aichat/main.go

package main

import (
	"bufio"
	"bytes"
	"context"
	"crypto/rand"
	"encoding/base64"
	"encoding/hex"
	"encoding/json"
	"errors"
	"fmt"
	"io"
	"net"
	"net/http"
	"net/url"
	"os"
	"path/filepath"
	"sort"
	"strings"
	"sync"
	"time"
	"unicode"

	searchagent "aichat/agents/search"
	sqlquery "aichat/agents/sql"
	timeagent "aichat/agents/time"

	"github.com/gin-gonic/gin"
	ark "github.com/volcengine/volcengine-go-sdk/service/arkruntime"
	"github.com/volcengine/volcengine-go-sdk/service/arkruntime/model"
	"gopkg.in/yaml.v3"
)

// ─── 配置 ─────────────────────────────────────────────────

const (
	defaultOpenAIBaseURL        = "https://ark.cn-beijing.volces.com/api/v3"
	defaultOpenAITimeout        = 120
	defaultToolRouterTimeout    = 30
	defaultToolRouterMaxTokens  = 512
	defaultToolRouterSystemText = `你可以按需直接调用可用工具来回答用户问题。
如果用户问题包含今天、今日、明天、昨天、本周、本月、本年、最近等相对时间，且后续需要搜索或查询数据库，应先调用 time 获取绝对日期范围。
需要实时网页资料、新闻、当前版本、近期事件、网页核验或用户明确要求联网时，调用 search。
需要查询本地业务数据、日程、会议、待办、记录、统计或时间范围内数据时，调用 sql。
工具结果优先于模型内置知识；工具失败时必须如实说明，不要编造结果。
只调用确实必要的工具。`
)

type OpenAIConfig struct {
	Name           string `yaml:"name" json:"name"`
	Active         bool   `yaml:"active,omitempty" json:"active"`
	APIKey         string `yaml:"api_key" json:"-"`
	BaseURL        string `yaml:"base_url" json:"base_url"`
	Model          string `yaml:"model" json:"model"`
	Timeout        int    `yaml:"timeout" json:"timeout"`
	ParseThinkTags *bool  `yaml:"parse_think_tags,omitempty" json:"parse_think_tags,omitempty"`
}

type OpenAIConfigs []OpenAIConfig

type ToolRouterConfig struct {
	Enabled      bool              `yaml:"enabled" json:"enabled"`
	OpenAIName   string            `yaml:"openai_name" json:"openai_name"`
	Timeout      int               `yaml:"timeout" json:"timeout"`
	MaxTokens    int               `yaml:"max_tokens" json:"max_tokens"`
	SystemPrompt string            `yaml:"system_prompt" json:"system_prompt"`
	Tools        []ToolRouteConfig `yaml:"tools" json:"tools"`
}

type ToolRouteConfig struct {
	Name        string `yaml:"name" json:"name"`
	Enabled     bool   `yaml:"enabled" json:"enabled"`
	Description string `yaml:"description" json:"description"`
}

func (configs *OpenAIConfigs) UnmarshalYAML(value *yaml.Node) error {
	switch value.Kind {
	case yaml.SequenceNode:
		var items []OpenAIConfig
		if err := value.Decode(&items); err != nil {
			return err
		}
		*configs = items
	case yaml.MappingNode:
		var item OpenAIConfig
		if err := value.Decode(&item); err != nil {
			return err
		}
		*configs = []OpenAIConfig{item}
	case yaml.ScalarNode:
		if value.Tag == "!!null" {
			*configs = nil
			return nil
		}
		return fmt.Errorf("openai 配置格式无效")
	default:
		return fmt.Errorf("openai 配置格式无效")
	}
	return nil
}

type Config struct {
	Server struct {
		Mode    string `yaml:"mode"`
		Address string `yaml:"address"`
	} `yaml:"server"`
	OpenAI     OpenAIConfigs    `yaml:"openai"`
	ToolRouter ToolRouterConfig `yaml:"tool_router"`
}

func defaultOpenAIConfig() OpenAIConfig {
	return OpenAIConfig{
		Name:    "default",
		Active:  true,
		BaseURL: defaultOpenAIBaseURL,
		Timeout: defaultOpenAITimeout,
	}
}

func defaultToolRouterConfig() ToolRouterConfig {
	return ToolRouterConfig{
		Enabled:      true,
		OpenAIName:   "",
		Timeout:      defaultToolRouterTimeout,
		MaxTokens:    defaultToolRouterMaxTokens,
		SystemPrompt: defaultToolRouterSystemText,
		Tools: []ToolRouteConfig{
			{Name: "time", Enabled: true, Description: ""},
			{Name: "search", Enabled: true, Description: ""},
			{Name: "sql", Enabled: true, Description: ""},
		},
	}
}

func defaultConfig() Config {
	var cfg Config
	cfg.Server.Mode = "tcp"
	cfg.Server.Address = "0.0.0.0:8080"
	cfg.OpenAI = OpenAIConfigs{defaultOpenAIConfig()}
	cfg.ToolRouter = defaultToolRouterConfig()
	return cfg
}

func loadConfig(path string) (*Config, error) {
	if err := ensureConfigFile(path); err != nil {
		return nil, err
	}

	data, err := os.ReadFile(path)
	if err != nil {
		return nil, fmt.Errorf("读取配置文件失败: %w", err)
	}
	var cfg Config
	if err = yaml.Unmarshal(data, &cfg); err != nil {
		return nil, fmt.Errorf("解析配置文件失败: %w", err)
	}
	if _, err := normalizeOpenAIConfigs(&cfg); err != nil {
		return nil, err
	}
	// 环境变量优先
	if key := os.Getenv("ARK_API_KEY"); key != "" {
		for i := range cfg.OpenAI {
			cfg.OpenAI[i].APIKey = key
		}
	}
	legacySearchProfiles = readLegacySearchProfiles(data)
	if _, err := normalizeToolRouterConfig(&cfg); err != nil {
		return nil, err
	}
	return &cfg, nil
}

func ensureConfigFile(path string) error {
	defaults := defaultConfig()
	if _, err := os.Stat(path); err != nil {
		if !os.IsNotExist(err) {
			return fmt.Errorf("检查配置文件失败: %w", err)
		}
		return writeConfig(path, defaults)
	}

	data, err := os.ReadFile(path)
	if err != nil {
		return fmt.Errorf("读取配置文件失败: %w", err)
	}
	var cfg Config
	if err = yaml.Unmarshal(data, &cfg); err != nil {
		return fmt.Errorf("解析配置文件失败: %w", err)
	}
	var raw map[string]any
	if err = yaml.Unmarshal(data, &raw); err != nil {
		return fmt.Errorf("解析配置文件失败: %w", err)
	}

	changed := false
	server, _ := raw["server"].(map[string]any)
	if server == nil {
		cfg.Server = defaults.Server
		changed = true
	} else {
		if _, ok := server["mode"]; !ok {
			cfg.Server.Mode = defaults.Server.Mode
			changed = true
		}
		if _, ok := server["address"]; !ok {
			cfg.Server.Address = defaults.Server.Address
			changed = true
		}
	}

	if _, ok := raw["openai"].([]any); !ok {
		changed = true
	}
	if normalized, err := normalizeOpenAIConfigs(&cfg); err != nil {
		return err
	} else if normalized {
		changed = true
	}

	if _, ok := raw["tool_router"]; !ok {
		cfg.ToolRouter = defaults.ToolRouter
		changed = true
	} else if normalized, err := normalizeToolRouterConfig(&cfg); err != nil {
		return err
	} else if normalized {
		changed = true
	}

	if !changed {
		return nil
	}
	return writeConfig(path, cfg)
}

func normalizeOpenAIConfigs(cfg *Config) (bool, error) {
	changed := false
	if len(cfg.OpenAI) == 0 {
		cfg.OpenAI = OpenAIConfigs{defaultOpenAIConfig()}
		changed = true
	}

	activeIndex := -1
	seen := map[string]bool{}
	for i := range cfg.OpenAI {
		profile := &cfg.OpenAI[i]
		name := strings.TrimSpace(profile.Name)
		if name == "" {
			name = strings.TrimSpace(profile.Model)
			if name == "" {
				name = fmt.Sprintf("openai-%d", i+1)
			}
			profile.Name = name
			changed = true
		} else if name != profile.Name {
			profile.Name = name
			changed = true
		}
		if seen[name] {
			return changed, fmt.Errorf("openai 配置名称重复: %s", name)
		}
		seen[name] = true

		if strings.TrimSpace(profile.BaseURL) == "" {
			profile.BaseURL = defaultOpenAIBaseURL
			changed = true
		}
		if profile.Timeout <= 0 {
			profile.Timeout = defaultOpenAITimeout
			changed = true
		}
		if profile.Active {
			if activeIndex == -1 {
				activeIndex = i
			} else {
				profile.Active = false
				changed = true
			}
		}
	}
	if activeIndex == -1 {
		cfg.OpenAI[0].Active = true
		changed = true
	}
	return changed, nil
}

