fix: Думать - the model is reminded in the question itself, the label no longer lands in the arrow, and the thinking with its time is kept to show again later

internal/server/ai.go

@@ -41,6 +41,8 @@ type aiConfig struct{ Type, URL, Key, Model string }
 type aiMsg struct {
 	Role    string `json:"role"`
 	Content string `json:"content"`
+	// ThinkMS: how long the model thought before this answer (kept to show it).
+	ThinkMS int64 `json:"think_ms,omitempty"`
 }
 
 func (s *Server) aiConfig() aiConfig {
@@ -440,6 +442,11 @@ func (s *Server) aiDeleteChat(owner, chat int64) {
 
 // aiAppendTo keeps a message of owner's chat; the first question names it.
 func (s *Server) aiAppendTo(owner, chat int64, role, content string) {
+	s.aiKeepMsg(owner, chat, aiMsg{Role: role, Content: content})
+}
+
+func (s *Server) aiKeepMsg(owner, chat int64, m aiMsg) {
+	role, content := m.Role, m.Content
 	if content == "" || !s.aiOwns(owner, chat) {
 		return
 	}
@@ -447,7 +454,7 @@ func (s *Server) aiAppendTo(owner, chat int64, role, content string) {
 		content = utf8Safe(content[:64<<10])
 	}
 	now := time.Now().Unix()
-	_, _ = s.db.sql.Exec(`INSERT INTO ai_messages (admin_id, chat_id, role, content, at) VALUES (?, ?, ?, ?, ?)`, owner, chat, role, content, now)
+	_, _ = s.db.sql.Exec(`INSERT INTO ai_messages (admin_id, chat_id, role, content, at, think_ms) VALUES (?, ?, ?, ?, ?, ?)`, owner, chat, role, content, now, m.ThinkMS)
 	_, _ = s.db.sql.Exec(`DELETE FROM ai_messages WHERE chat_id = ? AND id <= (SELECT id FROM ai_messages WHERE chat_id = ? ORDER BY id DESC LIMIT 1 OFFSET ?)`, chat, chat, aiKeep)
 	_, _ = s.db.sql.Exec(`UPDATE ai_chats SET updated = ? WHERE id = ?`, now, chat)
 	if role == "user" {
@@ -461,7 +468,7 @@ func (s *Server) aiAppendTo(owner, chat int64, role, content string) {
 
 // aiHistoryOf is a chat's talk, oldest first, the last limit messages.
 func (s *Server) aiHistoryOf(owner, chat int64, limit int) []aiMsg {
-	rows, err := s.db.sql.Query(`SELECT role, content FROM ai_messages WHERE admin_id = ? AND chat_id = ? ORDER BY id DESC LIMIT ?`, owner, chat, limit)
+	rows, err := s.db.sql.Query(`SELECT role, content, think_ms FROM ai_messages WHERE admin_id = ? AND chat_id = ? ORDER BY id DESC LIMIT ?`, owner, chat, limit)
 	out := []aiMsg{}
 	if err != nil {
 		return out
@@ -469,7 +476,7 @@ func (s *Server) aiHistoryOf(owner, chat int64, limit int) []aiMsg {
 	defer rows.Close()
 	for rows.Next() {
 		var m aiMsg
-		if rows.Scan(&m.Role, &m.Content) == nil {
+		if rows.Scan(&m.Role, &m.Content, &m.ThinkMS) == nil {
 			out = append(out, m)
 		}
 	}
@@ -541,24 +548,44 @@ func (s *Server) handleAIChat(w http.ResponseWriter, r *http.Request) {
 	}
 	if in.Think {
 		system += aiThink
+		// Models heed the last words most: the reminder goes there too (not
+		// kept with the talk).
+		if n := len(msgs); n > 0 && msgs[n-1].Role == "user" {
+			msgs[n-1].Content += "\n\n(Сначала подумай: первая строка ответа - ::думаю::, дальше рассуждение, потом строка ::ответ:: и сам ответ.)"
+		}
 	}
 	w.Header().Set("Content-Type", "text/plain; charset=utf-8")
 	w.Header().Set("Cache-Control", "no-store")
 	w.Header().Set("X-Accel-Buffering", "no")
 	fl, _ := w.(http.Flusher)
 	var answer strings.Builder
+	var thinkFrom, thinkTo time.Time // when the thinking began and ended
 	emit := func(t string) {
 		if t == "" {
 			return
 		}
 		answer.WriteString(t)
+		if a := answer.String(); thinkFrom.IsZero() && (strings.Contains(a, "::думаю::") || strings.Contains(a, "<think>")) {
+			thinkFrom = time.Now()
+		} else if !thinkFrom.IsZero() && thinkTo.IsZero() && (strings.Contains(a, "::ответ::") || strings.Contains(a, "</think>")) {
+			thinkTo = time.Now()
+		}
 		_, _ = io.WriteString(w, t)
 		if fl != nil {
 			fl.Flush()
 		}
 	}
 	err := aiStream(r.Context(), s.aiConfig(), system+"\n\n--- Панель сейчас ---\n"+in.Context, msgs, emit)
-	s.aiAppendTo(a.ID, in.Chat, "assistant", stripThink(answer.String()))
+	// The answer is kept whole, thinking too (to show it again); the model
+	// gets it without (aiWindow).
+	ans := aiMsg{Role: "assistant", Content: answer.String()}
+	if !thinkFrom.IsZero() {
+		if thinkTo.IsZero() {
+			thinkTo = time.Now()
+		}
+		ans.ThinkMS = thinkTo.Sub(thinkFrom).Milliseconds()
+	}
+	s.aiKeepMsg(a.ID, in.Chat, ans)
 	if err != nil && r.Context().Err() == nil {
 		if answer.Len() == 0 {
 			w.WriteHeader(http.StatusBadGateway)
@@ -679,7 +706,9 @@ func aiWindow(rows []aiMsg, budget int) []aiMsg {
 			}
 			m = aiMsg{Role: "user", Content: "Результат действия:\n" + cut(m.Content, max)}
 		case m.Role == "assistant" && i < len(rows)-2:
-			m.Content = cut(m.Content, 3000)
+			m.Content = cut(stripThink(m.Content), 3000)
+		case m.Role == "assistant":
+			m.Content = stripThink(m.Content)
 		case m.Role != "assistant":
 			m.Role = "user"
 		}