Ollama Add-on

Run Large Language Models (LLMs) locally on your Home Assistant hardware with complete privacy and no cloud dependency.

Features

• 🔒 Complete Privacy: All processing happens locally
• 🚀 Multiple Models: Support for Llama, Mistral, Phi, and more
• 🎯 REST API: Easy integration with Home Assistant and other services
• 💾 Model Management: Download, update, and remove models easily
• ⚡ GPU Support: Acceleration for supported hardware

Installation

1. Add the J0rsa repository to your Home Assistant
2. Search for “Ollama” in the Add-on Store
3. Click Install and wait for the download to complete
4. Configure the add-on (see Configuration below)
5. Start the add-on

Configuration

# Example configuration
gpu_support: false  # Enable if you have compatible GPU
models_path: /data/models  # Where to store models
api_host: 0.0.0.0  # API listening address
api_port: 11434  # API port

Configuration Options

Usage

Downloading Models

After starting the add-on, you can download models using the API:

curl http://homeassistant.local:11434/api/pull -d '{
  "name": "llama2"
}'

Popular Models

• llama2: Meta’s Llama 2 model (7B parameters)
• mistral: Mistral AI’s 7B model
• phi: Microsoft’s Phi-2 model (2.7B)
• codellama: Specialized for code generation
• neural-chat: Intel’s conversational model

Running Inference

Send a prompt to the model:

curl http://homeassistant.local:11434/api/generate -d '{
  "model": "llama2",
  "prompt": "Why is the sky blue?"
}'

Integration with Home Assistant

You can integrate Ollama with Home Assistant using REST commands:

rest_command:
  ask_ollama:
    url: "http://localhost:11434/api/generate"
    method: POST
    headers:
      Content-Type: "application/json"
    payload: '{"model": "llama2", "prompt": ""}'

Hardware Requirements

Minimum Requirements

• CPU: 4 cores recommended
• RAM: 8GB minimum (16GB recommended)
• Storage: 10GB+ depending on models

Model Size Guidelines

• 7B models: ~4GB RAM
• 13B models: ~8GB RAM
• 30B models: ~16GB RAM

API Documentation

Endpoints

Example: List Models

curl http://homeassistant.local:11434/api/tags

Tips and Tricks

1. Model Selection: Start with smaller models (7B) and upgrade if needed
2. Performance: Enable GPU support if you have compatible hardware
3. Storage: Models are stored persistently in /data/models
4. Memory: Monitor RAM usage, especially with larger models
5. API Key: Consider adding authentication for external access

Troubleshooting

Add-on Won’t Start

• Check logs for error messages
• Ensure sufficient RAM is available
• Verify port 11434 is not in use

Model Download Fails

• Check internet connectivity
• Ensure sufficient storage space
• Try downloading a smaller model first

Slow Performance

• Consider using smaller models
• Enable GPU acceleration if available
• Close other resource-intensive add-ons

Support

• GitHub Issues
• Ollama Documentation
• Community Forum