fix: 模型思考模式配置 + Unicode 截断 + 审计修复 (13 files)

模型模式:
- 新增 LLMConfig/Source.ThinkingEnabled 配置，通过 ExtraBody
  控制 DeepSeek thinking mode，默认关闭
- SeedDefaults/ToConfig 读写 core.llm.thinking_enabled
- deepseek.lua 移除硬编码 temperature=0

Unicode 截断:
- truncateStr 改按 rune 计数，修复中文截断乱码

审计修复 (Critical):
- graph.go: defer rows.Close 在 for 循环 → 显式 Close (连接池泄漏)
- cli/openclaw/plugin.go: bare type assertion → comma-ok (panic)
- channel.go: payload["type"].(string) → comma-ok (panic)
- webui/handler.go: .(string) → fmt.Sprint (panic)
- agent.go: 添加 nil provider 错误返回

审计修复 (High):
- events/bus.go: copy handler slice under RLock (data race)
- webui/handler.go: SSE 通过 channel 串行化写入 (data race)
- timer/plugin.go: time.Sleep → select with stopCh (Stop 阻塞)
- provider.go: stream ch <- 添加 select ctx.Done (goroutine 泄漏)
- main.go: outputCh goroutine 添加 ctx.Done 退出路径

cmd/homed/main.go

@@ -1,6 +1,7 @@
 package main
 
 import (
+	"context"
 	"flag"
 	"log"
 	"os"
@@ -162,34 +163,45 @@ func main() {
 		log.Printf("[homed] text memory active at %s", filepath.Join(cfg.Daemon.DataDir, "memory", "text"))
 	}
 
+	ctx, stop := context.WithCancel(context.Background())
+	defer stop()
+
 	go func() {
-		for evt := range iom.OutputChan() {
+		for {
-			if evt.Target == "memory" && evt.Type == "memory_candidate" {
+			select {
-				source, _ := evt.Payload["source"].(string)
+			case <-ctx.Done():
-				input, _ := evt.Payload["input"].(string)
+				return
-				response, _ := evt.Payload["response"].(string)
+			case evt, ok := <-iom.OutputChan():
-				toolsUsed, _ := evt.Payload["tools_used"].([]string)
+				if !ok {
-				agentID, _ := evt.Payload["agent_id"].(string)
+					return
-
-				if input != "" && textMem != nil {
-					te := text.Event{
-						Timestamp: time.Now().Unix(),
-						Source:    source,
-						Input:     input,
-						Response:  response,
-						ToolsUsed: toolsUsed,
-						AgentID:   agentID,
-					}
-					if err := textMem.Append(te); err != nil {
-						log.Printf("[homed] text memory append: %v", err)
-					}
 				}
+				if evt.Target == "memory" && evt.Type == "memory_candidate" {
+					source, _ := evt.Payload["source"].(string)
+					input, _ := evt.Payload["input"].(string)
+					response, _ := evt.Payload["response"].(string)
+					toolsUsed, _ := evt.Payload["tools_used"].([]string)
+					agentID, _ := evt.Payload["agent_id"].(string)
 
-				if input != "" {
+					if input != "" && textMem != nil {
-					distiller.Append("agent", "user", input)
+						te := text.Event{
-				}
+							Timestamp: time.Now().Unix(),
-				if response != "" {
+							Source:    source,
-					distiller.Append("agent", "assistant", response)
+							Input:     input,
+							Response:  response,
+							ToolsUsed: toolsUsed,
+							AgentID:   agentID,
+						}
+						if err := textMem.Append(te); err != nil {
+							log.Printf("[homed] text memory append: %v", err)
+						}
+					}
+
+					if input != "" {
+						distiller.Append("agent", "user", input)
+					}
+					if response != "" {
+						distiller.Append("agent", "assistant", response)
+					}
 				}
 			}
 		}
@@ -345,6 +357,7 @@ func main() {
 		ContextSavePath: filepath.Join(cfg.Daemon.DataDir, "memory", "context.json"),
 		StageHost:       stageHost,
 		EventBus:        evBus,
+		ThinkingEnabled: cfg.LLM.ThinkingEnabled,
 	})
 	agent.Start()
 	defer agent.Stop()

