The AI-wearable category is crowded with closed, cloud-locked gadgets. Omi by BasedHardware is the exception worth paying attention to: a genuinely open-source AI wearable platform, MIT-licensed, where the firmware, hardware designs, apps, and backend all live in one public repository.
What It Is #
Omi is a small necklace-style wearable that captures conversations and pairs with apps on your phone and desktop. The platform transcribes in real time, generates summaries and action items, and gives you an AI chat that remembers everything it has heard — a searchable memory of your days, meetings, and ideas. Newer builds push further: the project now describes an assistant that “sees your screen, listens to your conversations, and tells you what to do.”
The part that sets it apart from the Humane/Rabbit wave of devices: everything is open. The mobile apps (Flutter), the desktop apps (Swift/Python on macOS, Node on Windows), the backend, the firmware, and the hardware designs are all in the repo under MIT. There’s a hosted Omi Cloud with paid plans if you want turnkey, but you don’t have to use it.
How It Fits Together #
The architecture is refreshingly legible:
- The Omi wearable captures audio and talks overBluetooth Low Energy (BLE) to your devices.
- Desktop apps (macOS Swift/Python, Windows Node) and amobile app (Flutter, iOS/Android) relay to the backend overHTTPS/WebSockets .
- The backend does the transcription, summarization, memory, and chat — and you canself-host it .
Getting the desktop app running from source is close to trivial. On macOS it’s a single clone-and-run against the public cloud config — no env files, no credentials, no local backend needed:
git clone https://github.com/BasedHardware/omi.git && cd omi/desktop/macos && ./run.sh --yolo
Windows is an npm install and npm run dev with the bundled example config. For serious work there’s a full installation path that spins up the complete backend stack locally with uv and the provided .env.template. There’s also a browser build at app.omi.me if you just want to poke at it.
Why Open Source Is the Whole Point Here #
An always-listening — and increasingly always-watching — device records the most sensitive stream of data you have. For a closed gadget, you take the vendor’s word on what happens to that stream. With Omi you can:
- Audit the firmware and backend to see exactly what’s captured, stored, and sent.
- Self-host the backend so transcripts and the resulting memory stay on infrastructure you control.
- Own the hardware design — the schematics are in the repo, so this is a platform to build on, not just a product to consume.
For an individual, that’s a privacy posture no closed competitor can match. For a team or organization — anyone handling regulated, proprietary, or customer data — keeping capture and memory off a third-party cloud isn’t a nicety, it’s often a hard requirement.
The Catch #
It’s a developer-forward platform, not a polished consumer appliance. The one-line installs are great, but the full self-hosted stack still asks you to manage a backend, keys, and updates. Battery life, transcription accuracy in noisy rooms, and the social reality of wearing a recording device are the usual open questions for this category. And always-on capture carries real consent and recording-law obligations — put a visible indicator on it and set clear policy before you wear it around others.
The Takeaway #
Omi is the most credible open-source answer to the AI-wearable moment: a necklace that captures, transcribes, and remembers — with the firmware, apps, and backend fully in your hands under MIT. If you want an AI that remembers your day and you want to own the data pipeline behind it, this is the one to clone first.
Explore it: github.com/BasedHardware/omi · omi.me · docs.omi.me.