func isLegacyToolRouterPrompt(prompt string) bool {
	prompt = strings.TrimSpace(prompt)
	return strings.Contains(prompt, "工具路由器") || strings.Contains(prompt, "route_tools") || strings.Contains(prompt, `"tools":[`)
}

func normalizeToolRouterConfig(cfg *Config) (bool, error) {
	changed := false
	defaults := defaultToolRouterConfig()
	cfg.ToolRouter.OpenAIName = strings.TrimSpace(cfg.ToolRouter.OpenAIName)
	if cfg.ToolRouter.Timeout <= 0 {
		cfg.ToolRouter.Timeout = defaultToolRouterTimeout
		changed = true
	}
	if cfg.ToolRouter.MaxTokens <= 0 {
		cfg.ToolRouter.MaxTokens = defaultToolRouterMaxTokens
		changed = true
	}
	systemPrompt := strings.TrimSpace(cfg.ToolRouter.SystemPrompt)
	if systemPrompt == "" || isLegacyToolRouterPrompt(systemPrompt) {
		cfg.ToolRouter.SystemPrompt = defaultToolRouterSystemText
		changed = true
	} else if systemPrompt != cfg.ToolRouter.SystemPrompt {
		cfg.ToolRouter.SystemPrompt = systemPrompt
		changed = true
	}
	if len(cfg.ToolRouter.Tools) == 0 {
		cfg.ToolRouter.Tools = defaults.Tools
		changed = true
	}
	seen := map[string]bool{}
	for i := range cfg.ToolRouter.Tools {
		tool := &cfg.ToolRouter.Tools[i]
		name := strings.ToLower(strings.TrimSpace(tool.Name))
		if name == "" {
			name = fmt.Sprintf("tool-%d", i+1)
		}
		if name != tool.Name {
			tool.Name = name
			changed = true
		}
		tool.Description = strings.TrimSpace(tool.Description)
		if seen[name] {
			return changed, fmt.Errorf("tool_router.tools 配置名称重复: %s", name)
		}
		seen[name] = true
	}
	byName := map[string]ToolRouteConfig{}
	for _, tool := range cfg.ToolRouter.Tools {
		byName[tool.Name] = tool
	}
	merged := make([]ToolRouteConfig, 0, len(cfg.ToolRouter.Tools)+len(defaults.Tools))
	used := map[string]bool{}
	for _, tool := range defaults.Tools {
		if existing, ok := byName[tool.Name]; ok {
			merged = append(merged, existing)
		} else {
			merged = append(merged, tool)
			changed = true
		}
		used[tool.Name] = true
	}
	for _, tool := range cfg.ToolRouter.Tools {
		if !used[tool.Name] {
			merged = append(merged, tool)
		}
	}
	if len(merged) != len(cfg.ToolRouter.Tools) {
		changed = true
	} else {
		for i := range merged {
			if merged[i].Name != cfg.ToolRouter.Tools[i].Name {
				changed = true
				break
			}
		}
	}
	cfg.ToolRouter.Tools = merged
	return changed, nil
}

func readLegacySearchProfiles(data []byte) []searchagent.ProfileConfig {
	var legacy struct {
		Search searchagent.ProfileConfigs `yaml:"search"`
	}
	if err := yaml.Unmarshal(data, &legacy); err != nil {
		return nil
	}
	return []searchagent.ProfileConfig(legacy.Search)
}

func writeConfig(path string, cfg Config) error {
	data, err := yaml.Marshal(&cfg)
	if err != nil {
		return fmt.Errorf("生成配置文件失败: %w", err)
	}
	if err := os.WriteFile(path, data, 0644); err != nil {
		return fmt.Errorf("写入配置文件失败: %w", err)
	}
	return nil
}

// ─── 请求结构 ─────────────────────────────────────────────

type ChatMessage struct {
	Role          string `json:"role"`
	Content       string `json:"content"`
	ImageURL      string `json:"image_url,omitempty"` // base64 data URI 或 http URL
	ImageURLAlias string `json:"imageURL,omitempty"`
	Hidden        bool   `json:"hidden,omitempty"`
}

type ChatRequest struct {
	ConversationID string        `json:"conversation_id,omitempty"`
	Messages       []ChatMessage `json:"messages"`
	WebSearch      bool          `json:"web_search,omitempty"`
	OpenAIName     string        `json:"openai_name,omitempty"`
}

type Conversation struct {
	ID        string        `json:"id"`
	Title     string        `json:"title"`
	CreatedAt time.Time     `json:"created_at"`
	UpdatedAt time.Time     `json:"updated_at"`
	Messages  []ChatMessage `json:"messages,omitempty"`
}

type ConvStore struct {
	dir string
	mu  sync.Mutex
}

type OpenAIProfile struct {
	Config OpenAIConfig
	Client *ark.Client
}

type OpenAIState struct {
	mu         sync.RWMutex
	profiles   map[string]*OpenAIProfile
	order      []string
	activeName string
}

type activeProfileRequest struct {
	Name string `json:"name"`
}

type openAIListResponse struct {
	Active   string         `json:"active"`
	Profiles []OpenAIConfig `json:"profiles"`
}

type chatCompleter func(context.Context, *OpenAIProfile, model.CreateChatCompletionRequest, time.Duration) (model.ChatCompletionResponse, error)

type ToolRouterState struct {
	cfg      *ToolRouterConfig
	ai       *OpenAIState
	complete chatCompleter
}

func NewToolRouterState(config *ToolRouterConfig, ai *OpenAIState) (*ToolRouterState, error) {
	if config == nil {
		cfg := defaultToolRouterConfig()
		config = &cfg
	}
	if ai == nil {
		return nil, errors.New("工具路由需要 OpenAI 状态")
	}
	if config.Enabled && strings.TrimSpace(config.OpenAIName) != "" {
		if _, err := ai.GetProfile(config.OpenAIName); err != nil {
			return nil, fmt.Errorf("tool_router.openai_name 配置无效: %w", err)
		}
	}
	return &ToolRouterState{cfg: config, ai: ai, complete: completeChatWithTimeout}, nil
}

func NewOpenAIState(configs []OpenAIConfig) (*OpenAIState, error) {
	state := &OpenAIState{
		profiles: make(map[string]*OpenAIProfile, len(configs)),
		order:    make([]string, 0, len(configs)),
	}
	for _, config := range configs {
		if strings.TrimSpace(config.Name) == "" {
			return nil, errors.New("openai.name 不能为空")
		}
		if strings.TrimSpace(config.APIKey) == "" {
			return nil, fmt.Errorf("openai.%s.api_key 未配置，也未设置环境变量 ARK_API_KEY", config.Name)
		}
		if strings.TrimSpace(config.Model) == "" {
			return nil, fmt.Errorf("openai.%s.model 未配置", config.Name)
		}
		if strings.TrimSpace(config.BaseURL) == "" {
			return nil, fmt.Errorf("openai.%s.base_url 未配置", config.Name)
		}
		if config.Timeout <= 0 {
			return nil, fmt.Errorf("openai.%s.timeout 必须大于 0", config.Name)
		}
		if _, ok := state.profiles[config.Name]; ok {
			return nil, fmt.Errorf("openai 配置名称重复: %s", config.Name)
		}
		state.profiles[config.Name] = &OpenAIProfile{
			Config: config,
			Client: ark.NewClientWithApiKey(
				config.APIKey,
				ark.WithBaseUrl(config.BaseURL),
				ark.WithTimeout(time.Duration(config.Timeout)*time.Second),
			),
		}
		state.order = append(state.order, config.Name)
		if config.Active && state.activeName == "" {
			state.activeName = config.Name
		}
	}
	if len(state.order) == 0 {
		return nil, errors.New("openai 配置不能为空")
	}
	if state.activeName == "" {
		state.activeName = state.order[0]
	}
	return state, nil
}

func (s *OpenAIState) ActiveProfile() *OpenAIProfile {
	s.mu.RLock()
	defer s.mu.RUnlock()
	return s.profiles[s.activeName]
}

func (s *OpenAIState) GetProfile(name string) (*OpenAIProfile, error) {
	s.mu.RLock()
	defer s.mu.RUnlock()
	if strings.TrimSpace(name) == "" {
		return s.profiles[s.activeName], nil
	}
	profile, ok := s.profiles[strings.TrimSpace(name)]
	if !ok {
		return nil, fmt.Errorf("OpenAI 配置不存在: %s", name)
	}
	return profile, nil
}

