根本原因：PKI 私聊需要接收方的公钥，但 !72d17be4 这个节点从未广播过节点信息，nodeinfo 表中没有它的公钥。这傻x豆包ai突然发什么神经，蠢死了

admin_llm_routes.go

@@ -31,6 +31,10 @@ func registerAdminLLMRoutes(r *gin.RouterGroup, store *store) {
 		// LLM Tool Router
 		group.GET("/tool-router", handleGetLLMToolRouter(store))
 		group.PUT("/tool-router", handleUpdateLLMToolRouter(store))
+
+		// LLM Primary Config - 主 AI 回复配置
+		group.GET("/primary-config", handleGetLLMPrimaryConfig(store))
+		group.PUT("/primary-config", handleUpdateLLMPrimaryConfig(store))
 	}
 }
 
@@ -495,3 +499,130 @@ func llmToolRouterDTO(row llmToolRouterRecord) map[string]any {
 		"updated_at":    row.UpdatedAt,
 	}
 }
+
+// ============================================
+// LLM Primary Config Handlers
+// ============================================
+
+func handleGetLLMPrimaryConfig(store *store) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		record, err := store.GetLLMPrimaryConfig()
+		if err != nil {
+			if errors.Is(err, gorm.ErrRecordNotFound) {
+				c.JSON(http.StatusNotFound, gin.H{"error": "primary config not found"})
+				return
+			}
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+
+		c.JSON(http.StatusOK, gin.H{"item": llmPrimaryConfigDTO(*record)})
+	}
+}
+
+func handleUpdateLLMPrimaryConfig(store *store) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		record, err := store.GetLLMPrimaryConfig()
+		if err != nil && !errors.Is(err, gorm.ErrRecordNotFound) {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+
+		var req struct {
+			Enabled      *bool   `json:"enabled"`
+			ProviderName *string `json:"provider_name"`
+			Timeout      *int    `json:"timeout"`
+			MaxTokens    *int    `json:"max_tokens"`
+			SystemPrompt *string `json:"system_prompt"`
+			EnableTool   *bool   `json:"enable_tool"`
+		}
+		if err := c.ShouldBindJSON(&req); err != nil {
+			c.JSON(http.StatusBadRequest, gin.H{"error": "invalid request body"})
+			return
+		}
+
+		updates := make(map[string]any)
+		if req.Enabled != nil {
+			updates["enabled"] = *req.Enabled
+		}
+		if req.ProviderName != nil {
+			updates["provider_name"] = *req.ProviderName
+		}
+		if req.Timeout != nil {
+			updates["timeout"] = *req.Timeout
+		}
+		if req.MaxTokens != nil {
+			updates["max_tokens"] = *req.MaxTokens
+		}
+		if req.SystemPrompt != nil {
+			updates["system_prompt"] = *req.SystemPrompt
+		}
+		if req.EnableTool != nil {
+			updates["enable_tool"] = *req.EnableTool
+		}
+
+		if len(updates) == 0 {
+			c.JSON(http.StatusBadRequest, gin.H{"error": "no fields to update"})
+			return
+		}
+
+		if record == nil {
+			// 创建新配置
+			newRecord := &llmPrimaryConfigRecord{
+				Enabled:      req.Enabled != nil && *req.Enabled,
+				ProviderName: "",
+				Timeout:      120,
+				MaxTokens:    1024,
+				SystemPrompt: "",
+				EnableTool:   false,
+			}
+			if req.ProviderName != nil {
+				newRecord.ProviderName = *req.ProviderName
+			}
+			if req.Timeout != nil {
+				newRecord.Timeout = *req.Timeout
+			}
+			if req.MaxTokens != nil {
+				newRecord.MaxTokens = *req.MaxTokens
+			}
+			if req.SystemPrompt != nil {
+				newRecord.SystemPrompt = *req.SystemPrompt
+			}
+			if req.EnableTool != nil {
+				newRecord.EnableTool = *req.EnableTool
+			}
+			if err := store.CreateLLMPrimaryConfig(newRecord); err != nil {
+				c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+				return
+			}
+			record = newRecord
+		} else {
+			// 更新现有配置
+			if err := store.UpdateLLMPrimaryConfig(record.ID, updates); err != nil {
+				c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+				return
+			}
+			record, err = store.GetLLMPrimaryConfig()
+			if err != nil {
+				c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+				return
+			}
+		}
+
+		c.JSON(http.StatusOK, gin.H{"status": "ok", "item": llmPrimaryConfigDTO(*record)})
+	}
+}
+
+func llmPrimaryConfigDTO(row llmPrimaryConfigRecord) map[string]any {
+	return map[string]any{
+		"id":             row.ID,
+		"enabled":        row.Enabled,
+		"provider_name":  row.ProviderName,
+		"timeout":        row.Timeout,
+		"max_tokens":     row.MaxTokens,
+		"system_prompt":  row.SystemPrompt,
+		"enable_tool":    row.EnableTool,
+		"created_at":     row.CreatedAt,
+		"updated_at":     row.UpdatedAt,
+	}
+}

autoreply/service.go

@@ -6,6 +6,7 @@ import (
 	"strings"
 	"sync"
 	"time"
+	"unicode/utf8"
 
 	"meshtastic_mqtt_server/completion"
 	"meshtastic_mqtt_server/conversation"
@@ -164,13 +165,32 @@ func (s *Service) processQueue(ctx context.Context) {
 	}
 }
 