internal/agent/api/provider.go

@@ -265,13 +265,17 @@ func (p *OpenAIProvider) ChatStream(ctx context.Context, req *CompletionRequest)
 			}
 
 			if err := decoder.Decode(&line); err != nil {
-				break
+				return
 			}
 
 			if len(line.Choices) > 0 {
-				ch <- StreamChunk{
+				select {
+				case ch <- StreamChunk{
 					Content: line.Choices[0].Delta.Content,
 					Done:    line.Choices[0].FinishReason != nil,
+				}:
+				case <-ctx.Done():
+					return
 				}
 			}
 		}
@@ -498,9 +502,13 @@ func (p *LuaAdaptedProvider) ChatStream(ctx context.Context, req *CompletionRequ
 					continue
 				}
 				if len(raw.Choices) > 0 {
-					ch <- StreamChunk{
+					select {
+					case ch <- StreamChunk{
 						Content: raw.Choices[0].Delta.Content,
 						Done:    raw.Choices[0].FinishReason != nil,
+					}:
+					case <-ctx.Done():
+						return
 					}
 				}
 				continue
@@ -509,7 +517,11 @@ func (p *LuaAdaptedProvider) ChatStream(ctx context.Context, req *CompletionRequ
 			// Lua 返回了变换后的统一格式
 			var chunk StreamChunk
 			if err := json.Unmarshal([]byte(unified), &chunk); err == nil {
-				ch <- chunk
+				select {
+				case ch <- chunk:
+				case <-ctx.Done():
+					return
+				}
 			}
 		}
 	}()

internal/agent/core/agent.go

@@ -7,6 +7,7 @@ import (
 	"strings"
 	"sync"
 	"time"
+	"unicode/utf8"
 
 	agentAPI "gitcode.com/JianFeeeee/HomeAgent/internal/agent/api"
 	agentPkg "gitcode.com/JianFeeeee/HomeAgent/internal/agent"
@@ -89,6 +90,9 @@ type Agent struct {
 	// 进行中的 LLM 请求取消函数，interceptLoop 可调用以在请求中打断
 	cancelLLM context.CancelFunc
 	llmMu     sync.Mutex
+
+	// 模型思考模式（thinking/reasoning）
+	thinkingEnabled bool
 }
 
 type AgentConfig struct {
@@ -115,6 +119,7 @@ type AgentConfig struct {
 	ContextSavePath string           // 上下文持久化路径，空则不持久化
 	StageHost     *StageHost
 	EventBus      *events.Bus
+	ThinkingEnabled bool
 }
 
 func New(cfg AgentConfig) *Agent {
@@ -156,6 +161,7 @@ func New(cfg AgentConfig) *Agent {
 		selfInputCh:      make(chan string, 64),
 		childResults:     make(map[string]string),
 		interceptCh:      make(chan string, 64),
+		thinkingEnabled:  cfg.ThinkingEnabled,
 	}
 }
 
@@ -383,6 +389,10 @@ func (a *Agent) process(input string, stageCtx *sdk.StageContext) (response stri
 	a.mu.Lock()
 	defer a.mu.Unlock()
 
+	if a.provider == nil {
+		return "", nil, fmt.Errorf("agent: no LLM provider configured")
+	}
+
 	memContext := a.buildMemoryContext(input)
 	sysPrompt := a.buildSystemPrompt(memContext, input)
 	tools := a.buildToolDefs()
@@ -418,14 +428,16 @@ func (a *Agent) process(input string, stageCtx *sdk.StageContext) (response stri
 			log.Printf("[agent] interrupt injected before LLM call (turn %d)", turn)
 		}
 
+		eb := map[string]interface{}{}
+		if !a.thinkingEnabled {
+			eb["thinking"] = map[string]interface{}{"type": "disabled"}
+		}
 		req := &agentAPI.CompletionRequest{
 			Messages:   msgs,
 			MaxTokens:  4096,
 			Tools:      tools,
 			ToolChoice: "auto",
-			ExtraBody: map[string]interface{}{
+			ExtraBody:  eb,
-				"thinking": map[string]interface{}{"type": "disabled"},
-			},
 		}
 
