MuyueWorkspace/internal/api/handlers_chat.go

package api

import (
	"bytes"
	"context"
	"encoding/json"
	"fmt"
	"io"
	"net/http"
	"os"
	"path/filepath"
	"regexp"
	"strings"
	"time"
	"unicode/utf8"

	"github.com/muyue/muyue/internal/agent"
	"github.com/muyue/muyue/internal/orchestrator"
	"github.com/muyue/muyue/internal/platform"
)

var thinkingTagRegex = regexp.MustCompile(`(?s)<[Tt]hink[^>]*>.*?</[Tt]hink>`)
var fileMentionRegex = regexp.MustCompile(`@(\S+\.[a-zA-Z0-9]+)`)

type ImageAttachment struct {
	Data     string `json:"data"`
	Filename string `json:"filename"`
	MimeType string `json:"mime_type"`
}

func resolveFileMentions(text string) string {
	return fileMentionRegex.ReplaceAllStringFunc(text, func(match string) string {
		filePath := match[1:]
		if strings.HasPrefix(filePath, "~/") {
			if home, err := os.UserHomeDir(); err == nil {
				filePath = filepath.Join(home, filePath[2:])
			}
		}
		if !filepath.IsAbs(filePath) {
			if home, err := os.UserHomeDir(); err == nil {
				filePath = filepath.Join(home, filePath)
			}
		}
		data, err := os.ReadFile(filePath)
		if err != nil {
			return match + fmt.Sprintf(" (erreur: fichier non trouve)")
		}
		content := string(data)
		if len(content) > 50000 {
			content = content[:50000] + "\n... (tronque a 50Ko)"
		}
		return fmt.Sprintf("[Fichier: %s]\n%s\n[Fin du fichier: %s]", filepath.Base(filePath), content, filepath.Base(filePath))
	})
}

var vlmClient = &http.Client{Timeout: 60 * time.Second}

func (s *Server) describeImages(images []ImageAttachment) []string {
	var apiKey string
	for i := range s.config.AI.Providers {
		if s.config.AI.Providers[i].Active {
			apiKey = s.config.AI.Providers[i].APIKey
			break
		}
	}
	if apiKey == "" {
		return nil
	}

	descriptions := make([]string, 0, len(images))
	for _, img := range images {
		desc, err := s.callVLM(apiKey, img)
		if err != nil {
			descriptions = append(descriptions, fmt.Sprintf("(description unavailable: %v)", err))
		} else {
			descriptions = append(descriptions, desc)
		}
	}
	return descriptions
}

func (s *Server) callVLM(apiKey string, img ImageAttachment) (string, error) {
	payload := map[string]string{
		"prompt":    "Describe this image in detail. Include all text, UI elements, code, diagrams, or data visible. Be thorough and specific.",
		"image_url": img.Data,
	}
	body, err := json.Marshal(payload)
	if err != nil {
		return "", fmt.Errorf("marshal vlm request: %w", err)
	}

	ctx, cancel := context.WithTimeout(context.Background(), 55*time.Second)
	defer cancel()

	req, err := http.NewRequestWithContext(ctx, "POST", "https://api.minimax.io/v1/coding_plan/vlm", bytes.NewReader(body))
	if err != nil {
		return "", fmt.Errorf("create vlm request: %w", err)
	}
	req.Header.Set("Content-Type", "application/json")
	req.Header.Set("Authorization", "Bearer "+apiKey)

	resp, err := vlmClient.Do(req)
	if err != nil {
		return "", fmt.Errorf("vlm request: %w", err)
	}
	defer resp.Body.Close()

	respBody, err := io.ReadAll(resp.Body)
	if err != nil {
		return "", fmt.Errorf("read vlm response: %w", err)
	}

	if resp.StatusCode != http.StatusOK {
		return "", fmt.Errorf("vlm API error (%d): %s", resp.StatusCode, string(respBody))
	}

	var result struct {
		Content string `json:"content"`
	}
	if err := json.Unmarshal(respBody, &result); err != nil {
		return "", fmt.Errorf("parse vlm response: %w", err)
	}