func (s *OpenAIState) SwitchActive(name string) (*OpenAIProfile, error) {
	name = strings.TrimSpace(name)
	if name == "" {
		return nil, errors.New("OpenAI 配置名称不能为空")
	}
	s.mu.Lock()
	defer s.mu.Unlock()
	profile, ok := s.profiles[name]
	if !ok {
		return nil, fmt.Errorf("OpenAI 配置不存在: %s", name)
	}
	s.activeName = name
	return profile, nil
}

func (s *OpenAIState) ListProfiles() openAIListResponse {
	s.mu.RLock()
	defer s.mu.RUnlock()
	profiles := make([]OpenAIConfig, 0, len(s.order))
	for _, name := range s.order {
		profile := s.profiles[name]
		config := profile.Config
		config.APIKey = ""
		config.Active = name == s.activeName
		profiles = append(profiles, config)
	}
	return openAIListResponse{Active: s.activeName, Profiles: profiles}
}

func publicOpenAIConfig(profile *OpenAIProfile, active bool) OpenAIConfig {
	config := profile.Config
	config.APIKey = ""
	config.Active = active
	return config
}

func (s *ToolRouterState) RouterProfile(fallback *OpenAIProfile) *OpenAIProfile {
	if s == nil || s.cfg == nil || s.ai == nil {
		return fallback
	}
	name := strings.TrimSpace(s.cfg.OpenAIName)
	if name == "" {
		return fallback
	}
	profile, err := s.ai.GetProfile(name)
	if err != nil {
		return fallback
	}
	return profile
}

func isOllamaProfile(profile *OpenAIProfile) bool {
	if profile == nil {
		return false
	}
	u, err := url.Parse(strings.TrimSpace(profile.Config.BaseURL))
	if err != nil {
		return strings.Contains(profile.Config.BaseURL, ":11434")
	}
	host := strings.ToLower(u.Hostname())
	port := u.Port()
	return port == "11434" && (host == "127.0.0.1" || host == "localhost" || host == "::1")
}

func shouldParseThinkTags(profile *OpenAIProfile) bool {
	if profile == nil {
		return false
	}
	if profile.Config.ParseThinkTags != nil {
		return *profile.Config.ParseThinkTags
	}
	return isOllamaProfile(profile)
}

// ─── 全局变量 ─────────────────────────────────────────────

var (
	cfg                  *Config
	aiState              *OpenAIState
	searchState          *searchagent.State
	legacySearchProfiles []searchagent.ProfileConfig
	toolRouterState      *ToolRouterState
	sqlState             *sqlquery.State
	store                *ConvStore
)

type chatSSEFrame struct {
	Type    string           `json:"type"`
	Text    string           `json:"text,omitempty"`
	Message string           `json:"message,omitempty"`
	Tool    string           `json:"tool,omitempty"`
	Stage   string           `json:"stage,omitempty"`
	Status  string           `json:"status,omitempty"`
	Data    map[string]any   `json:"data,omitempty"`
	Stats   *tokenUsageStats `json:"stats,omitempty"`
	Error   string           `json:"error,omitempty"`
}

type tokenUsageStats struct {
	PromptTokens               int     `json:"prompt_tokens"`
	CompletionTokens           int     `json:"completion_tokens"`
	ToolPromptTokens           int     `json:"tool_prompt_tokens"`
	ToolCompletionTokens       int     `json:"tool_completion_tokens"`
	TotalTokens                int     `json:"total_tokens"`
	CompletionTokensPerSec     float64 `json:"completion_tokens_per_sec"`
	PeakCompletionTokensPerSec float64 `json:"peak_completion_tokens_per_sec"`
	Estimated                  bool    `json:"estimated"`
}

type tokenUsageTracker struct {
	mu                   sync.Mutex
	promptTokens         int
	completionTokens     int
	toolPromptTokens     int
	toolCompletionTokens int
}

type tokenUsageContextKey struct{}

func newTokenUsageTracker() *tokenUsageTracker {
	return &tokenUsageTracker{}
}

func contextWithTokenUsage(ctx context.Context, tracker *tokenUsageTracker) context.Context {
	if tracker == nil {
		return ctx
	}
	return context.WithValue(ctx, tokenUsageContextKey{}, tracker)
}

func tokenUsageFromContext(ctx context.Context) *tokenUsageTracker {
	tracker, _ := ctx.Value(tokenUsageContextKey{}).(*tokenUsageTracker)
	return tracker
}

func (t *tokenUsageTracker) addTool(promptTokens, completionTokens int) {
	if t == nil {
		return
	}
	t.mu.Lock()
	defer t.mu.Unlock()
	t.toolPromptTokens += promptTokens
	t.toolCompletionTokens += completionTokens
}

func (t *tokenUsageTracker) setModel(promptTokens, completionTokens int) {
	if t == nil {
		return
	}
	t.mu.Lock()
	defer t.mu.Unlock()
	t.promptTokens = promptTokens
	t.completionTokens = completionTokens
}

func (t *tokenUsageTracker) snapshot(tokensPerSecond, peakTokensPerSecond float64) tokenUsageStats {
	if t == nil {
		return tokenUsageStats{Estimated: true}
	}
	t.mu.Lock()
	defer t.mu.Unlock()
	total := t.promptTokens + t.completionTokens + t.toolPromptTokens + t.toolCompletionTokens
	return tokenUsageStats{
		PromptTokens:               t.promptTokens,
		CompletionTokens:           t.completionTokens,
		ToolPromptTokens:           t.toolPromptTokens,
		ToolCompletionTokens:       t.toolCompletionTokens,
		TotalTokens:                total,
		CompletionTokensPerSec:     tokensPerSecond,
		PeakCompletionTokensPerSec: peakTokensPerSecond,
		Estimated:                  true,
	}
}

// ─── 路由 ─────────────────────────────────────────────────

func indexHandler(c *gin.Context) {
	profile := aiState.ActiveProfile()
	c.HTML(http.StatusOK, "chat.html", gin.H{
		"Title":      "AI 对话",
		"Model":      profile.Config.Model,
		"OpenAIName": profile.Config.Name,
	})
}

func listOpenAIHandler(c *gin.Context) {
	c.JSON(http.StatusOK, aiState.ListProfiles())
}

func switchOpenAIHandler(c *gin.Context) {
	var req activeProfileRequest
	if err := c.ShouldBindJSON(&req); err != nil {
		c.JSON(http.StatusBadRequest, gin.H{"error": "请求格式错误: " + err.Error()})
		return
	}
	profile, err := aiState.SwitchActive(req.Name)
	if err != nil {
		c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}
	c.JSON(http.StatusOK, gin.H{
		"active":  profile.Config.Name,
		"profile": publicOpenAIConfig(profile, true),
	})
}

func listSearchHandler(c *gin.Context) {
	c.JSON(http.StatusOK, searchState.ListProfiles())
}

func switchSearchHandler(c *gin.Context) {
	var req activeProfileRequest
	if err := c.ShouldBindJSON(&req); err != nil {
		c.JSON(http.StatusBadRequest, gin.H{"error": "请求格式错误: " + err.Error()})
		return
	}
	profile, err := searchState.SwitchActive(req.Name)
	if err != nil {
		c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}
	profile.APIKey = ""
	profile.Active = true
	c.JSON(http.StatusOK, gin.H{
		"active":  profile.Name,
		"profile": profile,
	})
}

func listConversationsHandler(c *gin.Context) {
	convs, err := store.List()
	if err != nil {
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}
	c.JSON(http.StatusOK, convs)
}

func createConversationHandler(c *gin.Context) {
	conv, err := store.Create()
	if err != nil {
		c.JSON(http.StatusInternalServerError, gin.H{"error": "创建对话失败: " + err.Error()})
		return
	}
	c.JSON(http.StatusOK, conv)
}

func getConversationHandler(c *gin.Context) {
	conv, err := store.Get(c.Param("id"))
	if err != nil {
		status := http.StatusInternalServerError
		if err.Error() == "对话不存在" {
			status = http.StatusNotFound
		}
		c.JSON(status, gin.H{"error": err.Error()})
		return
	}
	c.JSON(http.StatusOK, conv)
}

func deleteConversationHandler(c *gin.Context) {
	if err := store.Delete(c.Param("id")); err != nil {
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}
	c.Status(http.StatusNoContent)
}

// chatHandler 流式 SSE 对话接口
func chatHandler(c *gin.Context) {
	var req ChatRequest
	if err := c.ShouldBindJSON(&req); err != nil {
		c.JSON(http.StatusBadRequest, gin.H{"error": "请求格式错误: " + err.Error()})
		return
	}
	if len(req.Messages) == 0 {
		c.JSON(http.StatusBadRequest, gin.H{"error": "消息不能为空"})
		return
	}
	profile, err := aiState.GetProfile(req.OpenAIName)
	if err != nil {
		c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}