 		// 可取消的 LLM 调用：interceptLoop 通过 cancelLLM 打断进行中的请求
@@ -1839,6 +1851,10 @@ func (a *Agent) executeSpawnChild(tc agentAPI.ToolCall) string {
 
 // runChildTask 后台运行子 Agent 任务，完成后将结果存储并通过 selfInputCh 通知主 Agent
 func (a *Agent) runChildTask(taskID, task string) {
+	if a.provider == nil {
+		log.Printf("[child] %s failed: no LLM provider configured", taskID)
+		return
+	}
 	log.Printf("[child] %s started: %s", taskID, truncateStr(task, 80))
 
 	sysPrompt := fmt.Sprintf(`你是 HomeAgent 的子任务助手。
@@ -1871,14 +1887,16 @@ func (a *Agent) runChildTask(taskID, task string) {
 
 	var finalResult string
 	for turn := 0; turn < 5; turn++ {
+		eb := map[string]interface{}{}
+		if !a.thinkingEnabled {
+			eb["thinking"] = map[string]interface{}{"type": "disabled"}
+		}
 		req := &agentAPI.CompletionRequest{
 			Messages:   msgs,
 			MaxTokens:  4096,
 			Tools:      childTools,
 			ToolChoice: "auto",
-			ExtraBody: map[string]interface{}{
+			ExtraBody:  eb,
-				"thinking": map[string]interface{}{"type": "disabled"},
-			},
 		}
 
 		resp, err := a.provider.Chat(a.ctx, req)
@@ -2038,10 +2056,17 @@ func getFloat(m map[string]interface{}, key string) float64 {
 }
 
 func truncateStr(s string, max int) string {
-	if len(s) <= max {
+	if utf8.RuneCountInString(s) <= max {
 		return s
 	}
-	return s[:max] + "..."
+	var truncated int
+	for i := range s {
+		if truncated >= max {
+			return s[:i] + "..."
+		}
+		truncated++
+	}
+	return s
 }
 
 // drainInterrupt 非阻塞读取 interceptCh 中的一条打断消息。

internal/agent/io/channel.go

@@ -246,10 +246,11 @@ func (m *IOManager) InjectInterrupt(source, channel string, payload map[string]i
 	if payload == nil {
 		payload = map[string]interface{}{}
 	}
+	evtType, _ := payload["type"].(string)
 	m.interruptCh <- &InputEvent{
 		RequestID:     m.nextRequestID(),
 		Source:        source,
-		Type:          payload["type"].(string),
+		Type:          evtType,
 		Payload:       payload,
 		OutputChannel: channel,
 	}

internal/config/registry.go

@@ -186,25 +186,29 @@ func (r *ConfigRegistry) SeedDefaults(dataDir string) {
 	set("core.llm.provider", "deepseek")
 	set("core.llm.model", "deepseek-v4-flash")
 	set("core.llm.base_url", "https://api.deepseek.com")
+	set("core.llm.api_key", "")
 	set("core.llm.adapter", "deepseek")
 	set("core.llm.temperature", "0.7")
 	set("core.llm.max_tokens", "4096")
+	set("core.llm.thinking_enabled", "false")
 
