fix(ai): configure OLLAMA_KEEP_ALIVE=5m to automatically release RAM after inactivity

app/ai.py

@@ -251,7 +251,7 @@ async def translate_sentence_ai(
         "prompt": prompt,
         "stream": False,
         "format": "json",
-        "keep_alive": "24h",
+        "keep_alive": config.OLLAMA_KEEP_ALIVE,
         "options": {
             "temperature": 0.2,
             "num_predict": 300,
@@ -367,10 +367,10 @@ async def warmup_ollama_model() -> bool:
         client = get_ollama_client()
         resp = await client.post(
             f"{config.OLLAMA_BASE_URL}/api/generate",
-            json={"model": config.OLLAMA_MODEL, "keep_alive": "24h"},
+            json={"model": config.OLLAMA_MODEL, "keep_alive": config.OLLAMA_KEEP_ALIVE},
         )
         if resp.status_code == 200:
-            logger.info(f"Ollama model '{config.OLLAMA_MODEL}' warmed up successfully (keep_alive=24h).")
+            logger.info(f"Ollama model '{config.OLLAMA_MODEL}' warmed up successfully (keep_alive={config.OLLAMA_KEEP_ALIVE}).")
             return True
         else:
             logger.warning(f"Ollama warmup returned HTTP {resp.status_code}: {resp.text}")
@@ -567,7 +567,7 @@ async def generate_lesson_ai_debrief(
         "prompt": prompt,
         "stream": False,
         "format": "json",
-        "keep_alive": "24h",
+        "keep_alive": config.OLLAMA_KEEP_ALIVE,
         "options": {
             "temperature": 0.3,
             "num_predict": 450,

app/config.py

@@ -43,4 +43,5 @@ DB_PATH.parent.mkdir(parents=True, exist_ok=True)
 
 OLLAMA_BASE_URL = os.getenv("OLLAMA_BASE_URL", "http://127.0.0.1:11434").rstrip("/")
 OLLAMA_MODEL = os.getenv("OLLAMA_MODEL", "qwen2.5:3b")
+OLLAMA_KEEP_ALIVE = os.getenv("OLLAMA_KEEP_ALIVE", "5m")