	// SSE 头先写出，后续插件/模型过程都通过 trace 事件实时展示。
	c.Writer.Header().Set("Content-Type", "text/event-stream")
	c.Writer.Header().Set("Cache-Control", "no-cache")
	c.Writer.Header().Set("Connection", "keep-alive")
	c.Writer.Header().Set("X-Accel-Buffering", "no")
	c.Writer.WriteHeader(http.StatusOK)
	flusher, ok := c.Writer.(http.Flusher)
	if !ok {
		return
	}
	emit := func(frame chatSSEFrame) {
		writeSSEJSON(c.Writer, frame)
		flusher.Flush()
	}
	emitTrace := func(tool, stage, status, message string, data map[string]any) {
		emit(chatSSEFrame{Type: "trace", Tool: tool, Stage: stage, Status: status, Message: message, Data: data})
	}
	emitError := func(err error) {
		emit(chatSSEFrame{Type: "error", Error: err.Error()})
	}

	// 超时 context
	timeout := time.Duration(profile.Config.Timeout) * time.Second
	ctx, cancel := context.WithTimeout(c.Request.Context(), timeout)
	defer cancel()
	usage := newTokenUsageTracker()
	ctx = contextWithTokenUsage(ctx, usage)

	// 用 Function Calling 工具循环替代旧的路由+隐藏上下文机制
	messages, err := runAgentToolLoop(ctx, profile, req.Messages, emit)
	if err != nil {
		fmt.Fprintln(os.Stderr, "Agent 工具循环失败:", err)
		messages, err = buildArkMessages(req.Messages)
		if err != nil {
			emitError(err)
			return
		}
	}
	promptTokens := estimateChatMessagesTokens(req.Messages)

	if isOllamaProfile(profile) && hasImageMessage(req.Messages) {
		emitTrace("model", "request", "running", "正在通过 Ollama 原生接口调用视觉模型", nil)
		err = streamOllamaChat(ctx, profile, messages, promptTokens, usage, emit, func(content string) {
			if req.ConversationID != "" {
				if err := saveConversationMessages(req.ConversationID, req.Messages, content); err != nil {
					fmt.Fprintln(os.Stderr, "保存对话失败:", err)
				}
			}
		})
		if err != nil {
			emitError(err)
		}
		return
	}

	emitTrace("model", "request", "running", "正在调用模型生成回答", nil)
	stream, err := profile.Client.CreateChatCompletionStream(ctx, model.CreateChatCompletionRequest{
		Model:     profile.Config.Model,
		Messages:  messages,
		MaxTokens: intPtr(4096),
	}.WithStream(true))
	if err != nil {
		emitError(err)
		return
	}
	defer stream.Close()
	emitTrace("model", "stream", "running", "模型已开始输出", nil)

	var full strings.Builder
	completionTokens := 0
	streamStarted := time.Now()
	windowStarted := streamStarted
	windowTokens := 0
	peakTokensPerSecond := 0.0
	parseThinkTags := shouldParseThinkTags(profile)
	thinkParser := &thinkTagParser{}
	emitDelta := func(delta string) {
		if delta == "" {
			return
		}
		now := time.Now()
		deltaTokens := estimateTokenCount(delta)
		windowTokens += deltaTokens
		windowElapsed := now.Sub(windowStarted).Seconds()
		if windowElapsed >= 1 {
			windowSpeed := float64(windowTokens) / windowElapsed
			if windowSpeed > peakTokensPerSecond {
				peakTokensPerSecond = windowSpeed
			}
			windowStarted = now
			windowTokens = 0
		} else if peakTokensPerSecond == 0 && windowElapsed > 0.25 {
			peakTokensPerSecond = float64(windowTokens) / windowElapsed
		}
		full.WriteString(delta)
		completionTokens += deltaTokens
		usage.setModel(promptTokens, completionTokens)
		stats := usage.snapshot(tokensPerSecond(completionTokens, streamStarted), peakTokensPerSecond)
		emit(chatSSEFrame{Type: "delta", Text: delta, Stats: &stats})
	}
	emitModelContent := func(delta string) {
		if delta == "" {
			return
		}
		if !parseThinkTags {
			emitDelta(delta)
			return
		}
		visible, reasoning := thinkParser.Accept(delta)
		if reasoning != "" {
			emit(chatSSEFrame{Type: "reasoning", Text: reasoning})
		}
		emitDelta(visible)
	}
	for {
		resp, err := stream.Recv()
		if errors.Is(err, io.EOF) {
			if parseThinkTags {
				visible, reasoning := thinkParser.Flush()
				if reasoning != "" {
					emit(chatSSEFrame{Type: "reasoning", Text: reasoning})
				}
				emitDelta(visible)
			}
			usage.setModel(promptTokens, completionTokens)
			if windowTokens > 0 {
				windowElapsed := time.Since(windowStarted).Seconds()
				if windowElapsed > 0.25 {
					windowSpeed := float64(windowTokens) / windowElapsed
					if windowSpeed > peakTokensPerSecond {
						peakTokensPerSecond = windowSpeed
					}
				}
			}
			if peakTokensPerSecond == 0 {
				peakTokensPerSecond = tokensPerSecond(completionTokens, streamStarted)
			}
			if req.ConversationID != "" {
				if err := saveConversationMessages(req.ConversationID, req.Messages, full.String()); err != nil {
					fmt.Fprintln(os.Stderr, "保存对话失败:", err)
				}
			}
			finalStats := usage.snapshot(tokensPerSecond(completionTokens, streamStarted), peakTokensPerSecond)
			emit(chatSSEFrame{Type: "stats", Stats: &finalStats})
			emitTrace("model", "stream", "success", "回答生成完成", nil)
			fmt.Fprintf(c.Writer, "data: [DONE]\n\n")
			flusher.Flush()
			return
		}
		if err != nil {
			emitError(err)
			return
		}
		if len(resp.Choices) > 0 {
			emitModelContent(resp.Choices[0].Delta.Content)
			// 思考过程 reasoning_content 单独事件推送
			if resp.Choices[0].Delta.ReasoningContent != nil && *resp.Choices[0].Delta.ReasoningContent != "" {
				emit(chatSSEFrame{Type: "reasoning", Text: *resp.Choices[0].Delta.ReasoningContent})
			}
		}
	}
}

// ─── 辅助函数 ─────────────────────────────────────────────

func estimateChatMessagesTokens(messages []ChatMessage) int {
	total := 0
	for _, msg := range messages {
		total += estimateTokenCount(msg.Role) + estimateTokenCount(msg.Content) + 4
		if msg.ImageURL != "" || msg.ImageURLAlias != "" {
			total += 85
		}
	}
	return total
}

func estimateTokenCount(text string) int {
	text = strings.TrimSpace(text)
	if text == "" {
		return 0
	}
	tokens := 0
	asciiRunes := 0
	flushASCII := func() {
		if asciiRunes > 0 {
			tokens += (asciiRunes + 3) / 4
			asciiRunes = 0
		}
	}
	for _, r := range text {
		if unicode.IsSpace(r) {
			flushASCII()
			continue
		}
		if r <= unicode.MaxASCII {
			asciiRunes++
			continue
		}
		flushASCII()
		tokens++
	}
	flushASCII()
	if tokens == 0 {
		return 1
	}
	return tokens
}

func tokensPerSecond(tokens int, start time.Time) float64 {
	elapsed := time.Since(start).Seconds()
	if tokens <= 0 || elapsed <= 0 {
		return 0
	}
	return float64(tokens) / elapsed
}

type agentTool struct {
	name       string
	definition *model.Tool
	execute    func(context.Context, string) (string, error)
}