	if result.Content == "" {
		return "(empty description)", nil
	}
	return result.Content, nil
}

func (s *Server) handleChat(w http.ResponseWriter, r *http.Request) {
	if r.Method != "POST" {
		writeError(w, "POST only", http.StatusMethodNotAllowed)
		return
	}
	r.Body = http.MaxBytesReader(w, r.Body, 50*1024*1024)
	var body struct {
		Message            string            `json:"message"`
		Stream             bool              `json:"stream"`
		Images             []ImageAttachment `json:"images"`
		AdvancedReflection bool              `json:"advanced_reflection"`
	}
	if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
		writeError(w, err.Error(), http.StatusBadRequest)
		return
	}
	if body.Message == "" {
		writeError(w, "no message", http.StatusMethodNotAllowed)
		return
	}
	if len(body.Images) > 3 {
		writeError(w, "max 3 images", http.StatusBadRequest)
		return
	}

	enrichedMessage := resolveFileMentions(body.Message)

	var imageIDs []string
	if len(body.Images) > 0 {
		descriptions := s.describeImages(body.Images)
		var imgContext strings.Builder
		for i, desc := range descriptions {
			imgContext.WriteString(fmt.Sprintf("\n[Image %d (%s): %s]\n", i+1, body.Images[i].Filename, desc))

			id, err := saveImage(body.Images[i].Data, body.Images[i].Filename, body.Images[i].MimeType)
			if err != nil {
				_ = err
			} else {
				imageIDs = append(imageIDs, id)
			}
		}
		enrichedMessage = imgContext.String() + enrichedMessage
	}

	displayMsg := body.Message
	if len(body.Images) > 0 {
		imgNames := make([]string, len(body.Images))
		for i, img := range body.Images {
			imgNames[i] = img.Filename
		}
		displayMsg += " [" + strings.Join(imgNames, ", ") + "]"
	}

	if len(imageIDs) > 0 {
		s.convStore.AddWithImages("user", displayMsg, imageIDs)
	} else {
		s.convStore.Add("user", displayMsg)
	}

	if s.convStore.NeedsSummarization() {
		s.autoSummarize()
	}

	orb, err := orchestrator.New(s.config)
	if err != nil {
		writeError(w, err.Error(), http.StatusServiceUnavailable)
		return
	}
	var studioPrompt strings.Builder
	studioPrompt.WriteString(agent.StudioSystemPrompt())
	sysInfo := platform.Detect()
	osName := sysInfo.OSName
	if osName == "" {
		osName = string(sysInfo.OS)
	}
	studioPrompt.WriteString(fmt.Sprintf("\nDate: %s\nHeure: %s\nSystème: %s\n", time.Now().Format("02/01/2006"), time.Now().Format("15:04:05"), osName))
	canSudo := !agent.NeedsSudoPassword()
	studioPrompt.WriteString(fmt.Sprintf("Root: %t\n", !canSudo))
	if !canSudo {
		studioPrompt.WriteString("⚠️ Session sans sudo sans mot de passe — les commandes sudo/doas nécessitent une autorisation. N'utilise PAS sudo ou doas sans demander.\n")
	} else {
		studioPrompt.WriteString("⚠️ Session avec privilèges sudo sans mot de passe — les commandes sudo s'exécuteront directement.\n")
	}
	orb.SetSystemPrompt(studioPrompt.String())
	orb.SetTools(s.agentToolsJSON)

	if memBlock := s.buildMemoryContext(enrichedMessage); memBlock != "" {
		orb.AppendHistory(orchestrator.Message{
			Role:    "system",
			Content: orchestrator.TextContent(memBlock),
		})
	}

