An open-source AI second brain that sees, hears, and remembers

What it does

Omi runs across your devices to record what you see and hear, transcribe speech in real time, and distil it into summaries, action items, and a persistent chat history. The project ships a macOS app built in Swift and Rust, a Flutter mobile app, a Python backend, and open-source wearable hardware including a necklace and camera-equipped glasses. Everything is MIT-licensed and designed to work as a continuous, cross-platform memory layer.

The interesting bit

Most AI memory tools stop at the browser; Omi goes all the way to the firmware, treating a Zephyr-powered wearable as a 24-hour audio capture mic and your macOS app as a screen-recording node. It is a vertically integrated life-logger that keeps the entire stack open from silicon to summary.

Key highlights

• Full-stack open source: Zephyr-based wearable firmware, ESP32-S3 smart glasses, Swift/Rust desktop app, Flutter mobile app, and Python/FastAPI backend.
• Real-time transcription and diarization pipeline backed by Deepgram, with GPU-accelerated VAD.
• Cross-platform memory layer that claims 300,000+ professional users.
• Extensible via SDKs for Python, Swift, and React Native, plus an MCP server for Model Context Protocol integration.
• Hardware designs and buying guides are published, so you can build the capture device yourself instead of buying it.

Verdict

Developers interested in open hardware, real-time transcription pipelines, or owning your AI memory stack should look here; if you just want a SaaS notetaker, the hardware assembly and multi-repo setup are likely overkill.