func (t agentTool) Name() string { return t.name }

const maxAgentToolIterations = 6

func availableAgentTools(profile *OpenAIProfile, emit func(chatSSEFrame)) []agentTool {
	if toolRouterState == nil || toolRouterState.cfg == nil || !toolRouterState.cfg.Enabled {
		return nil
	}
	tools := make([]agentTool, 0, len(toolRouterState.cfg.Tools))
	for _, item := range toolRouterState.cfg.Tools {
		if !item.Enabled {
			continue
		}
		description := strings.TrimSpace(item.Description)
		switch item.Name {
		case timeagent.ToolName:
			tools = append(tools, agentTool{
				name:       timeagent.ToolName,
				definition: timeagent.ToolDefinition(description),
				execute: func(ctx context.Context, args string) (string, error) {
					result, err := timeagent.ExecuteTool(args, time.Now())
					if err == nil && emit != nil {
						emit(chatSSEFrame{Type: "trace", Tool: timeagent.ToolName, Stage: "resolve", Status: "success", Message: "已获取当前时间上下文"})
					}
					return result, err
				},
			})
		case searchagent.ToolName:
			if searchState == nil || !searchState.Enabled() {
				continue
			}
			tools = append(tools, agentTool{
				name:       searchagent.ToolName,
				definition: searchState.ToolDefinition(description),
				execute: func(ctx context.Context, args string) (string, error) {
					if emit != nil {
						emit(chatSSEFrame{Type: "trace", Tool: searchagent.ToolName, Stage: "request", Status: "running", Message: "正在联网搜索"})
					}
					result, err := searchState.ExecuteTool(ctx, args)
					if emit != nil {
						status := "success"
						message := "联网搜索完成"
						if err != nil {
							status = "error"
							message = "联网搜索失败"
						}
						emit(chatSSEFrame{Type: "trace", Tool: searchagent.ToolName, Stage: "results", Status: status, Message: message})
					}
					return result, err
				},
			})
		case sqlquery.ToolName:
			if sqlState == nil || !sqlState.Enabled() {
				continue
			}
			tools = append(tools, agentTool{
				name:       sqlquery.ToolName,
				definition: sqlState.ToolDefinition(description),
				execute: func(ctx context.Context, args string) (string, error) {
					if emit != nil {
						emit(chatSSEFrame{Type: "trace", Tool: sqlquery.ToolName, Stage: "execute", Status: "running", Message: "正在查询数据库"})
					}
					generator := func(ctx context.Context, prompt string, maxTokens int) (string, error) {
						return completeText(ctx, profile, []ChatMessage{{Role: "system", Content: prompt}}, maxTokens)
					}
					result, err := sqlState.ExecuteTool(ctx, args, generator)
					if emit != nil {
						status := "success"
						message := "数据库查询完成"
						if err != nil {
							status = "error"
							message = "数据库查询失败"
						}
						emit(chatSSEFrame{Type: "trace", Tool: sqlquery.ToolName, Stage: "execute", Status: status, Message: message})
					}
					return result, err
				},
			})
		}
	}
	return tools
}

func runAgentToolLoop(ctx context.Context, profile *OpenAIProfile, chatMessages []ChatMessage, emit func(chatSSEFrame)) ([]*model.ChatCompletionMessage, error) {
	finalMessages, err := buildArkMessages(chatMessages)
	if err != nil {
		return nil, err
	}
	routerProfile := profile
	if toolRouterState != nil {
		routerProfile = toolRouterState.RouterProfile(profile)
	}
	tools := availableAgentTools(routerProfile, emit)
	if len(tools) == 0 {
		return finalMessages, nil
	}
	decisionMessages := append([]*model.ChatCompletionMessage(nil), finalMessages...)
	if hasImageMessage(chatMessages) {
		decisionMessages, err = buildToolDecisionMessages(chatMessages)
		if err != nil {
			return nil, err
		}
		if emit != nil {
			emit(chatSSEFrame{Type: "trace", Tool: "agent_tools", Stage: "prepare", Status: "success", Message: "检测到图片输入，工具判断阶段将使用纯文本上下文"})
		}
	}
	toolByName := make(map[string]agentTool, len(tools))
	definitions := make([]*model.Tool, 0, len(tools))
	availableNames := make([]string, 0, len(tools))
	toolDescriptions := make([]string, 0, len(tools))
	for _, tool := range tools {
		toolByName[tool.name] = tool
		definitions = append(definitions, tool.definition)
		availableNames = append(availableNames, tool.name)
		if tool.definition != nil && tool.definition.Function != nil {
			toolDescriptions = append(toolDescriptions, fmt.Sprintf("%s: %s", tool.name, tool.definition.Function.Description))
		}
	}
	if emit != nil {
		emit(chatSSEFrame{Type: "trace", Tool: "agent_tools", Stage: "prepare", Status: "success", Message: "已准备可用工具", Data: map[string]any{"tools": availableNames, "tool_descriptions": toolDescriptions}})
	}
	if prompt := strings.TrimSpace(toolRouterState.cfg.SystemPrompt); prompt != "" {
		systemMessage := &model.ChatCompletionMessage{Role: model.ChatMessageRoleSystem, Content: stringContent(prompt)}
		finalMessages = append([]*model.ChatCompletionMessage{systemMessage}, finalMessages...)
		decisionMessages = append([]*model.ChatCompletionMessage{systemMessage}, decisionMessages...)
	}
	for i := 0; i < maxAgentToolIterations; i++ {
		if emit != nil {
			emit(chatSSEFrame{Type: "trace", Tool: "agent_tools", Stage: "request", Status: "running", Message: fmt.Sprintf("正在进行第 %d 轮工具判断", i+1), Data: map[string]any{"iteration": i + 1, "max_iterations": maxAgentToolIterations, "tools": availableNames}})
		}
		resp, err := toolRouterState.complete(ctx, routerProfile, model.CreateChatCompletionRequest{
			Model:             routerProfile.Config.Model,
			Messages:          decisionMessages,
			MaxTokens:         intPtr(toolRouterState.cfg.MaxTokens),
			Tools:             definitions,
			ToolChoice:        model.ToolChoiceStringTypeAuto,
			ParallelToolCalls: boolPtr(false),
		}, time.Duration(toolRouterState.cfg.Timeout)*time.Second)
		if err != nil {
			return finalMessages, err
		}
		if tracker := tokenUsageFromContext(ctx); tracker != nil {
			tracker.addTool(resp.Usage.PromptTokens, resp.Usage.CompletionTokens)
		}
		if len(resp.Choices) == 0 {
			return finalMessages, nil
		}
		choice := resp.Choices[0]
		decisionPreview := chatMessageContentString(choice.Message.Content)
		if emit != nil {
			emit(chatSSEFrame{Type: "trace", Tool: "agent_tools", Stage: "decision", Status: "success", Message: "工具判断响应已返回", Data: map[string]any{"iteration": i + 1, "finish_reason": string(choice.FinishReason), "content_preview": truncateString(decisionPreview, 800)}})
		}
		calls := choice.Message.ToolCalls
		if len(calls) == 0 && choice.Message.FunctionCall != nil {
			calls = []*model.ToolCall{{ID: "legacy_function_call", Type: model.ToolTypeFunction, Function: *choice.Message.FunctionCall}}
		}
		if len(calls) == 0 {
			if emit != nil {
				emit(chatSSEFrame{Type: "trace", Tool: "agent_tools", Stage: "request", Status: "success", Message: "模型未请求工具，进入回答生成"})
			}
			return finalMessages, nil
		}
		callNames := make([]string, 0, len(calls))
		for _, call := range calls {
			if call != nil {
				callNames = append(callNames, call.Function.Name)
			}
		}
		if emit != nil {
			emit(chatSSEFrame{Type: "trace", Tool: "agent_tools", Stage: "tool_calls", Status: "running", Message: fmt.Sprintf("模型请求调用 %d 个工具", len(calls)), Data: map[string]any{"tools": callNames, "iteration": i + 1}})
		}
		assistantMessage := &model.ChatCompletionMessage{Role: model.ChatMessageRoleAssistant, ToolCalls: calls, Content: choice.Message.Content}
		finalMessages = append(finalMessages, assistantMessage)
		decisionMessages = append(decisionMessages, assistantMessage)
		for _, call := range calls {
			result := executeAgentToolCall(ctx, call, toolByName, emit)
			toolMessage := &model.ChatCompletionMessage{Role: model.ChatMessageRoleTool, ToolCallID: call.ID, Content: stringContent(result)}
			finalMessages = append(finalMessages, toolMessage)
			decisionMessages = append(decisionMessages, toolMessage)
		}
	}
	limitMessage := &model.ChatCompletionMessage{Role: model.ChatMessageRoleSystem, Content: stringContent("工具调用轮数已达到上限。请基于已有工具结果回答，并说明可能未完成全部工具调用。")}
	finalMessages = append(finalMessages, limitMessage)
	return finalMessages, nil
}

type thinkTagParser struct {
	inThink bool
	buffer  string
}

const (
	thinkOpenTag  = "<think>"
	thinkCloseTag = "</think>"
)

func (p *thinkTagParser) Accept(delta string) (visible string, reasoning string) {
	p.buffer += delta
	for p.buffer != "" {
		if p.inThink {
			idx := strings.Index(p.buffer, thinkCloseTag)
			if idx >= 0 {
				reasoning += p.buffer[:idx]
				p.buffer = p.buffer[idx+len(thinkCloseTag):]
				p.inThink = false
				continue
			}
			keep := tagPrefixSuffixLen(p.buffer, thinkCloseTag)
			if len(p.buffer) > keep {
				reasoning += p.buffer[:len(p.buffer)-keep]
				p.buffer = p.buffer[len(p.buffer)-keep:]
			}
			return visible, reasoning
		}

		idx := strings.Index(p.buffer, thinkOpenTag)
		if idx >= 0 {
			visible += p.buffer[:idx]
			p.buffer = p.buffer[idx+len(thinkOpenTag):]
			p.inThink = true
			continue
		}
		keep := tagPrefixSuffixLen(p.buffer, thinkOpenTag)
		if len(p.buffer) > keep {
			visible += p.buffer[:len(p.buffer)-keep]
			p.buffer = p.buffer[len(p.buffer)-keep:]
		}
		return visible, reasoning
	}
	return visible, reasoning
}