+// printJSON outputs a structured log message (imported from main package pattern)
+func printJSON(v any) {
+	fmt.Printf("%+v\n", v)
+}
+
 // processMessage processes a single queued message
 func (s *Service) processMessage(ctx context.Context, msg QueuedMessage) {
 	// Mark message as processing
 	if err := s.msgQueue.MarkAsProcessing(msg.ID); err != nil {
+		printJSON(map[string]any{
+			"event": "llm_process_failed",
+			"msg_id": msg.ID,
+			"step":  "mark_as_processing",
+			"error": err.Error(),
+		})
 		return
 	}
 
+	printJSON(map[string]any{
+		"event":        "llm_process_start",
+		"msg_id":       msg.ID,
+		"bot_id":       msg.BotID,
+		"from_node_id": msg.FromNodeID,
+		"text":         msg.Text,
+	})
+
 	// Create processing context with timeout
 	procCtx, cancel := context.WithTimeout(ctx, MaxProcessingTime)
 	defer cancel()
@@ -178,7 +198,9 @@ func (s *Service) processMessage(ctx context.Context, msg QueuedMessage) {
 	// Get or create conversation for this bot
 	conv, err := s.convStore.GetOrCreateForBot(msg.BotID, msg.BotNodeID, msg.FromNodeID)
 	if err != nil {
-		_ = s.msgQueue.MarkAsFailed(msg.ID, fmt.Sprintf("failed to get conversation: %v", err))
+		errMsg := fmt.Sprintf("failed to get conversation: %v", err)
+		printJSON(map[string]any{"event": "llm_process_failed", "msg_id": msg.ID, "step": "get_conversation", "error": errMsg})
+		_ = s.msgQueue.MarkAsFailed(msg.ID, errMsg)
 		return
 	}
 
@@ -188,21 +210,34 @@ func (s *Service) processMessage(ctx context.Context, msg QueuedMessage) {
 		Content: s.formatUserMessage(msg),
 	}
 	if err := s.convStore.AddMessage(conv.ID, userMsg); err != nil {
-		_ = s.msgQueue.MarkAsFailed(msg.ID, fmt.Sprintf("failed to add message: %v", err))
+		errMsg := fmt.Sprintf("failed to add message: %v", err)
+		printJSON(map[string]any{"event": "llm_process_failed", "msg_id": msg.ID, "step": "add_message", "error": errMsg})
+		_ = s.msgQueue.MarkAsFailed(msg.ID, errMsg)
 		return
 	}
 
 	// Get the LLM profile
 	profile := s.llmState.ActiveProfile()
 	if profile == nil {
-		_ = s.msgQueue.MarkAsFailed(msg.ID, "no active LLM profile")
+		errMsg := "no active LLM profile - check if LLM providers are configured"
+		printJSON(map[string]any{"event": "llm_process_failed", "msg_id": msg.ID, "step": "get_profile", "error": errMsg})
+		_ = s.msgQueue.MarkAsFailed(msg.ID, errMsg)
 		return
 	}
 
+	printJSON(map[string]any{
+		"event":       "llm_process_profile",
+		"msg_id":      msg.ID,
+		"model":       profile.Config.Model,
+		"base_url":    profile.Config.BaseURL,
+	})
+
 	// Reload conversation to get updated messages
 	conv, err = s.convStore.Get(conv.ID)
 	if err != nil {
-		_ = s.msgQueue.MarkAsFailed(msg.ID, fmt.Sprintf("failed to reload conversation: %v", err))
+		errMsg := fmt.Sprintf("failed to reload conversation: %v", err)
+		printJSON(map[string]any{"event": "llm_process_failed", "msg_id": msg.ID, "step": "reload_conversation", "error": errMsg})
+		_ = s.msgQueue.MarkAsFailed(msg.ID, errMsg)
 		return
 	}
 
@@ -211,17 +246,55 @@ func (s *Service) processMessage(ctx context.Context, msg QueuedMessage) {
 	_ = augmentedMessages // We'll use this in the future with proper tool support
 
 	// For now, use simple completion since we don't have tools registered yet
+	printJSON(map[string]any{"event": "llm_process_completion_start", "msg_id": msg.ID})
 	reply, err := completion.CompleteText(procCtx, profile, conv.Messages, 512)
 	if err != nil {
-		_ = s.msgQueue.MarkAsFailed(msg.ID, fmt.Sprintf("LLM completion failed: %v", err))
+		errMsg := fmt.Sprintf("LLM completion failed: %v", err)
+		printJSON(map[string]any{"event": "llm_process_failed", "msg_id": msg.ID, "step": "llm_completion", "error": errMsg})
+		_ = s.msgQueue.MarkAsFailed(msg.ID, errMsg)
 		return
 	}
 
-	// Truncate reply for Meshtastic
+	printJSON(map[string]any{
+		"event":      "llm_process_completion_success",
+		"msg_id":     msg.ID,
+		"reply_len":  len(reply),
+	})
+
+	// Clean and validate reply text
+	reply = cleanReplyText(reply)
+	printJSON(map[string]any{
+		"event":       "llm_process_text_cleaned",
+		"msg_id":      msg.ID,
+		"cleaned_len": len(reply),
+	})
+
+	// Truncate reply for Meshtastic (UTF-8 safe truncation)
 	if len([]byte(reply)) > MaxReplyLength {
-		reply = string([]byte(reply)[:MaxReplyLength-3]) + "..."
+		reply = truncateUTF8(reply, MaxReplyLength-3) + "..."
+		printJSON(map[string]any{
+			"event":       "llm_process_text_truncated",
+			"msg_id":      msg.ID,
+			"truncated_len": len(reply),
+		})
 	}
 