	// Auto-force advanced reflection while a browser-test session is active:
	// the user is doing AI-driven UI testing, where having a second model
	// produce a preliminary report (when one is configured) materially
	// improves which clicks the active model decides to perform. The toggle
	// remains user-controllable for non-test conversations.
	wantReflection := body.AdvancedReflection
	if !wantReflection && s.browserTestStore != nil && len(s.browserTestStore.List()) > 0 {
		wantReflection = true
	}

	if wantReflection {
		if report, ok := s.runReflectionReport(enrichedMessage); ok {
			enrichedMessage = enrichedMessage + "\n\n[RAPPORT PRÉALABLE — produit par un autre modèle, à valider]\n" + report + "\n[/RAPPORT PRÉALABLE]"
		}
	}

	if body.Stream {
		s.handleStreamChat(w, orb, enrichedMessage)
	} else {
		s.handleNonStreamChat(w, orb, enrichedMessage)
	}
}

func (s *Server) handleStreamChat(w http.ResponseWriter, orb *orchestrator.Orchestrator, userMessage string) {
	SetupSSEHeaders(w)
	flusher, canFlush := w.(http.Flusher)


	sseWriter := NewSSEWriter(w)


	ctx := context.Background()
	messages := s.buildContextMessages(userMessage)

	engine := NewChatEngine(orb, s.agentRegistry, s.agentToolsJSON)
	engine.SetLimiter(s.AcquireAgentSlot)
	engine.OnChunk(func(data map[string]interface{}) {
		if data == nil {
			return
		}
		sseWriter.Write(data)
		if canFlush {
			flusher.Flush()
		}
	})

	finalContent, allToolCalls, allToolResults, err := engine.RunWithTools(ctx, messages)
	if err != nil {
		sseWriter.Write(map[string]interface{}{"error": err.Error()})
		return
	}

	storeContent := finalContent
	if len(allToolCalls) > 0 {
		storeObj := map[string]interface{}{
			"content":      storeContent,
			"tool_calls":   allToolCalls,
			"tool_results": allToolResults,
		}
		storeJSON, _ := json.Marshal(storeObj)
		storeContent = string(storeJSON)
	}
	s.convStore.Add("assistant", storeContent)

	s.consumption.Record(engine.ProviderName(), engine.TotalTokens)

	sseWriter.Write(map[string]interface{}{"done": "true"})
}

func (s *Server) handleNonStreamChat(w http.ResponseWriter, orb *orchestrator.Orchestrator, userMessage string) {
	ctx := context.Background()
	messages := s.buildContextMessages(userMessage)

	engine := NewChatEngine(orb, s.agentRegistry, s.agentToolsJSON)
	engine.SetLimiter(s.AcquireAgentSlot)
	finalContent, err := engine.RunNonStream(ctx, messages)
	if err != nil {
		writeError(w, err.Error(), http.StatusInternalServerError)
		return
	}

	s.convStore.Add("assistant", finalContent)

	s.consumption.Record(engine.ProviderName(), engine.TotalTokens)

	writeJSON(w, map[string]string{"content": finalContent})
}

func cleanThinkingTags(content string) string {
	return strings.TrimSpace(thinkingTagRegex.ReplaceAllString(content, ""))
}

// runReflectionReport runs the inactive AI provider on the user message to
// produce a preliminary analysis report that the active provider will then
// use as additional context. Returns ("", false) if no inactive provider is
// configured or on error — the caller falls back to a normal chat flow.
func (s *Server) runReflectionReport(userMessage string) (string, bool) {
	orb, err := orchestrator.NewForInactiveProvider(s.config)
	if err != nil {
		return "", false
	}
	orb.SetSystemPrompt("Tu es un analyste. Pour la question ci-dessous, produis un rapport bref (max 15 lignes) qui : (1) reformule l'objectif de l'utilisateur, (2) liste les points à clarifier ou les risques, (3) suggère une approche structurée. Pas de code, pas d'action — uniquement de l'analyse.")
	resp, err := orb.SendNoTools(userMessage)
	if err != nil {
		return "", false
	}
	return strings.TrimSpace(resp), true
}

func (s *Server) buildContextMessages(userMessage string) []orchestrator.Message {
	history := s.convStore.Get()

