services/ollama_monitor.py

import requests
import time
from utils.config import config

def check_ollama_status():
    """
    Checks if Ollama is running and which model is loaded.
    
    Returns:
        dict: {
            "running": True/False,
            "model_loaded": "mistral-7b" or None,
            "ngrok_url": "https://f943b91f0a0c.ngrok-free.app/",
            "local_url": "http://localhost:11434/"
        }
    """
    # Use the working ngrok URL from your logs
    ngrok_url = config.ollama_host
    local_url = "http://localhost:11434/"  # Always check localhost as fallback

    def _get_model_from_url(base_url, retries=3, delay=1):
        """Try to get model info with retry logic"""
        for attempt in range(retries):
            try:
                # Add headers to skip ngrok browser warning
                headers = {
                    "ngrok-skip-browser-warning": "true",
                    "User-Agent": "AI-Life-Coach-App"
                }
                response = requests.get(f"{base_url}/api/tags", timeout=10, headers=headers)
                if response.status_code == 200:
                    models = response.json().get("models", [])
                    if models:
                        return models[0].get("name")
                elif response.status_code == 404:
                    # Try alternative endpoint
                    response2 = requests.get(f"{base_url}", timeout=10, headers=headers)
                    if response2.status_code == 200:
                        # Server is running but might not have /api/tags endpoint
                        return "unknown-model"
            except Exception as e:
                if attempt < retries - 1:  # Don't sleep on last attempt
                    time.sleep(delay * (2 ** attempt))  # Exponential backoff
                continue
        return None

    # First try remote ngrok URL, then fall back to local
    remote_model = _get_model_from_url(ngrok_url)
    local_model = None
    
    if not remote_model:  # Only check local if remote failed
        local_model = _get_model_from_url(local_url)

    model_loaded = remote_model or local_model
    running = bool(model_loaded)

    return {
        "running": running,
        "model_loaded": model_loaded,
        "ngrok_url": ngrok_url,
        "local_url": local_url,
    }