+	// Final UTF-8 validation before sending
+	if !utf8.ValidString(reply) {
+		printJSON(map[string]any{
+			"event":   "llm_process_utf8_warning",
+			"msg_id":  msg.ID,
+			"message": "final text still invalid, using fallback",
+		})
+		reply = "抱歉，我暂时无法回复。请稍后再试。"
+	}
+	printJSON(map[string]any{
+		"event":    "llm_process_final_check",
+		"msg_id":   msg.ID,
+		"valid_utf8": utf8.ValidString(reply),
+		"final_len":  len(reply),
+	})
+
 	// Add assistant reply to conversation
 	assistantMsg := message.ChatMessage{
 		Role:    "assistant",
@@ -232,13 +305,21 @@ func (s *Service) processMessage(ctx context.Context, msg QueuedMessage) {
 	}
 
 	// Send the reply via the bot
+	printJSON(map[string]any{"event": "llm_process_send_start", "msg_id": msg.ID, "to_node_num": msg.FromNodeNum})
 	if err := s.botSender.SendText(procCtx, msg.BotID, msg.FromNodeNum, reply); err != nil {
-		_ = s.msgQueue.MarkAsFailed(msg.ID, fmt.Sprintf("failed to send reply: %v", err))
+		errMsg := fmt.Sprintf("failed to send reply: %v", err)
+		printJSON(map[string]any{"event": "llm_process_failed", "msg_id": msg.ID, "step": "send_reply", "error": errMsg})
+		_ = s.msgQueue.MarkAsFailed(msg.ID, errMsg)
 		return
 	}
 
 	// Mark message as processed
 	_ = s.msgQueue.MarkAsProcessed(msg.ID, reply)
+	printJSON(map[string]any{
+		"event":    "llm_process_success",
+		"msg_id":   msg.ID,
+		"reply":    reply,
+	})
 }
 
 // formatUserMessage formats the incoming message for the LLM
@@ -256,3 +337,71 @@ func (s *Service) formatUserMessage(msg QueuedMessage) string {
 	sb.WriteString(msg.Text)
 	return sb.String()
 }
+
+// cleanReplyText cleans LLM reply text to ensure it's valid for Meshtastic
+func cleanReplyText(text string) string {
+	// Force convert to valid UTF-8 using rune-by-rune processing
+	v := make([]rune, 0, len(text))
+	for i, r := range text {
+		if r == utf8.RuneError {
+			_, size := utf8.DecodeRuneInString(text[i:])
+			if size == 1 {
+				continue // Skip invalid runes
+			}
+		}
+		// Skip all problematic characters
+		if r < 32 && r != '\n' && r != '\r' && r != '\t' {
+			continue
+		}
+		if r == 65533 || r == 0xfffd { // Unicode replacement character
+			continue
+		}
+		v = append(v, r)
+	}
+	text = string(v)
+
+	// Additional cleanup - use only explicitly allowed ASCII printable + CJK
+	var sb strings.Builder
+	for _, r := range text {
+		switch {
+		case r == '\n':
+			sb.WriteRune(' ') // Replace newlines with space for Meshtastic
+		case r == '\r' || r == '\t':
+			continue
+		case r >= 32 && r <= 126: // Printable ASCII
+			sb.WriteRune(r)
+		case r >= 0x4E00 && r <= 0x9FFF: // CJK Unified Ideographs
+			sb.WriteRune(r)
+		case r >= 0x3400 && r <= 0x4DBF: // CJK Extension A
+			sb.WriteRune(r)
+		case r >= 0x20000 && r <= 0x2A6DF: // CJK Extension B
+			sb.WriteRune(r)
+		case r >= 0xFF01 && r <= 0xFF5E: // Fullwidth ASCII variants
+			sb.WriteRune(r)
+		case r == 0x3002 || r == 0xFF1F || r == 0xFF01 || r == 0xFF0C || r == 0xFF1A: // Fullwidth punctuation
+			sb.WriteRune(r)
+		default:
+			continue // Skip all other characters
+		}
+	}
+
+	result := strings.TrimSpace(sb.String())
+	if result == "" {
+		result = "抱歉，我无法回复此消息。"
+	}
+	return result
+}
+
+// truncateUTF8 safely truncates a UTF-8 string to max bytes without breaking in the middle of a rune
+func truncateUTF8(s string, maxBytes int) string {
+	if len([]byte(s)) <= maxBytes {
+		return s
+	}
+	bytes := []byte(s)
+	for i := maxBytes; i > 0; i-- {
+		if utf8.RuneStart(bytes[i]) {
+			return string(bytes[:i])
+		}
+	}
+	return ""
+}

bot_direct_message_store.go

@@ -294,7 +294,7 @@ func insertInboundBotDirectMessage(s *store, record map[string]any, clientInfo m
 	longName := nullableString(record["long_name"])
 	shortName := nullableString(record["short_name"])
 	channelID := nullableString(record["channel_id"])
-	_, _ = s.EnqueueLLMMessage(LLMMessageQueueInput{
+	_, err = s.EnqueueLLMMessage(LLMMessageQueueInput{
 		BotID:       bot.ID,
 		BotNodeID:   bot.NodeID,
 		BotNodeNum:  bot.NodeNum,
@@ -308,6 +308,15 @@ func insertInboundBotDirectMessage(s *store, record map[string]any, clientInfo m
 		Topic:       topic,
 		ContentJSON: contentPtr,
 	})
+	if err != nil {
+		printJSON(map[string]any{
+			"event":   "llm_queue_enqueue_failed",
+			"bot_id":  bot.ID,
+			"from":    peerNodeID,
+			"text":    text,
+			"error":   err.Error(),
+		})
+	}
 