	sysPromptTokens := utf8.RuneCountInString(agent.StudioSystemPrompt())/charsPerToken + 50
	toolsTokens := utf8.RuneCountInString(string(s.agentToolsJSON)) / charsPerToken
	responseMargin := 4000
	userMsgTokens := utf8.RuneCountInString(userMessage) / charsPerToken

	overhead := sysPromptTokens + toolsTokens + responseMargin + userMsgTokens
	available := contextWindowTokens - overhead
	if available < 1000 {
		available = 1000
	}

	included := 0
	tokensUsed := 0
	for i := len(history) - 1; i >= 0; i-- {
		if history[i].Summarized {
			break
		}
		displayContent := extractDisplayContent(history[i].Role, history[i].Content)
		msgTokens := utf8.RuneCountInString(displayContent) / charsPerToken
		if msgTokens == 0 {
			msgTokens = 1
		}
		if tokensUsed+msgTokens > available {
			break
		}
		tokensUsed += msgTokens
		included++
	}

	start := len(history) - included
	if start < 0 {
		start = 0
	}

	hasSummarized := false
	for i := 0; i < start; i++ {
		if history[i].Summarized {
			hasSummarized = true
			break
		}
	}
	if start > 0 {
		_ = start
	}

	messages := make([]orchestrator.Message, 0, included+2)

	summary := s.convStore.GetSummary()
	if summary != "" && (start > 0 || hasSummarized) {
		messages = append(messages, orchestrator.Message{
			Role:    "system",
			Content: orchestrator.TextContent("Résumé de la conversation précédente:\n" + summary),
		})
	}

	for _, m := range history[start:] {
		if m.Role == "system" {
			continue
		}
		displayContent := extractDisplayContent(m.Role, m.Content)
		messages = append(messages, orchestrator.Message{
			Role:    m.Role,
			Content: orchestrator.TextContent(displayContent),
		})
	}

	messages = append(messages, orchestrator.Message{
		Role:    "user",
		Content: orchestrator.TextContent(userMessage),
	})

	return messages
}

func (s *Server) autoSummarize() {
	messages := s.convStore.Get()
	if len(messages) < 10 {
		return
	}

	half := len(messages) / 2
	var oldText string
	for _, m := range messages[:half] {
		oldText += m.Role + ": " + m.Content + "\n\n"
	}

	summary := s.convStore.GetSummary()
	if summary != "" {
		oldText = "Résumé précédent:\n" + summary + "\n\nNouveaux échanges:\n" + oldText
	}

	orb, err := orchestrator.New(s.config)
	if err != nil {
		return
	}
	orb.SetSystemPrompt(summarizePrompt)

	result, err := orb.Send(oldText)
	if err != nil {
		return
	}

	s.convStore.SetSummary(result)
	s.convStore.MarkSummarized(half)
}

func (s *Server) handleChatHistory(w http.ResponseWriter, r *http.Request) {
	if r.Method != "GET" {
		writeError(w, "GET only", http.StatusMethodNotAllowed)
		return
	}
	messages := s.convStore.Get()
	writeJSON(w, map[string]interface{}{
		"messages":     messages,
		"tokens":       s.convStore.ApproxTokenCount(),
		"max_tokens":   contextWindowTokens,
		"summarize_at": int(float64(contextWindowTokens) * summarizeRatio),
		"summary":      s.convStore.GetSummary(),
	})
}

func (s *Server) handleChatClear(w http.ResponseWriter, r *http.Request) {
	if r.Method != "POST" {
		writeError(w, "POST only", http.StatusMethodNotAllowed)
		return
	}
	s.convStore.Clear()
	writeJSON(w, map[string]string{"status": "ok"})
}

func (s *Server) handleChatSummarize(w http.ResponseWriter, r *http.Request) {
	if r.Method != "POST" {
		writeError(w, "POST only", http.StatusMethodNotAllowed)
		return
	}
	s.autoSummarize()
	writeJSON(w, map[string]interface{}{
		"status":  "ok",
		"tokens":  s.convStore.ApproxTokenCount(),
		"summary": s.convStore.GetSummary(),
	})
}