 	// llm sources
 	sources := map[string]map[string]string{
-		"deepseek": {"base_url": "https://api.deepseek.com", "model": "deepseek-v4-flash", "adapter": "deepseek", "adapter_path": "adapters/deepseek.lua"},
+		"deepseek": {"base_url": "https://api.deepseek.com", "model": "deepseek-v4-flash", "api_key": "", "thinking_enabled": "false", "adapter": "deepseek", "adapter_path": "adapters/deepseek.lua"},
-		"openai":   {"base_url": "https://api.openai.com/v1", "model": "gpt-4o", "adapter": "openai", "adapter_path": "adapters/openai.lua"},
+		"openai":   {"base_url": "https://api.openai.com/v1", "model": "gpt-4o", "api_key": "", "thinking_enabled": "false", "adapter": "openai", "adapter_path": "adapters/openai.lua"},
-		"anthropic": {"base_url": "https://api.anthropic.com", "model": "claude-sonnet-4-20250514", "adapter": "anthropic", "adapter_path": "adapters/anthropic.lua"},
+		"anthropic": {"base_url": "https://api.anthropic.com", "model": "claude-sonnet-4-20250514", "api_key": "", "thinking_enabled": "false", "adapter": "anthropic", "adapter_path": "adapters/anthropic.lua"},
-		"gemini":   {"base_url": "https://generativelanguage.googleapis.com", "model": "gemini-2.0-flash", "adapter": "gemini", "adapter_path": "adapters/gemini.lua"},
+		"gemini":   {"base_url": "https://generativelanguage.googleapis.com", "model": "gemini-2.0-flash", "api_key": "", "thinking_enabled": "false", "adapter": "gemini", "adapter_path": "adapters/gemini.lua"},
-		"mistral":  {"base_url": "https://api.mistral.ai", "model": "mistral-large-latest", "adapter": "mistral", "adapter_path": "adapters/mistral.lua"},
+		"mistral":  {"base_url": "https://api.mistral.ai", "model": "mistral-large-latest", "api_key": "", "thinking_enabled": "false", "adapter": "mistral", "adapter_path": "adapters/mistral.lua"},
-		"groq":     {"base_url": "https://api.groq.com", "model": "llama3-70b-8192", "adapter": "groq", "adapter_path": "adapters/groq.lua"},
+		"groq":     {"base_url": "https://api.groq.com", "model": "llama3-70b-8192", "api_key": "", "thinking_enabled": "false", "adapter": "groq", "adapter_path": "adapters/groq.lua"},
-		"github":   {"base_url": "https://models.inference.ai.azure.com", "model": "gpt-4o", "adapter": "github", "adapter_path": "adapters/github.lua"},
+		"github":   {"base_url": "https://models.inference.ai.azure.com", "model": "gpt-4o", "api_key": "", "thinking_enabled": "false", "adapter": "github", "adapter_path": "adapters/github.lua"},
-		"ollama":   {"base_url": "http://localhost:11434", "model": "llama3", "adapter": "ollama", "adapter_path": "adapters/ollama.lua"},
+		"ollama":   {"base_url": "http://localhost:11434", "model": "llama3", "api_key": "", "thinking_enabled": "false", "adapter": "ollama", "adapter_path": "adapters/ollama.lua"},
 	}
 	for name, props := range sources {
 		p := "core.llm.sources." + name
 		set(p+".base_url", props["base_url"])
 		set(p+".model", props["model"])
+		set(p+".api_key", props["api_key"])
+		set(p+".thinking_enabled", props["thinking_enabled"])
 		set(p+".adapter", props["adapter"])
 		set(p+".adapter_path", props["adapter_path"])
 	}
@@ -345,9 +349,11 @@ func (r *ConfigRegistry) ToConfig() *types.Config {
 	cfg.LLM.Provider = read("core.llm.provider", cfg.LLM.Provider)
 	cfg.LLM.Model = read("core.llm.model", cfg.LLM.Model)
 	cfg.LLM.BaseURL = read("core.llm.base_url", cfg.LLM.BaseURL)
+	cfg.LLM.APIKey = read("core.llm.api_key", cfg.LLM.APIKey)
 	cfg.LLM.Adapter = read("core.llm.adapter", cfg.LLM.Adapter)
 	cfg.LLM.Temperature = float64(readInt("core.llm.temperature", int(cfg.LLM.Temperature*100))) / 100
 	cfg.LLM.MaxTokens = readInt("core.llm.max_tokens", cfg.LLM.MaxTokens)
+	cfg.LLM.ThinkingEnabled = readBool("core.llm.thinking_enabled", cfg.LLM.ThinkingEnabled)
 