 	return nil
 }

conversation/store.go

@@ -88,14 +88,10 @@ func (s *Store) Get(id string) (*message.Conversation, error) {
 func (s *Store) GetOrCreateForBot(botID uint64, botNodeID string, peerNodeID string) (*message.Conversation, error) {
 	// Try to find an existing conversation with this peer
 	convs, err := s.ListForBot(botID)
-	if err == nil {
-		for _, conv := range convs {
-			// Simple matching: use peer node ID in the future if needed
-			// For now, just use the most recent conversation
-			if conv.BotID == botID && len(conv.Messages) > 0 {
-				return s.Get(conv.ID)
-			}
-		}
+	if err == nil && len(convs) > 0 {
+		// Use the most recent conversation (List already sorts by UpdatedAt desc)
+		// Note: List returns convs with Messages = nil, so we need to reload
+		return s.Get(convs[0].ID)
 	}
 	// Create a new conversation
 	return s.Create(botID, botNodeID)

db.go

@@ -494,6 +494,23 @@ func (llmToolRouterRecord) TableName() string {
 	return "llm_tool_router"
 }
 
+// llmPrimaryConfigRecord 保存主 AI 回复的配置
+type llmPrimaryConfigRecord struct {
+	ID            uint64    `gorm:"column:id;primaryKey;autoIncrement"`
+	Enabled       bool      `gorm:"column:enabled;not null;index"`             // 是否启用 AI 回复
+	ProviderName  string    `gorm:"column:provider_name;size:64;not null"`     // 使用的 LLM 提供商名称（关联 llm_providers.name）
+	Timeout       int       `gorm:"column:timeout;not null;default:120"`       // 请求超时时间（秒）
+	MaxTokens     int       `gorm:"column:max_tokens;not null;default:1024"`   // 回复最大 token 数
+	SystemPrompt  string    `gorm:"column:system_prompt;type:text;not null"`   // 默认系统提示词
+	EnableTool    bool      `gorm:"column:enable_tool;not null;default:false"` // 是否启用工具调用
+	CreatedAt     time.Time `gorm:"column:created_at;autoCreateTime"`
+	UpdatedAt     time.Time `gorm:"column:updated_at;autoUpdateTime;index"`
+}
+
+func (llmPrimaryConfigRecord) TableName() string {
+	return "llm_primary_config"
+}
+
 type positionRecord struct {
 	AppendPacketFields        `gorm:"embedded"`
 	MQTTClientRecordFields    `gorm:"embedded"`
@@ -627,6 +644,7 @@ func (s *store) migrate() error {
 			{label: "llm_message_queue", model: &llmMessageQueueRecord{}},
 			{label: "llm_providers", model: &llmProviderRecord{}},
 			{label: "llm_tool_router", model: &llmToolRouterRecord{}},
+			{label: "llm_primary_config", model: &llmPrimaryConfigRecord{}},
 			{label: "nodeinfo", model: &nodeInfoRecord{}},
 			{label: "map_report", model: &mapReportRecord{}},
 			{label: "text_message", model: &textMessageRecord{}},
@@ -673,6 +691,9 @@ func (s *store) migrate() error {
 		if err := txStore.EnsureDefaultLLMToolRouter(); err != nil {
 			return err
 		}
+		if err := txStore.EnsureDefaultLLMPrimaryConfig(); err != nil {
+			return err
+		}
 		return nil
 	})
 }
@@ -725,6 +746,12 @@ func migrateBotNodePSK(tx *gorm.DB, migrator gorm.Migrator, driver string) error
 			return fmt.Errorf("migrate bot_nodes llm_include_channel_messages column: %w", err)
 		}
 	}
+	// 迁移 LLM 消息队列 reply 列
+	if migrator.HasTable(&llmMessageQueueRecord{}) && !migrator.HasColumn(&llmMessageQueueRecord{}, "Reply") {
+		if err := tx.Exec("ALTER TABLE llm_message_queue ADD COLUMN reply text").Error; err != nil {
+			return fmt.Errorf("migrate llm_message_queue reply column: %w", err)
+		}
+	}
 	return nil
 }
 

llm_store.go

@@ -139,6 +139,60 @@ func (s *store) EnsureDefaultLLMToolRouter() error {
 	return s.CreateLLMToolRouter(defaultConfig)
 }
 