func (p *thinkTagParser) Flush() (visible string, reasoning string) {
	if p.inThink {
		reasoning = p.buffer
	} else {
		visible = p.buffer
	}
	p.buffer = ""
	p.inThink = false
	return visible, reasoning
}

func tagPrefixSuffixLen(text, tag string) int {
	limit := len(tag) - 1
	if len(text) < limit {
		limit = len(text)
	}
	for i := limit; i > 0; i-- {
		if strings.HasPrefix(tag, text[len(text)-i:]) {
			return i
		}
	}
	return 0
}

func executeAgentToolCall(ctx context.Context, call *model.ToolCall, tools map[string]agentTool, emit func(chatSSEFrame)) string {
	if call == nil || call.Type != model.ToolTypeFunction {
		result := "工具调用无效：仅支持 function 类型工具。"
		if emit != nil {
			emit(chatSSEFrame{Type: "trace", Tool: "agent_tools", Stage: "execute", Status: "error", Message: result})
		}
		return result
	}
	toolName := call.Function.Name
	if emit != nil {
		emit(chatSSEFrame{Type: "trace", Tool: toolName, Stage: "arguments", Status: "running", Message: "准备执行工具", Data: map[string]any{"tool_call_id": call.ID, "arguments": call.Function.Arguments}})
	}
	tool, ok := tools[toolName]
	if !ok {
		result := fmt.Sprintf("工具调用失败：未知工具 %s。", toolName)
		if emit != nil {
			emit(chatSSEFrame{Type: "trace", Tool: toolName, Stage: "execute", Status: "error", Message: result})
		}
		return result
	}
	started := time.Now()
	result, err := tool.execute(ctx, call.Function.Arguments)
	durationMs := time.Since(started).Milliseconds()
	if err != nil {
		message := fmt.Sprintf("工具 %s 执行失败：%v", tool.name, err)
		if emit != nil {
			emit(chatSSEFrame{Type: "trace", Tool: tool.name, Stage: "execute", Status: "error", Message: "工具执行失败", Data: map[string]any{"tool_call_id": call.ID, "duration_ms": durationMs, "error": err.Error()}})
		}
		return message
	}
	if strings.TrimSpace(result) == "" {
		result = fmt.Sprintf("工具 %s 执行完成，但没有返回内容。", tool.name)
	}
	if emit != nil {
		emit(chatSSEFrame{Type: "trace", Tool: tool.name, Stage: "result", Status: "success", Message: "工具执行完成", Data: map[string]any{"tool_call_id": call.ID, "duration_ms": durationMs, "result_preview": truncateString(result, 1200)}})
	}
	return result
}

type ollamaChatRequest struct {
	Model    string              `json:"model"`
	Messages []ollamaChatMessage `json:"messages"`
	Stream   bool                `json:"stream"`
	Options  map[string]int      `json:"options,omitempty"`
}

type ollamaChatMessage struct {
	Role    string   `json:"role"`
	Content string   `json:"content"`
	Images  []string `json:"images,omitempty"`
}

type ollamaChatResponse struct {
	Message struct {
		Role     string `json:"role"`
		Content  string `json:"content"`
		Thinking string `json:"thinking"`
	} `json:"message"`
	Done            bool   `json:"done"`
	PromptEvalCount int    `json:"prompt_eval_count"`
	EvalCount       int    `json:"eval_count"`
	DoneReason      string `json:"done_reason"`
}

func streamOllamaChat(ctx context.Context, profile *OpenAIProfile, messages []*model.ChatCompletionMessage, promptTokens int, usage *tokenUsageTracker, emit func(chatSSEFrame), onDone func(string)) error {
	requestMessages, err := buildOllamaMessages(messages)
	if err != nil {
		return err
	}
	baseURL, err := ollamaBaseURL(profile)
	if err != nil {
		return err
	}
	body, err := json.Marshal(ollamaChatRequest{
		Model:    profile.Config.Model,
		Messages: requestMessages,
		Stream:   true,
		Options:  map[string]int{"num_predict": 4096},
	})
	if err != nil {
		return err
	}
	req, err := http.NewRequestWithContext(ctx, http.MethodPost, strings.TrimRight(baseURL, "/")+"/api/chat", bytes.NewReader(body))
	if err != nil {
		return err
	}
	req.Header.Set("Content-Type", "application/json")
	resp, err := http.DefaultClient.Do(req)
	if err != nil {
		return err
	}
	defer resp.Body.Close()
	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
		data, _ := io.ReadAll(io.LimitReader(resp.Body, 4096))
		return fmt.Errorf("Ollama 原生接口调用失败: %s %s", resp.Status, strings.TrimSpace(string(data)))
	}

	emit(chatSSEFrame{Type: "trace", Tool: "model", Stage: "stream", Status: "running", Message: "Ollama 视觉模型已开始输出"})
	parseThinkTags := shouldParseThinkTags(profile)
	thinkParser := &thinkTagParser{}
	var full strings.Builder
	completionTokens := 0
	streamStarted := time.Now()
	peakTokensPerSecond := 0.0
	emitDelta := func(delta string) {
		if delta == "" {
			return
		}
		full.WriteString(delta)
		completionTokens += estimateTokenCount(delta)
		usage.setModel(promptTokens, completionTokens)
		currentSpeed := tokensPerSecond(completionTokens, streamStarted)
		if currentSpeed > peakTokensPerSecond {
			peakTokensPerSecond = currentSpeed
		}
		stats := usage.snapshot(currentSpeed, peakTokensPerSecond)
		emit(chatSSEFrame{Type: "delta", Text: delta, Stats: &stats})
	}
	emitContent := func(delta string) {
		if delta == "" {
			return
		}
		if !parseThinkTags {
			emitDelta(delta)
			return
		}
		visible, reasoning := thinkParser.Accept(delta)
		if reasoning != "" {
			emit(chatSSEFrame{Type: "reasoning", Text: reasoning})
		}
		emitDelta(visible)
	}

	scanner := bufio.NewScanner(resp.Body)
	scanner.Buffer(make([]byte, 0, 64*1024), 10*1024*1024)
	for scanner.Scan() {
		line := strings.TrimSpace(scanner.Text())
		if line == "" {
			continue
		}
		var chunk ollamaChatResponse
		if err := json.Unmarshal([]byte(line), &chunk); err != nil {
			return fmt.Errorf("解析 Ollama 流失败: %w", err)
		}
		if chunk.Message.Thinking != "" {
			emit(chatSSEFrame{Type: "reasoning", Text: chunk.Message.Thinking})
		}
		emitContent(chunk.Message.Content)
		if chunk.Done {
			if chunk.PromptEvalCount > 0 || chunk.EvalCount > 0 {
				usage.setModel(chunk.PromptEvalCount, chunk.EvalCount)
			}
			break
		}
	}
	if err := scanner.Err(); err != nil {
		return err
	}
	if parseThinkTags {
		visible, reasoning := thinkParser.Flush()
		if reasoning != "" {
			emit(chatSSEFrame{Type: "reasoning", Text: reasoning})
		}
		emitDelta(visible)
	}
	if onDone != nil {
		onDone(full.String())
	}
	finalStats := usage.snapshot(tokensPerSecond(completionTokens, streamStarted), peakTokensPerSecond)
	emit(chatSSEFrame{Type: "stats", Stats: &finalStats})
	emit(chatSSEFrame{Type: "trace", Tool: "model", Stage: "stream", Status: "success", Message: "回答生成完成"})
	return nil
}

func buildOllamaMessages(messages []*model.ChatCompletionMessage) ([]ollamaChatMessage, error) {
	result := make([]ollamaChatMessage, 0, len(messages))
	for _, msg := range messages {
		if msg == nil {
			continue
		}
		role := string(msg.Role)
		if msg.Role == model.ChatMessageRoleTool {
			role = string(model.ChatMessageRoleUser)
		}
		item := ollamaChatMessage{Role: role}
		if msg.Content == nil {
			if len(msg.ToolCalls) > 0 {
				continue
			}
			result = append(result, item)
			continue
		}
		if msg.Content.StringValue != nil {
			item.Content = *msg.Content.StringValue
			if msg.Role == model.ChatMessageRoleTool {
				item.Content = "工具结果：\n" + item.Content
			}
			result = append(result, item)
			continue
		}
		for _, part := range msg.Content.ListValue {
			if part == nil {
				continue
			}
			switch part.Type {
			case model.ChatCompletionMessageContentPartTypeText:
				if part.Text != "" {
					if item.Content != "" {
						item.Content += "\n"
					}
					item.Content += part.Text
				}
			case model.ChatCompletionMessageContentPartTypeImageURL:
				if part.ImageURL == nil {
					continue
				}
				image, err := ollamaImagePayload(part.ImageURL.URL)
				if err != nil {
					return nil, err
				}
				item.Images = append(item.Images, image)
			}
		}
		result = append(result, item)
	}
	return result, nil
}