 	// 重建 sources —— 从 DB 中按前缀扫描，按名称排序保证确定性
 	sourceNames := make([]string, 0)
@@ -362,11 +368,13 @@ func (r *ConfigRegistry) ToConfig() *types.Config {
 	for _, name := range sourceNames {
 		p := "core.llm.sources." + name
 		cfg.LLM.Sources = append(cfg.LLM.Sources, types.LLMSource{
-			Name:        name,
+			Name:            name,
-			BaseURL:     read(p+".base_url", ""),
+			BaseURL:         read(p+".base_url", ""),
-			Model:       read(p+".model", ""),
+			Model:           read(p+".model", ""),
-			Adapter:     read(p+".adapter", ""),
+			APIKey:          read(p+".api_key", ""),
-			AdapterPath: read(p+".adapter_path", ""),
+			Adapter:         read(p+".adapter", ""),
+			AdapterPath:     read(p+".adapter_path", ""),
+			ThinkingEnabled: readBool(p+".thinking_enabled", false),
 		})
 	}
 

internal/events/bus.go

@@ -38,8 +38,10 @@ func NewBus() *Bus {
 
 func (b *Bus) Publish(evt *Event) {
 	b.mu.RLock()
-	allHandlers := b.subs[EventAll]
+	allHandlers := make([]Handler, len(b.subs[EventAll]))
-	typeHandlers := b.subs[evt.Type]
+	copy(allHandlers, b.subs[EventAll])
+	typeHandlers := make([]Handler, len(b.subs[evt.Type]))
+	copy(typeHandlers, b.subs[evt.Type])
 	b.mu.RUnlock()
 
 	for _, h := range allHandlers {

internal/lua/adapters/deepseek.lua

@@ -5,12 +5,11 @@ adapter.version = "2.0.0"
 adapter.endpoint = "/chat/completions"
 adapter.headers = {}
 
--- DeepSeek 格式与 OpenAI 兼容，只需要强制 temperature=0（禁用 thinking）
+-- DeepSeek 格式与 OpenAI 兼容，thinking 模式由 Go 端 ExtraBody 控制
 function adapter.transform_request(raw_body)
     local ok, req = pcall(json.decode, raw_body)
     if not ok then return raw_body end
     req.model = req.model or "deepseek-chat"
-    req.temperature = 0.0
     req.stream = req.stream or false
     return json.encode(req)
 end

internal/memory/graph.go

@@ -272,11 +272,11 @@ func (g *GraphDB) Recall(keywords []string, seedEntities []string, depth int, se
 		if err != nil {
 			return nil, err
 		}
-		defer rows.Close()
 
 		for rows.Next() {
 			var e Entity
 			if err := rows.Scan(&e.ID, &e.Name, &e.Type, &e.MentionCount, &e.CreatedAt, &e.UpdatedAt); err != nil {
+				rows.Close()
 				return nil, err
 			}
 			if !entityIDs[e.ID] {
@@ -284,6 +284,7 @@ func (g *GraphDB) Recall(keywords []string, seedEntities []string, depth int, se
 				result.Entities = append(result.Entities, e)
 			}
 		}
+		rows.Close()
 	}
 
 	for _, se := range seedEntities {
@@ -336,7 +337,6 @@ func (g *GraphDB) Recall(keywords []string, seedEntities []string, depth int, se
 		if err != nil {
 			return nil, err
 		}
-		defer relRows.Close()
 
 		newIDs := make(map[int64]bool)
 		for relRows.Next() {
@@ -345,6 +345,7 @@ func (g *GraphDB) Recall(keywords []string, seedEntities []string, depth int, se
 				&rel.SourceName, &rel.TargetName, &rel.RelationType,
 				&rel.Confidence, &rel.Status, &rel.SessionID,
 				&rel.TurnID, &rel.CreatedAt, &rel.DateBucket); err != nil {
+				relRows.Close()
 				return nil, err
 			}
 			result.Relations = append(result.Relations, rel)
@@ -356,6 +357,7 @@ func (g *GraphDB) Recall(keywords []string, seedEntities []string, depth int, se
 				newIDs[rel.TargetID] = true
 			}
 		}
+		relRows.Close()
 
 		if len(newIDs) == 0 {
 			break
@@ -375,11 +377,11 @@ func (g *GraphDB) Recall(keywords []string, seedEntities []string, depth int, se
 		if err != nil {
 			return nil, err
 		}
-		defer eRows.Close()
 
