README.md

metadata

title: AI Life Coach
emoji: 🧘
colorFrom: purple
colorTo: blue
sdk: streamlit
sdk_version: 1.24.0
app_file: app.py
pinned: false

AI Life Coach 🧘

Your personal AI-powered life coaching assistant.

Features

• Personalized life coaching conversations
• Redis-based conversation memory
• Multiple LLM provider support (Ollama, Hugging Face, OpenAI)
• Dynamic model selection
• Remote Ollama integration via ngrok
• Automatic fallback between providers

How to Use

1. Select a user from the sidebar
2. Configure your Ollama connection (if using remote Ollama)
3. Choose your preferred model
4. Start chatting with your AI Life Coach!

Requirements

All requirements are specified in requirements.txt. The app automatically handles:

• Streamlit UI
• FastAPI backend (for future expansion)
• Redis connection for persistent memory
• Multiple LLM integrations

Environment Variables

Configure these in your Hugging Face Space secrets or local .env file:

• OLLAMA_HOST: Your Ollama server URL (default: ngrok URL)
• LOCAL_MODEL_NAME: Default model name (default: mistral)
• HF_TOKEN: Hugging Face API token (for Hugging Face models)
• HF_API_ENDPOINT_URL: Hugging Face inference API endpoint
• USE_FALLBACK: Whether to use fallback providers (true/false)
• REDIS_HOST: Redis server hostname (default: localhost)
• REDIS_PORT: Redis server port (default: 6379)
• REDIS_USERNAME: Redis username (optional)
• REDIS_PASSWORD: Redis password (optional)

Provider Details

Ollama (Primary Local Provider)

Setup:

1. Install Ollama: https://ollama.com/download
2. Pull a model: ollama pull mistral
3. Start server: ollama serve
4. Configure ngrok: ngrok http 11434
5. Set OLLAMA_HOST to your ngrok URL

Advantages:

• No cost for inference
• Full control over models
• Fast response times
• Privacy - all processing local

Hugging Face Inference API (Fallback)

Current Endpoint: https://zxzbfrlg3ssrk7d9.us-east-1.aws.endpoints.huggingface.cloud

Important Scaling Behavior:

• ⚠️ Scale-to-Zero: Endpoint automatically scales to zero after 15 minutes of inactivity
• ⏱️ Cold Start: Takes approximately 4 minutes to initialize when first requested
• 🔄 Automatic Wake-up: Sending any request will automatically start the endpoint
• 💰 Cost: $0.536/hour while running (not billed when scaled to zero)
• 📍 Location: AWS us-east-1 (Intel Sapphire Rapids, 16vCPUs, 32GB RAM)

Handling 503 Errors: When using the Hugging Face fallback, you may encounter 503 errors initially. This indicates the endpoint is initializing. Simply retry your request after 30-60 seconds, or wait for the initialization to complete (typically 4 minutes).

Model: OpenAI GPT OSS 20B (Uncensored variant)

OpenAI (Alternative Fallback)

Configure with OPENAI_API_KEY environment variable.

Switching Between Providers

For Local Development (Windows/Ollama):

1. Install Ollama:

   # Download from https://ollama.com/download/OllamaSetup.exe
   

Pull and run models:

ollama pull mistral ollama pull llama3 ollama serve Start ngrok tunnel:

ngrok http 11434 Update environment variables:

OLLAMA_HOST=https://your-ngrok-url.ngrok-free.app LOCAL_MODEL_NAME=mistral USE_FALLBACK=false For Production Deployment: The application automatically handles provider fallback:

Primary: Ollama (via ngrok) Secondary: Hugging Face Inference API Tertiary: OpenAI (if configured) Architecture This application consists of:

Streamlit frontend (app.py) Core LLM abstraction (core/llm.py) Memory management (core/memory.py) Configuration management (utils/config.py) API endpoints (in api/ directory for future expansion) Built with Python, Streamlit, FastAPI, and Redis.

Troubleshooting Common Issues: 503 Errors with Hugging Face Fallback: Wait 4 minutes for cold start initialization Retry request after endpoint warms up Ollama Connection Issues: Verify ollama serve is running locally Check ngrok tunnel status Confirm ngrok URL matches OLLAMA_HOST Test with test_ollama_connection.py Redis Connection Problems: Set USE_FALLBACK=true to disable Redis requirement Or configure proper Redis credentials Model Not Found: Pull required model: ollama pull Check available models: ollama list Diagnostic Scripts: Run python test_ollama_connection.py to verify Ollama connectivity. Run python diagnose_ollama.py for detailed connection diagnostics.