `POST /api/rag-chat/sessions/:id/stream` SSE endpoint | ### SSE 数据格式 ``` data: {"type":"thinking","content":"让我分析..."} data: {"type":"content","content":"你好！"} data: {"type":"done","usage":{"inputTokens":100,"outputTokens":50}} ``` ### 待完成 iOS 端 SSE 接收 + 思考过程 UI (ios-projects #38)

wangdl commented

2026-06-06 14:54:06 +08:00

后端完成 (2026-06-06)

新增端点

POST /api/rag-chat/sessions/:id/stream — SSE 流式输出

改动文件

ai-provider.interface.ts — StreamChunk 类型 + generateStream()
deepseek.provider.ts — stream:true，ReadableStream yield thinking/content
ai-gateway.service.ts — generateStream() 透传 + 记录用量
rag-chat.service.ts — sendMessageStream() 流式调用
rag-chat.controller.ts — SSE endpoint

状态

✅ 完成。

## 后端完成 (2026-06-06) ### 新增端点 `POST /api/rag-chat/sessions/:id/stream` — SSE 流式输出 ### 改动文件 - `ai-provider.interface.ts` — StreamChunk 类型 + generateStream() - `deepseek.provider.ts` — stream:true，ReadableStream yield thinking/content - `ai-gateway.service.ts` — generateStream() 透传 + 记录用量 - `rag-chat.service.ts` — sendMessageStream() 流式调用 - `rag-chat.controller.ts` — SSE endpoint ### 状态 ✅ 完成。

wangdl closed this issue

2026-06-06 14:54:06 +08:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: wangdl/api-server#71