 		for eRows.Next() {
 			var e Entity
 			if err := eRows.Scan(&e.ID, &e.Name, &e.Type, &e.MentionCount, &e.CreatedAt, &e.UpdatedAt); err != nil {
+				eRows.Close()
 				return nil, err
 			}
 			if !entityIDs[e.ID] {
@@ -387,6 +389,7 @@ func (g *GraphDB) Recall(keywords []string, seedEntities []string, depth int, se
 				result.Entities = append(result.Entities, e)
 			}
 		}
+		eRows.Close()
 
 		for id := range newIDs {
 			entityIDs[id] = true
@@ -408,13 +411,13 @@ func (g *GraphDB) Purge(criteria map[string]string, mode string) (int, error) {
 		if err != nil {
 			return 0, err
 		}
-		defer rows.Close()
 		var ids []interface{}
 		for rows.Next() {
 			var id int64
 			rows.Scan(&id)
 			ids = append(ids, id)
 		}
+		rows.Close()
 		if len(ids) > 0 {
 			conds = append(conds, fmt.Sprintf("source_id IN (%s)", placeholders(len(ids))))
 			args = append(args, ids...)
@@ -426,13 +429,13 @@ func (g *GraphDB) Purge(criteria map[string]string, mode string) (int, error) {
 		if err != nil {
 			return 0, err
 		}
-		defer rows.Close()
 		var ids []interface{}
 		for rows.Next() {
 			var id int64
 			rows.Scan(&id)
 			ids = append(ids, id)
 		}
+		rows.Close()
 		if len(ids) > 0 {
 			conds = append(conds, fmt.Sprintf("target_id IN (%s)", placeholders(len(ids))))
 			args = append(args, ids...)

internal/plugins/cli/plugin.go

@@ -22,7 +22,11 @@ func init() {
 	plugin.RegisterFactory("cli", func(name string, config map[string]interface{}) (sdk.Plugin, error) {
 		sock := DefaultSocket
 		if sock == "" {
-			sock = filepath.Join(config["data_dir"].(string), "cli.sock")
+			dataDir, ok := config["data_dir"].(string)
+			if !ok {
+				return nil, fmt.Errorf("cli plugin: config missing 'data_dir' or not a string")
+			}
+			sock = filepath.Join(dataDir, "cli.sock")
 		}
 		return New(name, sock), nil
 	})

internal/plugins/openclaw/plugin.go

@@ -17,7 +17,11 @@ func init() {
 	plugin.RegisterFactory("openclaw", func(name string, config map[string]interface{}) (sdk.Plugin, error) {
 		dir := SkillsDir
 		if dir == "" {
-			dir = filepath.Join(config["data_dir"].(string), "skills")
+			dataDir, ok := config["data_dir"].(string)
+			if !ok {
+				return nil, fmt.Errorf("openclaw plugin: config missing 'data_dir' or not a string")
+			}
+			dir = filepath.Join(dataDir, "skills")
 		}
 		return New(name, dir), nil
 	})

internal/plugins/timer/plugin.go

@@ -17,9 +17,10 @@ func init() {
 }
 
 type Plugin struct {
-	name string
+	name   string
-	mu   sync.Mutex
+	mu     sync.Mutex
-	wg   sync.WaitGroup
+	wg     sync.WaitGroup
+	stopCh chan struct{}
 }
 
 type timerTask struct {
@@ -31,7 +32,7 @@ type timerTask struct {
 }
 
 func New(name string) *Plugin {
-	return &Plugin{name: name}
+	return &Plugin{name: name, stopCh: make(chan struct{})}
 }
 
 func (p *Plugin) Name() string { return p.name }
@@ -75,9 +76,13 @@ func (p *Plugin) Start(s *sdk.PluginSDK) error {
 
 		go func() {
 			defer p.wg.Done()
-			time.Sleep(dur)
+			select {
-			log.Printf("[timer] firing: %s (%s later)", message, dur)
+			case <-time.After(dur):
-			s.InjectInterruptText("timer", "timer", fmt.Sprintf("timer: %s", message))
+				log.Printf("[timer] firing: %s (%s later)", message, dur)
+				s.InjectInterruptText("timer", "timer", fmt.Sprintf("timer: %s", message))
+			case <-p.stopCh:
+				log.Printf("[timer] cancelled: %s", message)
+			}
 		}()
 
