update fronted and fix llm

.env

@@ -1,3 +1,6 @@
 BOT_TOKEN=8156642467:AAEalekr-33lFS3tTTkXIEiaawfDzR6lBJc
 TZ=Europe/Moscow
 ODDS_API_KEY=c29bc26dee8aa9d95a5ce8d14ea63923
+LLAMA_API_URL=https://mirror.porno4free.ru/zovos-ai/
+LLAMA_FORCE_DISABLE_THINKING=0
+LLAMA_LOG_THINKING=1

AI/talk_handler.py

@@ -24,6 +24,7 @@ DB_PATH = os.getenv("CHAT_HISTORY_DB_PATH") or str(Path(__file__).resolve().with
 BOT_MEMORY_NAME = os.getenv("BOT_MEMORY_NAME", "бот")
 SKIP_TOKEN = "<skip>"
 FORCE_DISABLE_THINKING = os.getenv("LLAMA_FORCE_DISABLE_THINKING", "1").lower() not in {"0", "false", "no"}
+LOG_THINKING = os.getenv("LLAMA_LOG_THINKING", "1").lower() not in {"0", "false", "no"}
 
 RECENT_MESSAGES_LIMIT = 14
 SUMMARY_TRIGGER_MESSAGES = 24
@@ -614,6 +615,13 @@ async def _call_llm(
     content = message.get("content", "")
     reasoning_content = (message.get("reasoning_content") or "").strip()
     cleaned_content = content.strip()
+    if reasoning_content and LOG_THINKING:
+        logger.info(
+            "LLM reasoning detected. disable_thinking=%s finish_reason=%s reasoning=%s",
+            disable_thinking,
+            finish_reason,
+            _clip_text(reasoning_content, 800),
+        )
     if not cleaned_content and reasoning_content and not disable_thinking:
         retry_max_tokens = min(max(max_tokens * 2, RETRY_MIN_MAX_TOKENS), RETRY_MAX_MAX_TOKENS)
         logger.warning(

README.md

@@ -32,6 +32,8 @@ docker compose down
 
 Переменные окружения: задайте `BOT_TOKEN` в `.env`.
 Для ИИ-фич нужен `LLAMA_API_URL`.
+По умолчанию проект настроен на `https://mirror.porno4free.ru/zovos-ai/`.
+Если нужен лог reasoning/thinking модели, оставь `LLAMA_LOG_THINKING=1`; чтобы модель не резалась в режим без thinking, задай `LLAMA_FORCE_DISABLE_THINKING=0`.
 Для ставок нужен `ODDS_API_KEY` (бесплатно на https://the-odds-api.com).
 
 Контейнер хранит состояние в `./db`:

docker-compose.yaml

@@ -12,6 +12,9 @@ services:
       CHAT_HISTORY_DB_PATH: /db/chat_history.sqlite3
       POLYCHAETSI_STATS_PATH: /db/polychaetsi_stats.json
       BET_DB_PATH: /db/bets.sqlite3
+      LLAMA_API_URL: ${LLAMA_API_URL:-https://mirror.porno4free.ru/zovos-ai/}
+      LLAMA_FORCE_DISABLE_THINKING: ${LLAMA_FORCE_DISABLE_THINKING:-0}
+      LLAMA_LOG_THINKING: ${LLAMA_LOG_THINKING:-1}
     init: true
     restart: unless-stopped
 
@@ -30,6 +33,9 @@ services:
       POLYCHAETSI_STATS_PATH: /db/polychaetsi_stats.json
       BET_DB_PATH: /db/bets.sqlite3
       API_PORT: "8080"
+      LLAMA_API_URL: ${LLAMA_API_URL:-https://mirror.porno4free.ru/zovos-ai/}
+      LLAMA_FORCE_DISABLE_THINKING: ${LLAMA_FORCE_DISABLE_THINKING:-0}
+      LLAMA_LOG_THINKING: ${LLAMA_LOG_THINKING:-1}
     restart: unless-stopped
 
   webapp-frontend:

webapp/api.py

@@ -408,7 +408,14 @@ async def talk_api(
     user: dict = Depends(get_current_user),
 ):
     """Отправить сообщение ИИ."""
-    from AI.talk_handler import _generate_response, push_message
+    from AI.talk_handler import (
+        BOT_MEMORY_NAME,
+        EMPTY_RESPONSE_TEXT,
+        SYSTEM_PROMPT,
+        _generate_response,
+        _normalize_reply,
+        push_message,
+    )
 
     user_id = user["user_id"]
     user_name = user["first_name"] or "кент"
@@ -417,12 +424,27 @@ async def talk_api(
     chat_id = user_id
 
     try:
-        response = await _generate_response(chat_id, req.text, user_name)
-        if not response:
-            response = "Братуха, чёт базар не клеится."
-        await asyncio.to_thread(push_message, chat_id, "user", user_name, req.text)
-        await asyncio.to_thread(push_message, chat_id, "assistant", "бот", response)
-        return {"response": response}
+        await asyncio.to_thread(push_message, chat_id, "user", user_name, req.text, user_id=user_id)
+        response = await _generate_response(
+            chat_id,
+            system_prompt=SYSTEM_PROMPT,
+            user_id=user_id,
+            max_tokens=220,
+            temperature=0.8,
+            top_p=0.9,
+        )
+        normalized_response = _normalize_reply(response)
+        if not normalized_response:
+            normalized_response = EMPTY_RESPONSE_TEXT
+        await asyncio.to_thread(
+            push_message,
+            chat_id,
+            "assistant",
+            BOT_MEMORY_NAME,
+            normalized_response,
+            user_id=user_id,
+        )
+        return {"response": normalized_response}
     except Exception:
         logger.exception("Talk API failed")
         raise HTTPException(status_code=500, detail="LLM error")