Max Headbox is a compact, truly local, voice-activated LLM agent designed for Raspberry Pi 5. It enables users to configure and execute a variety of tools and actions via voice commands, offering a private and customizable AI assistant experience on low-power hardware. The project aims to provide a functional, on-device AI agent with a focus on accessibility, extensibility, and user control over AI interactions.

How It Works

The agent leverages Ollama for running local LLMs (e.g., Gemma 3:1b, Qwen 3:1.7b) and Vosk for robust wake-word detection, with faster-whisper handling efficient speech-to-text transcription. Its architecture features a JavaScript frontend managing agent logic and tool invocation, while a backend composed of Ruby (Sinatra) and Python handles hardware interactions and API routes. Tools are implemented as JavaScript modules, allowing for straightforward customization and extension of the agent's capabilities, with backend tools potentially using Ruby or Python.

Quick Start & Requirements

• Hardware: Raspberry Pi 5 (8GB/16GB recommended), microphone. Optional: GeeekPi Screen/Case/Cooler. Requires ~6GB RAM for LLMs.
• Software: Ruby 3.3.0, Node 22, Python 3, Ollama.
• Installation: Clone repo, npm install, cd backend/, bundle install, pip3 install -r requirements.txt. Pull necessary Ollama models (`ollama pull gemma3: