fix(sse): correct UTF-8 handling and chunking in SSE streaming\n\n- Updated splitContentIntoSegments to handle runes instead of bytes\n- Fixed buildWebUIStreamForwarder to send delta chunks instead of cumulative content\n- Ensures proper handling of multi-byte characters in SSE streams

README.md

@@ -27,6 +27,7 @@ Now supports mutually exclusive message channels:
 	- If `telegram`: set `TELEGRAM_BOT_TOKEN`, keep `FEISHU_*` empty.
 	- If `feishu`: set `FEISHU_APP_ID` and `FEISHU_APP_SECRET`, keep `TELEGRAM_BOT_TOKEN` empty.
 	- If `webui`: set `WEBUI_LISTEN_ADDR` (default `:8090`).
+	- Optional for `webui`: set `WEBUI_EXPOSE_REASONING=true` to expose `thought/tool_call/tool_result` events to frontend.
 3. Set log level with `LOG_LEVEL=debug|info|warn|error`.
 	- To inspect full skill/tool execution content and detailed ReAct step traces, use `LOG_LEVEL=debug`.
 4. Configure knowledge and reasoning:
@@ -95,6 +96,7 @@ Minimum env config in `workspace/agent_runtime/configs/env`:
 ```env
 MESSAGE_CHANNEL=webui
 WEBUI_LISTEN_ADDR=:8090
+WEBUI_EXPOSE_REASONING=false
 LLM_API_KEY=your_api_key
 ```
 
@@ -115,6 +117,7 @@ docker run --rm -d \
 	--env-file ./workspace/agent_runtime/configs/env \
 	-e MESSAGE_CHANNEL=webui \
 	-e WEBUI_LISTEN_ADDR=:8090 \
+	-e WEBUI_EXPOSE_REASONING=false \
 	-e AGENT_WORKSPACE_DIR=/app/workspace/agent_runtime \
 	-v "$(pwd)/workspace/agent_runtime:/app/workspace/agent_runtime" \
 	laodingbot:latest

cmd/bot/main.go

@@ -220,7 +220,7 @@ func runMessageChannel(ctx context.Context, cfg config.Config, engine *agent.Orc
 				return engine.HandleMessage(ctx, msg.ChatID, msg.UserID, msg.Text)
 			},
 			func(ctx context.Context, msg webui.IncomingMessage, callback webui.StreamEventCallback) (string, error) {
-				return engine.HandleMessageStream(ctx, msg.ChatID, msg.UserID, msg.Text, buildWebUIStreamForwarder(callback))
+				return engine.HandleMessageStream(ctx, msg.ChatID, msg.UserID, msg.Text, buildWebUIStreamForwarder(callback, cfg.WebUI.ExposeReasoning))
 			},
 			func(ctx context.Context, chatID, userID string, files []llm.InputFile) ([]string, error) {
 				return engine.UploadAndCacheFiles(ctx, chatID, userID, files)
@@ -231,16 +231,54 @@ func runMessageChannel(ctx context.Context, cfg config.Config, engine *agent.Orc
 	}
 }
 
-func buildWebUIStreamForwarder(callback webui.StreamEventCallback) agent.StreamEventCallback {
+func buildWebUIStreamForwarder(callback webui.StreamEventCallback, exposeReasoning bool) agent.StreamEventCallback {
+	const finalChunkRunes = 12
+	const finalChunkInterval = 25 * time.Millisecond
+
 	return func(event agent.StreamEvent) error {
 		if callback == nil {
 			return nil
 		}
 		switch event.Type {
-		case agent.StreamEventTypeFinal, agent.StreamEventTypeError:
+		case agent.StreamEventTypeThought, agent.StreamEventTypeToolCall, agent.StreamEventTypeToolResult:
+			if !exposeReasoning {
+				return nil
+			}
 			return callback(webui.StreamEvent{
 				Type:     webui.StreamEventType(event.Type),
 				Content:  event.Content,
+				Step:     event.Step,
+				ToolName: event.ToolName,
+			})
+		case agent.StreamEventTypeFinal:
+			runes := []rune(event.Content)
+			if len(runes) == 0 {
+				return callback(webui.StreamEvent{Type: webui.StreamEventTypeFinal, Content: "", Step: event.Step})
+			}
+			start := 0
+			for start < len(runes) {
+				end := start + finalChunkRunes
+				if end > len(runes) {
+					end = len(runes)
+				}
+				if err := callback(webui.StreamEvent{
+					Type:    webui.StreamEventTypeFinal,
+					Content: string(runes[start:end]),
+					Step:    event.Step,
+				}); err != nil {
+					return err
+				}
+				start = end
+				if start < len(runes) {
+					time.Sleep(finalChunkInterval)
+				}
+			}
+			return nil
+		case agent.StreamEventTypeError:
+			return callback(webui.StreamEvent{
+				Type:    webui.StreamEventTypeError,
+				Content: event.Content,
+				Step:    event.Step,
 			})
 		default:
 			return nil

configs/env.sample

@@ -21,6 +21,7 @@ FEISHU_LISTEN_ADDR=:8080
 FEISHU_EVENT_PATH=/feishu/events
 WEBUI_LISTEN_ADDR=:8090
 WEBUI_MAX_UPLOAD_MB=20
+WEBUI_EXPOSE_REASONING=false
 