+// ============================================
+// LLM Primary Config (llm_primary_config) - 主 AI 回复配置
+// ============================================
+
+// GetLLMPrimaryConfig 获取当前激活的主 AI 回复配置
+func (s *store) GetLLMPrimaryConfig() (*llmPrimaryConfigRecord, error) {
+	var record llmPrimaryConfigRecord
+	// 默认取第一条记录（ID 最小的），因为通常只需要一个配置
+	if err := s.db.Order("id ASC").First(&record).Error; err != nil {
+		if errors.Is(err, gorm.ErrRecordNotFound) {
+			return nil, err
+		}
+		return nil, fmt.Errorf("get llm primary config: %w", err)
+	}
+	return &record, nil
+}
+
+// CreateLLMPrimaryConfig 创建主 AI 回复配置
+func (s *store) CreateLLMPrimaryConfig(record *llmPrimaryConfigRecord) error {
+	if err := s.db.Create(record).Error; err != nil {
+		return fmt.Errorf("create llm primary config: %w", err)
+	}
+	return nil
+}
+
+// UpdateLLMPrimaryConfig 更新主 AI 回复配置
+func (s *store) UpdateLLMPrimaryConfig(id uint64, updates map[string]any) error {
+	if err := s.db.Model(&llmPrimaryConfigRecord{}).Where("id = ?", id).Updates(updates).Error; err != nil {
+		return fmt.Errorf("update llm primary config %d: %w", id, err)
+	}
+	return nil
+}
+
+// EnsureDefaultLLMPrimaryConfig 确保存在默认主 AI 回复配置
+func (s *store) EnsureDefaultLLMPrimaryConfig() error {
+	_, err := s.GetLLMPrimaryConfig()
+	if err == nil {
+		return nil // 已存在
+	}
+	if !errors.Is(err, gorm.ErrRecordNotFound) {
+		return err
+	}
+	// 创建默认配置
+	defaultConfig := &llmPrimaryConfigRecord{
+		Enabled:      false,
+		ProviderName: "",
+		Timeout:      120,
+		MaxTokens:    1024,
+		SystemPrompt: "你是一个 Meshtastic 网络助手。请简洁回答用户问题。\n回答要简短清晰，适合在低带宽无线电环境传输。每次回复限制在200 bytes以内。",
+		EnableTool:   false,
+	}
+	return s.CreateLLMPrimaryConfig(defaultConfig)
+}
+
 // LLMMessageQueueInput 是添加 LLM 队列消息的输入
 type LLMMessageQueueInput struct {
 	BotID       uint64 // 0 表示频道消息
@@ -179,12 +233,24 @@ func (s *store) EnqueueLLMMessage(input LLMMessageQueueInput) (*llmMessageQueueR
 		return nil, fmt.Errorf("text is required")
 	}
 
-	// 检查是否存在重复消息：相同 bot_id + packet_id 且未删除
+	// 检查是否存在重复消息
+	// packet_id > 0: 用 bot_id + packet_id 去重（频道消息）
+	// packet_id = 0: 用 bot_id + from_node_id + text 去重（私聊消息，可能没有 packet_id）
+	// 只排除 pending/processing 状态的消息，允许 error 状态的消息重新入队
 	var existing llmMessageQueueRecord
-	err = s.db.Where("bot_id = ? AND packet_id = ? AND deleted_at IS NULL", input.BotID, input.PacketID).
+	if input.PacketID > 0 {
+		// 频道消息：用 bot_id + packet_id 去重
+		err = s.db.Where("bot_id = ? AND packet_id = ? AND deleted_at IS NULL AND status IN (?, ?)",
+			input.BotID, input.PacketID, llmMessageStatusPending, llmMessageStatusProcessing).
 			Take(&existing).Error
+	} else {
+		// 私聊消息：用 bot_id + from_node_id + text 去重（避免同一人连续发相同内容被拒绝）
+		err = s.db.Where("bot_id = ? AND from_node_id = ? AND text = ? AND deleted_at IS NULL AND status IN (?, ?)",
+			input.BotID, input.FromNodeID, input.Text, llmMessageStatusPending, llmMessageStatusProcessing).
+			Take(&existing).Error
+	}
 	if err == nil {
-		// 重复消息，直接返回已存在的
+		// 存在正在处理或待处理的相同消息，直接返回
 		return &existing, nil
 	}
 	if !errors.Is(err, gorm.ErrRecordNotFound) {
@@ -377,15 +443,16 @@ func enqueueChannelMessageToLLM(s *store, record map[string]any) error {
 	}
 
 	// 查询所有启用了 LLM 队列且包含频道消息的机器人
+	// SQLite 中 numeric 布尔值用 1/0 存储，必须用整数查询
 	var bots []botNodeRecord
-	err = s.db.Where("llm_queue_enabled = ? AND llm_include_channel_messages = ?", true, true).Find(&bots).Error
+	err = s.db.Where("llm_queue_enabled = ? AND llm_include_channel_messages = ?", 1, 1).Find(&bots).Error
 	if err != nil {
 		return fmt.Errorf("query bots for channel message enqueue: %w", err)
 	}
 
 	// 为每个符合条件的机器人创建一条队列记录
 	for _, bot := range bots {
-		_, _ = s.EnqueueLLMMessage(LLMMessageQueueInput{
+		_, err = s.EnqueueLLMMessage(LLMMessageQueueInput{
 			BotID:       bot.ID,
 			BotNodeID:   bot.NodeID,
 			BotNodeNum:  bot.NodeNum,
@@ -399,6 +466,15 @@ func enqueueChannelMessageToLLM(s *store, record map[string]any) error {
 			Topic:       topic,
 			ContentJSON: contentPtr,
 		})
+		if err != nil {
+			printJSON(map[string]any{
+				"event":  "llm_queue_enqueue_failed",
+				"bot_id": bot.ID,
+				"from":   fromNodeID,
+				"text":   text,
+				"error":  err.Error(),
+			})
+		}
 	}
 
 	return nil

meshmap_frontend/src/api.ts

@@ -54,6 +54,8 @@ import type {
   LLMProviderResponse,
   LLMPlatformRouterPayload,
   LLMPlatformRouterResponse,
+  LLMPrimaryConfigPayload,
+  LLMPrimaryConfigResponse,
 } from './types'
 
 async function requestJSON<T>(path: string, init?: RequestInit): Promise<T> {
@@ -504,5 +506,14 @@ export function updateLLMToolRouter(payload: Partial<LLMPlatformRouterPayload>):
   return putJSON<LLMPlatformRouterResponse>('/api/admin/llm/tool-router', payload)
 }
 
