From 3b1cb85ba27cedef5f140a986da7e7b923a1eecf Mon Sep 17 00:00:00 2001
From: Antigravity <Antigravity@qpid.pro>
Date: Wed, 7 Oct 2026 15:30:36 +0300
Subject: [PATCH] fix(ai): configure OLLAMA_KEEP_ALIVE=5m to automatically
 release RAM after inactivity

---
 app/ai.py          | 8 ++++----
 app/config.py      | 1 +
 docker-compose.yml | 3 ++-
 requirements.txt   | 1 +
 4 files changed, 8 insertions(+), 5 deletions(-)

diff --git a/app/ai.py b/app/ai.py
index e596f52..bc3b7a8 100644
--- a/app/ai.py
+++ b/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,
diff --git a/app/config.py b/app/config.py
index f588f0f..77afa91 100644
--- a/app/config.py
+++ b/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")
 
diff --git a/docker-compose.yml b/docker-compose.yml
index d9001b1..5cf2d17 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -18,6 +18,7 @@ services:
       - DB_PATH=/app/data/notes.db
       - OLLAMA_BASE_URL=http://ollama:11434
       - OLLAMA_MODEL=qwen2.5:3b
+      - OLLAMA_KEEP_ALIVE=5m
     depends_on:
       - ollama
 
@@ -30,4 +31,4 @@ services:
     volumes:
       - ./data/ollama:/root/.ollama
     environment:
-      - OLLAMA_KEEP_ALIVE=24h
+      - OLLAMA_KEEP_ALIVE=5m
diff --git a/requirements.txt b/requirements.txt
index 6eb78c4..f78c4ad 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -16,3 +16,4 @@ typing-inspection==0.4.4
 typing_extensions==4.16.0
 uvicorn==0.53.0
 pypdf>=5.0.0
+pytest>=8.0.0