 LLM_BASE_URL=https://api.openai.com/v1
 LLM_API_KEY=

doc/WebUI_Stream_API_前端对接说明.md

@@ -57,13 +57,17 @@ data: {"type":"final","content":"当前目录是 C:/Project/MyProject","step":2}
 
 事件类型:
 
+- `thought`: LLM 思考片段（可选透传）
+- `tool_call`: 工具调用请求（可选透传）
+- `tool_result`: 工具执行结果（可选透传）
 - `final`: 最终回答
 - `error`: 错误信息
 
 说明:
 
-- 当前 WebUI 默认只向前端返回 `final` 和 `error`。
+- 默认情况下（`WEBUI_EXPOSE_REASONING=false`），WebUI 只向前端返回 `final` 和 `error`。
-- 中间推理轨迹（如 `thought`、`tool_call`、`tool_result` 以及对应的 `step`）会写入服务端 `debug` 日志，不再直接返回给用户界面。
+- 当设置 `WEBUI_EXPOSE_REASONING=true` 时，WebUI 会额外透传 `thought`、`tool_call`、`tool_result` 事件。
+- 无论是否透传推理事件，`final` 都会分段累计推送，以便前端实现打字机效果。
 
 ## 4. 连接生命周期
 
@@ -78,6 +82,11 @@ data: {"type":"final","content":"当前目录是 C:/Project/MyProject","step":2}
 - 答案区: 显示最后一个 `final`
 - 错误区: 显示 `error`
 
+如果开启了 `WEBUI_EXPOSE_REASONING=true`，建议额外提供“思考面板”：
+
+- 思考区: 渲染 `thought`
+- 工具区: 渲染 `tool_call` / `tool_result`
+
 建议状态机:
 
 - `idle`: 初始状态
@@ -88,7 +97,7 @@ data: {"type":"final","content":"当前目录是 C:/Project/MyProject","step":2}
 ## 6. TypeScript 对接示例 (fetch + ReadableStream)
 
 ```ts
-type StreamEventType = 'final' | 'error';
+type StreamEventType = 'thought' | 'tool_call' | 'tool_result' | 'final' | 'error';
 
 interface StreamEvent {
   type: StreamEventType;
@@ -165,7 +174,7 @@ export async function streamChat(
    - step?: number
    - tool_name?: string
 5) 收到 final 视为本轮完成；收到 error 视为失败
-6) 不要再假设前端会收到 thought/tool_call/tool_result；这些内部轨迹已改为服务端 debug 日志
+6) 默认不要假设前端一定会收到 thought/tool_call/tool_result；仅当 `WEBUI_EXPOSE_REASONING=true` 才会透传
 
 你的改造要求:
 1) 保留现有 UI 风格和组件结构，不做无关重构

internal/agent/orchestrator.go

@@ -731,14 +731,18 @@ func (o *Orchestrator) runUnifiedReActStream(ctx context.Context, chatID, userID
 
 		// 推送思考过程事件
 		if completion.Content != "" {
+			// 分割内容为逐步推送的片段
+			segments := splitContentIntoSegments(completion.Content, 50) // 每段50字符
+			for _, segment := range segments {
 				if err := callback(StreamEvent{
 					Type:    StreamEventTypeThought,
-				Content: completion.Content,
+					Content: segment,
 					Step:    step,
 				}); err != nil {
 					return "", fmt.Errorf("callback error: %w", err)
 				}
 			}
+		}
 
 		// ========== 无 tool_calls → 最终回答 ==========
 		if len(completion.ToolCalls) == 0 {
@@ -1568,3 +1572,17 @@ func sanitizeUserFacingAnswer(raw string) string {
 	}
 	return strings.TrimSpace(strings.Join(cleaned, "\n"))
 }
+
+// splitContentIntoSegments splits a string into smaller segments of the specified size (by rune count).
+func splitContentIntoSegments(content string, segmentSize int) []string {
+	runes := []rune(content)
+	var segments []string
+	for start := 0; start < len(runes); start += segmentSize {
+		end := start + segmentSize
+		if end > len(runes) {
+			end = len(runes)
+		}
+		segments = append(segments, string(runes[start:end]))
+	}
+	return segments
+}

internal/config/config.go

@@ -50,6 +50,7 @@ type FeishuConfig struct {
 type WebUIConfig struct {
 	ListenAddr      string
 	MaxUploadBytes  int64
+	ExposeReasoning bool
 }
 
 type LLMConfig struct {
@@ -112,6 +113,7 @@ func Load() (Config, error) {
 		WebUI: WebUIConfig{
 			ListenAddr:      defaultIfEmpty(os.Getenv("WEBUI_LISTEN_ADDR"), ":8090"),
 			MaxUploadBytes:  int64(intFromEnv("WEBUI_MAX_UPLOAD_MB", 20)) * 1024 * 1024,
+			ExposeReasoning: boolFromEnv("WEBUI_EXPOSE_REASONING", false),
 		},
 		LLM: LLMConfig{
 			BaseURL:     strings.TrimRight(defaultIfEmpty(os.Getenv("LLM_BASE_URL"), "https://api.openai.com/v1"), "/"),