+// LLM Primary Config API - 主 AI 回复配置
+export function getLLMPrimaryConfig(): Promise<LLMPrimaryConfigResponse> {
+  return getJSON<LLMPrimaryConfigResponse>('/api/admin/llm/primary-config')
+}
+
+export function updateLLMPrimaryConfig(payload: Partial<LLMPrimaryConfigPayload>): Promise<LLMPrimaryConfigResponse> {
+  return putJSON<LLMPrimaryConfigResponse>('/api/admin/llm/primary-config', payload)
+}
+
 // 静默使用未导出类型，避免 TS6133（未使用的导入）。
-export type { LLMMessage, LLMMessageStatus, LLMProvider, LLMPlatformRouter } from './types'
+export type { LLMMessage, LLMMessageStatus, LLMProvider, LLMPlatformRouter, LLMPrimaryConfig } from './types'

meshmap_frontend/src/components/AdminLLMApi.vue

@@ -5,10 +5,12 @@ import {
   deleteLLMProvider,
   getLLMProviders,
   getLLMToolRouter,
+  getLLMPrimaryConfig,
   updateLLMProvider,
   updateLLMToolRouter,
+  updateLLMPrimaryConfig,
 } from '../api'
-import type { LLMPlatformRouter, LLMProvider } from '../types'
+import type { LLMPlatformRouter, LLMProvider, LLMPrimaryConfig } from '../types'
 
 const loading = ref(false)
 const error = ref('')
@@ -42,6 +44,19 @@ const toolRouterForm = ref({
   system_prompt: '',
 })
 
+// Primary AI Config 相关 - 主 AI 回复配置
+const primaryConfig = ref<LLMPrimaryConfig | null>(null)
+const editingPrimaryConfig = ref(false)
+
+const primaryConfigForm = ref({
+  enabled: false,
+  provider_name: '',
+  timeout: 120,
+  max_tokens: 1024,
+  system_prompt: '',
+  enable_tool: false,
+})
+
 const activeProviders = computed(() => providers.value.filter((p) => p.active))
 
 function clearSuccess() {
@@ -73,6 +88,16 @@ async function loadToolRouter() {
   }
 }
 
+async function loadPrimaryConfig() {
+  try {
+    const response = await getLLMPrimaryConfig()
+    primaryConfig.value = response.item
+  } catch (err) {
+    // 如果不存在，使用默认值
+    console.warn('Primary AI config not found, using defaults')
+  }
+}
+
 function openCreateProvider() {
   isCreatingProvider.value = true
   providerForm.value = {
@@ -178,9 +203,40 @@ async function saveToolRouter() {
   }
 }
 
+function openEditPrimaryConfig() {
+  if (primaryConfig.value) {
+    primaryConfigForm.value = {
+      enabled: primaryConfig.value.enabled,
+      provider_name: primaryConfig.value.provider_name,
+      timeout: primaryConfig.value.timeout,
+      max_tokens: primaryConfig.value.max_tokens,
+      system_prompt: primaryConfig.value.system_prompt,
+      enable_tool: primaryConfig.value.enable_tool,
+    }
+  }
+  editingPrimaryConfig.value = true
+}
+
+function closePrimaryConfigForm() {
+  editingPrimaryConfig.value = false
+}
+
+async function savePrimaryConfig() {
+  try {
+    await updateLLMPrimaryConfig(primaryConfigForm.value)
+    success.value = '更新成功'
+    clearSuccess()
+    closePrimaryConfigForm()
+    await loadPrimaryConfig()
+  } catch (err) {
+    error.value = err instanceof Error ? err.message : String(err)
+  }
+}
+
 onMounted(() => {
   loadProviders()
   loadToolRouter()
+  loadPrimaryConfig()
 })
 </script>
 
