diff --git a/AI/talk_handler.py b/AI/talk_handler.py
index 4f94380..48e7cb4 100644
--- a/AI/talk_handler.py
+++ b/AI/talk_handler.py
@@ -19,6 +19,7 @@ LLAMA_API_URL = os.getenv("LLAMA_API_URL", "https://mirror.porno4free.ru/zovos-a
 DB_PATH = os.getenv("CHAT_HISTORY_DB_PATH") or str(Path(__file__).resolve().with_name("chat_history.sqlite3"))
 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"}
 
 RECENT_MESSAGES_LIMIT = 14
 SUMMARY_TRIGGER_MESSAGES = 24
@@ -313,8 +314,9 @@ async def _call_llm(
     max_tokens: int,
     temperature: float,
     top_p: float,
-    disable_thinking: bool = False,
+    disable_thinking: bool | None = None,
 ) -> str:
+    disable_thinking = FORCE_DISABLE_THINKING if disable_thinking is None else disable_thinking
     url = f"{LLAMA_API_URL.rstrip('/')}/v1/chat/completions"
     payload = {
         "messages": messages,