omi

Omi functions as a personal AI 'second brain' by continuously capturing user screen activity and ambient audio conversations. It uses real-time transcription (via Deepgram), voice activity detection (VAD), speaker diarization, and large language models to generate summaries, extract action items, and maintain context-aware chat history. The system runs across macOS, Windows, iOS, Android, and custom wearable hardware (Omi Wearable and Omi Glass). The backend is Python-based with FastAPI, Firebase, Redis, and GPU-accelerated audio processing. Mobile apps are built with Flutter; macOS app combines Swift/SwiftUI with a local Python backend. Open-source SDKs support multi-language device integration via BLE and WebSocket protocols. Hardware designs are published for DIY assembly. The project includes documentation for development, API usage, app building, and hardware flashing. It is licensed under MIT and trusted by over 300,000 professionals.