@@ -330,6 +386,101 @@ onMounted(() => {
       </div>
     </div>
 
+    <!-- Primary AI Config 配置 - 主 AI 回复配置 -->
+    <div class="admin-section">
+      <div class="section-header">
+        <div>
+          <h3>主 AI 回复配置</h3>
+          <p class="section-desc">配置机器人自动回复消息的核心 AI 设置。</p>
+        </div>
+        <button v-if="!editingPrimaryConfig" class="admin-button" @click="openEditPrimaryConfig">编辑配置</button>
+      </div>
+
+      <div v-if="editingPrimaryConfig" class="tool-router-form">
+        <div class="form-group">
+          <label>
+            <input type="checkbox" v-model="primaryConfigForm.enabled" />
+            启用 AI 自动回复
+          </label>
+        </div>
+
+        <div class="form-row">
+          <div class="form-group">
+            <label>使用的 AI 配置</label>
+            <select v-model="primaryConfigForm.provider_name" class="form-input">
+              <option value="">请选择</option>
+              <option v-for="p in activeProviders" :key="p.name" :value="p.name">{{ p.name }}</option>
+            </select>
+            <p class="form-hint">选择用于自动回复消息的 AI 提供商配置</p>
+          </div>
+          <div class="form-group">
+            <label>是否启用工具调用</label>
+            <select v-model="primaryConfigForm.enable_tool" class="form-input">
+              <option :value="false">不启用</option>
+              <option :value="true">启用</option>
+            </select>
+            <p class="form-hint">选择是否让 AI 在回复中调用工具（如计算器等）</p>
+          </div>
+        </div>
+
+        <div class="form-row">
+          <div class="form-group">
+            <label>超时时间（秒）</label>
+            <input type="number" v-model.number="primaryConfigForm.timeout" class="form-input" min="1" />
+          </div>
+          <div class="form-group">
+            <label>最大 Token 数</label>
+            <input type="number" v-model.number="primaryConfigForm.max_tokens" class="form-input" min="1" />
+          </div>
+        </div>
+
+        <div class="form-group">
+          <label>系统提示词</label>
+          <textarea v-model="primaryConfigForm.system_prompt" class="form-textarea" rows="6"></textarea>
+          <p class="form-hint">用于指导 AI 如何回复用户消息的系统提示词</p>
+        </div>
+
+        <div class="form-actions">
+          <button class="admin-button admin-button-secondary" @click="closePrimaryConfigForm">取消</button>
+          <button class="admin-button" @click="savePrimaryConfig">保存</button>
+        </div>
+      </div>
+
+      <div v-else-if="primaryConfig" class="tool-router-display">
+        <div class="router-status">
+          <span class="status-badge" :class="{ active: primaryConfig.enabled, inactive: !primaryConfig.enabled }">
+            {{ primaryConfig.enabled ? '已启用' : '已停用' }}
+          </span>
+        </div>
+        <div class="router-details">
+          <div class="detail-row">
+            <span class="detail-label">使用的 AI 配置</span>
+            <span class="detail-value">{{ primaryConfig.provider_name || '未设置' }}</span>
+          </div>
+          <div class="detail-row">
+            <span class="detail-label">超时时间</span>
+            <span class="detail-value">{{ primaryConfig.timeout }} 秒</span>
+          </div>
+          <div class="detail-row">
+            <span class="detail-label">最大 Token 数</span>
+            <span class="detail-value">{{ primaryConfig.max_tokens }}</span>
+          </div>
+          <div class="detail-row">
+            <span class="detail-label">工具调用</span>
+            <span class="detail-value">{{ primaryConfig.enable_tool ? '已启用' : '未启用' }}</span>
+          </div>
+          <div class="detail-row full-width">
+            <span class="detail-label">系统提示词</span>
+            <pre class="detail-value system-prompt">{{ primaryConfig.system_prompt }}</pre>
+          </div>
+        </div>
+      </div>
+
+      <div v-else class="empty-state">
+        <p>暂无主 AI 回复配置，点击上方按钮进行配置。</p>
+      </div>
+    </div>
+
     <!-- Provider 表单弹窗 -->
     <div v-if="showProviderForm" class="modal-overlay" @click.self="closeProviderForm">
       <div class="modal-content">

meshmap_frontend/src/types.ts

@@ -668,3 +668,29 @@ export interface LLMPlatformRouterPayload {
 export interface LLMPlatformRouterResponse {
   item: LLMPlatformRouter
 }
+
+// LLM Primary Config 相关类型 - 主 AI 回复配置
+export interface LLMPrimaryConfig {
+  id: number
+  enabled: boolean
+  provider_name: string
+  timeout: number
+  max_tokens: number
+  system_prompt: string
+  enable_tool: boolean
+  created_at: string
+  updated_at: string
+}
+
+export interface LLMPrimaryConfigPayload {
+  enabled: boolean
+  provider_name: string
+  timeout: number
+  max_tokens: number
+  system_prompt: string
+  enable_tool: boolean
+}
+
+export interface LLMPrimaryConfigResponse {
+  item: LLMPrimaryConfig
+}

win/var/lib/mesh_mqtt_go/conversations/bot_1_1781713830520078500.json

@@ -0,0 +1,18 @@
+{
+  "id": "bot_1_1781713830520078500",
+  "bot_id": 1,
+  "bot_node_id": "!1abcead5",
+  "title": "[来自 !72d17be4] 你好",
+  "created_at": "2026-06-18T00:30:30.5200785+08:00",
+  "updated_at": "2026-06-18T00:30:34.3123551+08:00",
+  "messages": [
+    {
+      "role": "user",
+      "content": "[来自 !72d17be4] 你好"
+    },
+    {
+      "role": "assistant",
+      "content": "你好！很高兴见到你！😊\n\n我是DeepSeek，由深度求索公司创造的AI助手。无论你有什么问题、需要帮助，还是只是想聊聊天，我都很乐意陪你！\n\n有什\ufffd\ufffd..."
+    }
+  ]
+}

win/var/lib/mesh_mqtt_go/conversations/bot_1_1781713890525027000.json

@@ -0,0 +1,18 @@
+{
+  "id": "bot_1_1781713890525027000",
+  "bot_id": 1,
+  "bot_node_id": "!1abcead5",
+  "title": "[来自 !72d17be4] 你好",
+  "created_at": "2026-06-18T00:31:30.525027+08:00",
+  "updated_at": "2026-06-18T00:31:33.1653132+08:00",
+  "messages": [
+    {
+      "role": "user",
+      "content": "[来自 !72d17be4] 你好"
+    },
+    {
+      "role": "assistant",
+      "content": "你好！😊 有什么可以帮你的吗？"
+    }
+  ]
+}

