Meeting transcription that never leaves your Mac.

OpenOats records your calls, transcribes them on-device, and saves the transcript as a Markdown file on your machine. No cloud, no bot joining the call. No subscription.

1,180+
GitHub stars
MIT
Licensed, free forever
30
Languages
100%
Local transcription
OpenOats running during a call, showing a live transcript with relevant context pulled from the user's notes

Your audio stays on your machine

Everything runs on your Mac. Transcription, AI summaries, the whole thing. Nothing touches the internet unless you choose a cloud model over a local one.

Your transcripts are just files

Every recording becomes a .md file in a folder you control. Search them, back them up, feed them to whatever you want.

It stays out of your way

The app sits in your menu bar and hides from screen share. It detects when you open Zoom, Teams, or Slack and prompts you to start recording. No bot enters the meeting, nothing shows up for other participants.

What you do with this

Most people who say AI changed how they work are feeding their tools a lot of context. Transcripts of every meeting you've had in the last six months is a lot of context.

Record your calls for a few weeks. Claude can search across all of them. Cursor can pull up what you said in a planning session last Tuesday. That number a client mentioned offhand? Searchable.

Without a transcript, all of it disappears the moment the call ends.

Download for Mac

Your notes show up when you need them

Point OpenOats at a folder of notes or research docs. As the conversation moves, it pulls up what's relevant. You don't have to stop and go looking.

"How does it compare to Granola?"

Granola stores your data in their cloud and gives you access through their interface. OpenOats stores your data as files on your machine. You notice the difference when you want to do something the tool wasn't designed for.

"Is local transcription actually good?"

Yes. Local models have caught up. The engines OpenOats ships hit 3-5% word error rates on clear audio, close to what cloud services get. Qwen3 handles 30 languages with automatic detection, Parakeet TDT v3 covers 25 European languages. Both run on your Mac's GPU. When a better model comes out, you swap it in.

"Do I need to be technical?"

Download the app. Grant microphone access. Press record.

"Is it legal to record without telling people?"

Depends on where you are. Most US states and many countries allow one-party consent. Some require all parties to agree. Check your local laws.

"What hardware do I need?"

Any Mac with Apple Silicon (M1 or later). An M1 MacBook Air handles it fine.

"What if I stop using OpenOats?"

Your transcripts are Markdown files in a folder. They don't depend on OpenOats existing.

Under the hood

Transcription engines
Qwen3, Parakeet TDT v3, Whisper. CoreML on-device. Swappable in settings.
AI integration
Ollama for fully local by default. OpenRouter for cloud models (Claude, GPT, Gemini). Any OpenAI-compatible endpoint.
Knowledge base
Vector search with reranking. Surfaces relevant notes mid-call on a 90-second cooldown.
Output
Markdown transcripts plus session logs. AI meeting notes optional. Output goes to any folder you choose, including your Obsidian vault.
Stack
Swift 6.2, macOS 15+.

MIT licensed. Open source. Contributions welcome.

Every meeting you don't capture is context your AI will never have.