alexkapadia/omni

Omni is a meeting assistant for Windows that runs entirely on your own computer instead of sending your calls to someone else's cloud. It does not join your meetings as a visible bot. Instead it captures the other participants' audio and your own microphone separately, using your computer's built in audio loopback, so it still works properly even if you are wearing headphones. Everything is transcribed on your own device using a speech recognition model that runs on your graphics card, and by default the raw audio is thrown away right after the words are extracted, unless you choose to keep it. Your rough handwritten notes taken during a meeting are then combined with that transcript into a cleaner, more complete set of notes, with your own words kept untouched and the added text placed clearly around them so nothing you wrote is silently changed. Omni treats an Obsidian vault, a folder of your own markdown notes, as its single source of truth. You can ask it questions about your meetings and past notes, and it answers using only that vault, always pointing back to the exact note and line where the answer came from. It can also suggest actions like calendar events, contact updates, or draft emails, but it never sends anything on its own. Every suggested action needs your explicit approval, and every action that does run gets written to a permanent log you can review later. There is also a voice mode called Naomi that lets you talk to the assistant hands free and get answers pulled from your vault the same way. Under the hood, a React and Tauri interface talks to a separate Python program doing all the real work, including transcription, note search, and calls to AI providers like Groq, Gemini, and optionally Claude, whichever you supply your own API keys for. The project is open source under the MIT license.