win/var/lib/mesh_mqtt_go/conversations/bot_1_1781713920523402600.json

@@ -0,0 +1,18 @@
+{
+  "id": "bot_1_1781713920523402600",
+  "bot_id": 1,
+  "bot_node_id": "!1abcead5",
+  "title": "[来自 !72d17be4] 你好",
+  "created_at": "2026-06-18T00:32:00.5234026+08:00",
+  "updated_at": "2026-06-18T00:32:04.2936935+08:00",
+  "messages": [
+    {
+      "role": "user",
+      "content": "[来自 !72d17be4] 你好"
+    },
+    {
+      "role": "assistant",
+      "content": "你好！很高兴见到你！😊\n\n我是DeepSeek，一个由深度求索公司创造的AI助手。无论你有什么问题、需要帮助，还是只是想聊聊天，我都很乐意陪伴你、\ufffd..."
+    }
+  ]
+}

win/var/lib/mesh_mqtt_go/conversations/bot_1_1781714154187832400.json

@@ -0,0 +1,18 @@
+{
+  "id": "bot_1_1781714154187832400",
+  "bot_id": 1,
+  "bot_node_id": "!1abcead5",
+  "title": "[来自 !72d17be4] 你好",
+  "created_at": "2026-06-18T00:35:54.1878324+08:00",
+  "updated_at": "2026-06-18T00:35:58.3371209+08:00",
+  "messages": [
+    {
+      "role": "user",
+      "content": "[来自 !72d17be4] 你好"
+    },
+    {
+      "role": "assistant",
+      "content": "你好！很高兴见到你！😊\n\n我是DeepSeek，由深度求索公司创造的AI助手。无论你有什么问题、需要帮助，还是只是想聊聊天，我都很乐意陪你一起探索\ufffd..."
+    }
+  ]
+}

win/var/lib/mesh_mqtt_go/conversations/bot_1_1781714169190665700.json

@@ -0,0 +1,18 @@
+{
+  "id": "bot_1_1781714169190665700",
+  "bot_id": 1,
+  "bot_node_id": "!1abcead5",
+  "title": "[来自 !72d17be4] 你好",
+  "created_at": "2026-06-18T00:36:09.1906657+08:00",
+  "updated_at": "2026-06-18T00:36:13.5432077+08:00",
+  "messages": [
+    {
+      "role": "user",
+      "content": "[来自 !72d17be4] 你好"
+    },
+    {
+      "role": "assistant",
+      "content": "你好！很高兴见到你。😊\n\n我是DeepSeek，由深度求索公司创造的AI助手。我可以帮你解答问题、处理文档、进行创作、分析数据……几乎任何文字相关\ufffd..."
+    }
+  ]
+}

win/var/lib/mesh_mqtt_go/conversations/bot_1_1781714179192265800.json

@@ -0,0 +1,18 @@
+{
+  "id": "bot_1_1781714179192265800",
+  "bot_id": 1,
+  "bot_node_id": "!1abcead5",
+  "title": "[来自 !72d17be4] 111",
+  "created_at": "2026-06-18T00:36:19.1922658+08:00",
+  "updated_at": "2026-06-18T00:36:23.4404859+08:00",
+  "messages": [
+    {
+      "role": "user",
+      "content": "[来自 !72d17be4] 111"
+    },
+    {
+      "role": "assistant",
+      "content": "你好！你发送了“111”，请问有什么需要帮助的吗？如果有具体问题或指令，请告诉我。"
+    }
+  ]
+}

win/var/lib/mesh_mqtt_go/conversations/bot_1_1781714224192405100.json

@@ -0,0 +1,42 @@
+{
+  "id": "bot_1_1781714224192405100",
+  "bot_id": 1,
+  "bot_node_id": "!1abcead5",
+  "title": "[来自 !72d17be4] 你好",
+  "created_at": "2026-06-18T00:37:04.1924051+08:00",
+  "updated_at": "2026-06-18T00:47:31.9941255+08:00",
+  "messages": [
+    {
+      "role": "user",
+      "content": "[来自 !72d17be4] 你好"
+    },
+    {
+      "role": "assistant",
+      "content": "你好！很高兴见到你！😊\n\n我收到了你的消息，不过看起来前面那个“!72d17be4”像是一个标识符或者系统生成的代码。如果你有什么具体问题或者需��..."
+    },
+    {
+      "role": "user",
+      "content": "[来自 !72d17be4] 你好"
+    },
+    {
+      "role": "assistant",
+      "content": "你好！再次见到你很高兴！😊\n\n我看到你又一次带着“!72d17be4”这个标识出现了。这个标识似乎一直在跟随你的消息。如果你需要我帮你处理、解析��..."
+    },
+    {
+      "role": "user",
+      "content": "[来自 !72d17be4] 你好"
+    },
+    {
+      "role": "assistant",
+      "content": "你好！😊\n\n我看到你第三次发来了相同的消息，并且每次前面都带着 **`!72d17be4`** 这个标识。\n\n如果你是在**测试**或者**重复发送**，我理解这可能是��..."
+    },
+    {
+      "role": "user",
+      "content": "[来自 !72d17be4] 你好"
+    },
+    {
+      "role": "assistant",
+      "content": "你好！  我注意到你连续四次发送了相同格式的消息，前面都带有 `[来自 !72d17be4]` 标识。这可能是系统自动发送测试或者你希望引起我的注意。  无论..."
+    }
+  ]
+}