This project provides a completely offline voice assistant for macOS, leveraging Ollama for LLM inference (Mistral 7b) and Whisper for speech recognition. It's designed for users seeking a private, local AI assistant experience.

How It Works

The assistant integrates Ollama's Mistral 7b model for natural language understanding and response generation, coupled with OpenAI's Whisper for accurate speech-to-text transcription. It processes user voice input, sends it to the LLM via Ollama, and then uses text-to-speech with macOS's built-in system voices to provide spoken responses.

Quick Start & Requirements

• Install Ollama and pull the mistral model (ollama pull mistral).
• Download an OpenAI Whisper model (e.g., base.en) and place it in a /whisper directory.
• Install Homebrew and run brew install portaudio for PyAudio Apple Silicon support.
• Clone the repository and run pip install -r requirements.txt.
• Start the assistant with python assistant.py.
• Requires Python 3.x, Pip, Ollama, and a Whisper model.

Highlighted Details

• Offline operation for enhanced privacy.
• Utilizes Mistral 7b via Ollama for local LLM inference.
• Integrates OpenAI Whisper for speech recognition.
• Supports macOS system voices for text-to-speech, with premium voices like "Zoe (Premium)" recommended for improved quality.
• Configuration for other languages is available via assistant.yaml.

Maintenance & Community

This project builds upon the work of maudoin. Further community or maintenance details are not specified in the README.

Licensing & Compatibility

The repository's license is not explicitly stated in the README. Compatibility for commercial use or closed-source linking is not detailed.

Limitations & Caveats

The project is specifically for macOS. While it mentions improvements over a previous version, specific performance benchmarks or known issues are not detailed. The README implies that higher quality TTS requires downloading premium macOS system voices.