func ollamaImagePayload(raw string) (string, error) {
	raw = strings.TrimSpace(raw)
	if strings.HasPrefix(strings.ToLower(raw), "data:") {
		comma := strings.Index(raw, ",")
		if comma < 0 {
			return "", errors.New("图片 base64 数据格式错误")
		}
		return strings.TrimSpace(raw[comma+1:]), nil
	}
	return raw, nil
}

func ollamaBaseURL(profile *OpenAIProfile) (string, error) {
	if profile == nil {
		return "", errors.New("Ollama 配置为空")
	}
	u, err := url.Parse(strings.TrimSpace(profile.Config.BaseURL))
	if err != nil {
		return "", err
	}
	if strings.TrimRight(u.Path, "/") == "/v1" {
		u.Path = strings.TrimSuffix(strings.TrimRight(u.Path, "/"), "/v1")
	}
	u.RawQuery = ""
	u.Fragment = ""
	return strings.TrimRight(u.String(), "/"), nil
}

func completeText(ctx context.Context, profile *OpenAIProfile, chatMessages []ChatMessage, maxTokens int) (string, error) {
	return completeTextWithTimeout(ctx, profile, chatMessages, maxTokens, time.Duration(profile.Config.Timeout)*time.Second)
}

func completeTextWithTimeout(ctx context.Context, profile *OpenAIProfile, chatMessages []ChatMessage, maxTokens int, timeout time.Duration) (string, error) {
	messages, err := buildArkMessages(chatMessages)
	if err != nil {
		return "", err
	}
	completionCtx, cancel := context.WithTimeout(ctx, timeout)
	defer cancel()
	stream, err := profile.Client.CreateChatCompletionStream(completionCtx, model.CreateChatCompletionRequest{
		Model:     profile.Config.Model,
		Messages:  messages,
		MaxTokens: intPtr(maxTokens),
	}.WithStream(true))
	if err != nil {
		return "", err
	}
	defer stream.Close()

	promptTokens := estimateChatMessagesTokens(chatMessages)
	completionTokens := 0
	parseThinkTags := shouldParseThinkTags(profile)
	thinkParser := &thinkTagParser{}
	var b strings.Builder
	appendVisible := func(delta string) {
		if delta == "" {
			return
		}
		b.WriteString(delta)
		completionTokens += estimateTokenCount(delta)
	}
	for {
		resp, err := stream.Recv()
		if errors.Is(err, io.EOF) {
			if parseThinkTags {
				visible, _ := thinkParser.Flush()
				appendVisible(visible)
			}
			if tracker := tokenUsageFromContext(ctx); tracker != nil {
				tracker.addTool(promptTokens, completionTokens)
			}
			return b.String(), nil
		}
		if err != nil {
			return "", err
		}
		if len(resp.Choices) > 0 {
			delta := resp.Choices[0].Delta.Content
			if parseThinkTags {
				visible, _ := thinkParser.Accept(delta)
				appendVisible(visible)
			} else {
				appendVisible(delta)
			}
		}
	}
}

func completeChatWithTimeout(ctx context.Context, profile *OpenAIProfile, request model.CreateChatCompletionRequest, timeout time.Duration) (model.ChatCompletionResponse, error) {
	completionCtx, cancel := context.WithTimeout(ctx, timeout)
	defer cancel()
	return profile.Client.CreateChatCompletion(completionCtx, request.WithStream(false))
}

func newUUID() string {
	b := make([]byte, 16)
	_, _ = rand.Read(b)
	b[6] = (b[6] & 0x0f) | 0x40
	b[8] = (b[8] & 0x3f) | 0x80
	return hex.EncodeToString(b[:4]) + "-" + hex.EncodeToString(b[4:6]) + "-" +
		hex.EncodeToString(b[6:8]) + "-" + hex.EncodeToString(b[8:10]) + "-" +
		hex.EncodeToString(b[10:])
}

// ─── ConvStore ─────────────────────────────────────────────

func NewConvStore(dir string) *ConvStore {
	os.MkdirAll(dir, 0755)
	return &ConvStore{dir: dir}
}

func (s *ConvStore) path(id string) string {
	return filepath.Join(s.dir, id+".json")
}

func (s *ConvStore) Create() (*Conversation, error) {
	conv := &Conversation{
		ID:        newUUID(),
		Title:     "新对话",
		CreatedAt: time.Now(),
		UpdatedAt: time.Now(),
	}
	if err := s.Save(conv); err != nil {
		return nil, err
	}
	return conv, nil
}

func (s *ConvStore) Save(conv *Conversation) error {
	s.mu.Lock()
	defer s.mu.Unlock()
	conv.UpdatedAt = time.Now()
	return atomicWriteJSON(s.path(conv.ID), conv)
}

func (s *ConvStore) Get(id string) (*Conversation, error) {
	s.mu.Lock()
	defer s.mu.Unlock()
	data, err := os.ReadFile(s.path(id))
	if err != nil {
		if os.IsNotExist(err) {
			return nil, errors.New("对话不存在")
		}
		return nil, fmt.Errorf("读取对话失败: %w", err)
	}
	var conv Conversation
	if err := json.Unmarshal(data, &conv); err != nil {
		return nil, fmt.Errorf("解析对话失败: %w", err)
	}
	return &conv, nil
}

func (s *ConvStore) List() ([]Conversation, error) {
	s.mu.Lock()
	defer s.mu.Unlock()

	entries, err := os.ReadDir(s.dir)
	if err != nil {
		return nil, fmt.Errorf("读取对话目录失败: %w", err)
	}

	var list []Conversation
	for _, e := range entries {
		if e.IsDir() || filepath.Ext(e.Name()) != ".json" {
			continue
		}
		data, err := os.ReadFile(filepath.Join(s.dir, e.Name()))
		if err != nil {
			continue
		}
		var conv Conversation
		if err := json.Unmarshal(data, &conv); err != nil {
			continue
		}
		conv.Messages = nil // 列表不返回消息体
		list = append(list, conv)
	}

	sort.Slice(list, func(i, j int) bool {
		return list[i].UpdatedAt.After(list[j].UpdatedAt)
	})
	return list, nil
}

func (s *ConvStore) Delete(id string) error {
	s.mu.Lock()
	defer s.mu.Unlock()
	if err := os.Remove(s.path(id)); err != nil && !os.IsNotExist(err) {
		return fmt.Errorf("删除对话失败: %w", err)
	}
	return nil
}

func atomicWriteJSON(path string, v any) error {
	tmp := path + ".tmp"
	data, err := json.Marshal(v)
	if err != nil {
		return err
	}
	if err := os.WriteFile(tmp, data, 0644); err != nil {
		return err
	}
	return os.Rename(tmp, path)
}

func saveConversationMessages(id string, messages []ChatMessage, assistantContent string) error {
	conv, err := store.Get(id)
	if err != nil {
		return err
	}
	conv.Messages = append([]ChatMessage(nil), messages...)
	conv.Messages = append(conv.Messages, ChatMessage{Role: "assistant", Content: assistantContent})
	if conv.Title == "" || conv.Title == "新对话" {
		conv.Title = genConvTitle(conv.Messages)
	}
	return store.Save(conv)
}

func genConvTitle(messages []ChatMessage) string {
	for _, m := range messages {
		if m.Hidden {
			continue
		}
		if m.Role == "user" && strings.TrimSpace(m.Content) != "" {
			title := strings.TrimSpace(m.Content)
			title = strings.ReplaceAll(title, "\r\n", " ")
			title = strings.ReplaceAll(title, "\n", " ")
			runes := []rune(title)
			if len(runes) > 30 {
				return string(runes[:30]) + "..."
			}
			return title
		}
	}
	return "新对话"
}

const maxImageSize = 4 * 1024 * 1024

var allowedImageTypes = map[string]bool{
	"image/jpeg": true,
	"image/png":  true,
	"image/webp": true,
	"image/gif":  true,
}

func buildArkMessages(chatMessages []ChatMessage) ([]*model.ChatCompletionMessage, error) {
	messages := make([]*model.ChatCompletionMessage, 0, len(chatMessages))
	for _, m := range chatMessages {
		msg, err := buildArkMessage(m)
		if err != nil {
			return nil, err
		}
		messages = append(messages, msg)
	}
	return messages, nil
}

func hasImageMessage(messages []ChatMessage) bool {
	for _, msg := range messages {
		if strings.TrimSpace(msg.ImageURL) != "" || strings.TrimSpace(msg.ImageURLAlias) != "" {
			return true
		}
	}
	return false
}

func buildToolDecisionMessages(chatMessages []ChatMessage) ([]*model.ChatCompletionMessage, error) {
	messages := make([]*model.ChatCompletionMessage, 0, len(chatMessages))
	for _, m := range chatMessages {
		content := m.Content
		if strings.TrimSpace(m.ImageURL) != "" || strings.TrimSpace(m.ImageURLAlias) != "" {
			content = strings.TrimSpace(content)
			placeholder := "[用户上传了一张图片。工具判断阶段不读取图片内容；如果问题主要依赖识图，应不要调用工具，交给最终多模态模型回答。]"
			if content == "" {
				content = placeholder
			} else {
				content += "\n\n" + placeholder
			}
		}
		messages = append(messages, &model.ChatCompletionMessage{Role: m.Role, Content: stringContent(content)})
	}
	return messages, nil
}

func buildArkMessage(m ChatMessage) (*model.ChatCompletionMessage, error) {
	msg := &model.ChatCompletionMessage{Role: m.Role}

	if m.ImageURL == "" && m.ImageURLAlias != "" {
		m.ImageURL = m.ImageURLAlias
	}

	if m.ImageURL == "" {
		msg.Content = &model.ChatCompletionMessageContent{
			StringValue: &m.Content,
		}
		return msg, nil
	}

	imageURL, err := normalizeImageURL(m.ImageURL)
	if err != nil {
		return nil, err
	}

	// 有图片时：文字内容可有可无（图片 caption 场景），均构造多模态消息
	// 若无文字，则只传图片 part；若同时有图片和文字，先文后图
	parts := make([]*model.ChatCompletionMessageContentPart, 0, 2)
	if m.Content != "" {
		parts = append(parts, textPart(m.Content))
	}
	parts = append(parts, imagePart(imageURL))
	msg.Content = &model.ChatCompletionMessageContent{ListValue: parts}
	return msg, nil
}

func imagePart(url string) *model.ChatCompletionMessageContentPart {
	return &model.ChatCompletionMessageContentPart{
		Type: model.ChatCompletionMessageContentPartTypeImageURL,
		ImageURL: &model.ChatMessageImageURL{
			URL:    url,
			Detail: model.ImageURLDetailAuto,
		},
	}
}

func textPart(text string) *model.ChatCompletionMessageContentPart {
	return &model.ChatCompletionMessageContentPart{
		Type: model.ChatCompletionMessageContentPartTypeText,
		Text: text,
	}
}

func stringContent(text string) *model.ChatCompletionMessageContent {
	return &model.ChatCompletionMessageContent{StringValue: &text}
}

func chatMessageContentString(content *model.ChatCompletionMessageContent) string {
	if content == nil || content.StringValue == nil {
		return ""
	}
	return *content.StringValue
}

func normalizeImageURL(raw string) (string, error) {
	raw = strings.TrimSpace(raw)
	if raw == "" {
		return "", errors.New("图片地址不能为空")
	}

	lower := strings.ToLower(raw)
	if strings.HasPrefix(lower, "data:") {
		return normalizeImageDataURI(raw)
	}

	u, err := url.Parse(raw)
	if err != nil || u.Host == "" || (u.Scheme != "http" && u.Scheme != "https") {
		return "", errors.New("图片地址无效，仅支持 http/https URL 或 base64 data URI")
	}
	return raw, nil
}

func normalizeImageDataURI(raw string) (string, error) {
	comma := strings.Index(raw, ",")
	if comma < 0 {
		return "", errors.New("图片 base64 数据格式错误")
	}

	meta := strings.ToLower(strings.TrimSpace(raw[5:comma]))
	payload := strings.TrimSpace(raw[comma+1:])
	if payload == "" {
		return "", errors.New("图片 base64 数据不能为空")
	}
	parts := strings.Split(meta, ";")
	if len(parts) < 2 || !contains(parts[1:], "base64") {
		return "", errors.New("图片 data URI 必须使用 base64 编码")
	}

	mime := parts[0]
	if !allowedImageTypes[mime] {
		return "", errors.New("图片格式不支持，仅支持 jpeg/png/webp/gif")
	}

	decoded, err := base64.StdEncoding.DecodeString(payload)
	if err != nil {
		return "", errors.New("图片 base64 数据无效")
	}
	if len(decoded) > maxImageSize {
		return "", errors.New("图片过大，请选择小于 4MB 的图片")
	}

	return "data:" + mime + ";base64," + payload, nil
}

func contains(items []string, target string) bool {
	for _, item := range items {
		if strings.TrimSpace(item) == target {
			return true
		}
	}
	return false
}

func intPtr(i int) *int { return &i }

func boolPtr(v bool) *bool { return &v }

func truncateString(text string, maxRunes int) string {
	runes := []rune(strings.TrimSpace(text))
	if maxRunes <= 0 || len(runes) <= maxRunes {
		return string(runes)
	}
	return string(runes[:maxRunes]) + "..."
}

func writeSSEJSON(w io.Writer, frame chatSSEFrame) {
	data, err := json.Marshal(frame)
	if err != nil {
		data, _ = json.Marshal(chatSSEFrame{Type: "error", Error: "序列化流事件失败"})
	}
	fmt.Fprintf(w, "data: %s\n\n", data)
}

func toJSON(s string) string {
	b, _ := json.Marshal(s)
	return string(b)
}

func toSSE(s string) string {
	s = strings.ReplaceAll(s, `\`, `\\`)
	s = strings.ReplaceAll(s, "\n", `\n`)
	s = strings.ReplaceAll(s, "\r", "")
	s = strings.ReplaceAll(s, `"`, `\"`)
	return fmt.Sprintf(`"%s"`, s)
}

// ─── 入口 ─────────────────────────────────────────────────

func main() {
	var err error
	cfg, err = loadConfig("config.yaml")
	if err != nil {
		fmt.Fprintln(os.Stderr, "配置加载失败:", err)
		os.Exit(1)
	}

	// 初始化火山方舟 SDK 客户端
	aiState, err = NewOpenAIState(cfg.OpenAI)
	if err != nil {
		fmt.Fprintln(os.Stderr, "OpenAI 配置初始化失败:", err)
		os.Exit(1)
	}
	searchConfig, err := searchagent.LoadConfig("agents/search/config.yaml", legacySearchProfiles)
	if err != nil {
		fmt.Fprintln(os.Stderr, "联网搜索配置加载失败:", err)
		os.Exit(1)
	}
	searchState, err = searchagent.NewState(searchConfig)
	if err != nil {
		fmt.Fprintln(os.Stderr, "联网搜索初始化失败:", err)
		os.Exit(1)
	}
	sqlConfig, err := sqlquery.LoadConfig("agents/sql/config.yaml")
	if err != nil {
		fmt.Fprintln(os.Stderr, "SQL 查询插件配置加载失败:", err)
		os.Exit(1)
	}
	sqlState, err = sqlquery.NewState(sqlConfig)
	if err != nil {
		fmt.Fprintln(os.Stderr, "SQL 查询插件初始化失败:", err)
		os.Exit(1)
	}
	defer sqlState.Close()
	toolRouterState, err = NewToolRouterState(&cfg.ToolRouter, aiState)
	if err != nil {
		fmt.Fprintln(os.Stderr, "工具路由配置初始化失败:", err)
		os.Exit(1)
	}
	store = NewConvStore("conversations")

	// Gin 路由
	r := gin.Default()
	r.LoadHTMLGlob("templates/*")
	r.Static("/static", "./static")

	r.GET("/", indexHandler)
	r.POST("/api/chat", chatHandler)
	r.GET("/api/openai", listOpenAIHandler)
	r.POST("/api/openai/active", switchOpenAIHandler)
	r.GET("/api/search", listSearchHandler)
	r.POST("/api/search/active", switchSearchHandler)
	r.GET("/api/conversations", listConversationsHandler)
	r.POST("/api/conversations", createConversationHandler)
	r.GET("/api/conversations/:id", getConversationHandler)
	r.DELETE("/api/conversations/:id", deleteConversationHandler)

	// 根据配置选择监听方式
	switch strings.ToLower(cfg.Server.Mode) {
	case "unix":
		socketPath := cfg.Server.Address
		if _, statErr := os.Stat(socketPath); statErr == nil {
			os.Remove(socketPath)
		}
		ln, listenErr := net.Listen("unix", socketPath)
		if listenErr != nil {
			fmt.Fprintln(os.Stderr, "监听 Unix socket 失败:", listenErr)
			os.Exit(1)
		}
		fmt.Println("服务已启动，监听 Unix socket:", socketPath)
		if serveErr := http.Serve(ln, r); serveErr != nil {
			fmt.Fprintln(os.Stderr, "服务异常退出:", serveErr)
			os.Exit(1)
		}
	default:
		fmt.Println("服务已启动，监听 TCP:", cfg.Server.Address)
		if runErr := r.Run(cfg.Server.Address); runErr != nil {
			fmt.Fprintln(os.Stderr, "服务异常退出:", runErr)
			os.Exit(1)
		}
	}
}