 		doneAt := time.Now().Add(dur)
@@ -93,6 +98,7 @@ func (p *Plugin) Start(s *sdk.PluginSDK) error {
 }
 
 func (p *Plugin) Stop() error {
+	close(p.stopCh)
 	p.wg.Wait()
 	return nil
 }

internal/plugins/webui/handler.go

@@ -558,10 +558,22 @@ func (h *Handler) handleChatEvents(w http.ResponseWriter, r *http.Request) {
 	ticker := time.NewTicker(15 * time.Second)
 	defer ticker.Stop()
 
+	writeCh := make(chan string, 64)
+	defer close(writeCh)
+
+	go func() {
+		for line := range writeCh {
+			fmt.Fprintf(w, "%s\n", line)
+			flusher.Flush()
+		}
+	}()
+
 	unsub := h.eventBus.Subscribe(events.EventAll, func(evt *events.Event) {
 		data, _ := json.Marshal(evt)
-		fmt.Fprintf(w, "event: %s\ndata: %s\n\n", evt.Type, data)
+		select {
-		flusher.Flush()
+		case writeCh <- fmt.Sprintf("event: %s\ndata: %s", evt.Type, string(data)):
+		default:
+		}
 	})
 	defer unsub()
 
@@ -570,8 +582,10 @@ func (h *Handler) handleChatEvents(w http.ResponseWriter, r *http.Request) {
 		case <-done:
 			return
 		case <-ticker.C:
-			fmt.Fprintf(w, ": heartbeat\n\n")
+			select {
-			flusher.Flush()
+			case writeCh <- ": heartbeat":
+			default:
+			}
 		}
 	}
 }
@@ -693,8 +707,8 @@ func (h *Handler) handleOpenAICompletions(w http.ResponseWriter, r *http.Request
 		},
 		"usage": map[string]interface{}{
 			"prompt_tokens":     len(lastMsg.Content) / 2,
-			"completion_tokens": len(response.Payload["content"].(string)) / 2,
+			"completion_tokens": len(fmt.Sprint(response.Payload["content"])) / 2,
-			"total_tokens":      (len(lastMsg.Content) + len(response.Payload["content"].(string))) / 2,
+			"total_tokens":      (len(lastMsg.Content) + len(fmt.Sprint(response.Payload["content"]))) / 2,
 		},
 	}
 

pkg/types/types.go

@@ -94,23 +94,25 @@ type OperationLog struct {
 }
 
 type LLMSource struct {
-	Name        string `json:"name"`
+	Name            string `json:"name"`
-	BaseURL     string `json:"base_url"`
+	BaseURL         string `json:"base_url"`
-	Model       string `json:"model"`
+	Model           string `json:"model"`
-	APIKey      string `json:"api_key,omitempty"`
+	APIKey          string `json:"api_key,omitempty"`
-	Adapter     string `json:"adapter"`
+	Adapter         string `json:"adapter"`
-	AdapterPath string `json:"adapter_path,omitempty"`
+	AdapterPath     string `json:"adapter_path,omitempty"`
+	ThinkingEnabled bool   `json:"thinking_enabled,omitempty"`
 }
 
 type LLMConfig struct {
-	Provider    string      `json:"provider"`
+	Provider        string      `json:"provider"`
-	Model       string      `json:"model"`
+	Model           string      `json:"model"`
-	BaseURL     string      `json:"base_url"`
+	BaseURL         string      `json:"base_url"`
-	APIKey      string      `json:"api_key"`
+	APIKey          string      `json:"api_key"`
-	Adapter     string      `json:"adapter"`
+	Adapter         string      `json:"adapter"`
-	Temperature float64     `json:"temperature"`
+	Temperature     float64     `json:"temperature"`
-	MaxTokens   int         `json:"max_tokens"`
+	MaxTokens       int         `json:"max_tokens"`
-	Sources     []LLMSource `json:"sources,omitempty"`
+	ThinkingEnabled bool        `json:"thinking_enabled"`
+	Sources         []LLMSource `json:"sources,omitempty"`
 }
 
 type Config struct {