Private AI meeting notes that run locally: bot-free recording, live transcripts, summaries and Q&A. Free and open source.
- Language
- TypeScript
- License
- MIT
- Stars
- 989
Yogurt is a new desktop app from Y Combinator-backed Frigade that records your calls, transcribes them live and turns your notes and the conversation into a summary with action items. Everything, including the AI models, runs on your own machine. It's free, it's released under the MIT license, and it doesn't need an account or an API key. The repository was created on November 10. By early November 13 it had 989 stars, according to GitHub's API, and its Show HN post had 282 points about a day after it went up.
Why it matters: AI note-takers have become standard office software, but most of them send your meeting audio to someone's cloud or add a bot to the call. Yogurt, pitched as a free alternative to Granola, records without a bot and keeps recordings, transcripts and model inference on your own hardware, with source code you can audit.
What it does
According to the project site and the repository's feature list, as summarized by ScriptByAI and LavX News:
- No meeting bot. It records your microphone and your computer's audio, so other participants don't install anything.
- Call detection. It prompts you when a Zoom, Google Meet or Microsoft Teams call starts and stops automatically when the call ends. You can record other calling apps manually.
- Live transcript next to your notes, then a combined summary with decisions and follow-up tasks.
- Questions across meetings. You can ask about one meeting, a folder or your whole history, and answers link back to the source passages.
- Search, folders, audio replay and Markdown export.
- 99 languages when multi-language transcription is turned on.
- A local MCP server that lets an AI assistant you choose read your meeting records, only if you enable it.
How it works
| Piece | What Yogurt uses (per ScriptByAI and LavX) |
|---|---|
| Transcription | Whisper via whisper.cpp. Supported Macs can also use Apple's on-device SpeechAnalyzer |
| Summaries and Q&A | A local open-weight model run through llama.cpp, picked for your hardware from the Gemma 4, gpt-oss or Qwen3 families |
| Model size | About 5 GB for the starter notes model. Larger options (14 GB+) are available for capable machines |
| Network use | One-time model downloads and update checks only. ScriptByAI says the app collects no telemetry |
The site says transcription and summaries work offline once the models are downloaded, and that recordings, transcripts and notes are never uploaded.
Requirements
| Platform | Requirement (per the README and site) |
|---|---|
| macOS | Apple silicon (M1 or later), macOS 14.4+ |
| Windows | x64, Windows 10 version 2004+ |
| Linux | x64, Ubuntu 22.04 or similar, as an AppImage. ScriptByAI notes it needs PulseAudio or PipeWire and may need FUSE 2 |
| All | At least 8 GB of memory |
Installers for all three platforms are on the repo's Releases page, and the repo also documents how to build from source.
Limitations
- Some features are Mac-only. Calendar integration and speaker names, which are read from Zoom, Teams and Meet, only work on macOS for now.
- Windows builds aren't signed yet. The site says SmartScreen may warn you, and you have to choose "More info", then "Run anyway."
- A speed check can block setup. On Windows and Linux, setup runs a transcription speed test and stops if the machine is too slow.
- No accuracy data. LavX notes the repo publishes no benchmarks for noisy rooms, accents or different microphones. If you need verified transcripts, test it on your own calls first.
- Data stays after uninstalling. Removing the app leaves its data folder behind. A ZIP backup from Settings can be imported on any of the three platforms.
- Consent is your job. The site's FAQ advises telling other participants that you're recording, and recording laws vary by jurisdiction.
Who should try it
Yogurt is worth a look if you've wanted Granola-style notes but can't send meeting audio to a third party, for example because of client confidentiality, legal work or a strict IT policy. If you're on an 8 GB laptop, expect the bigger models to be off the table and test the speed check first. Developers who already use an MCP-capable assistant get a way to query their meetings locally without building their own pipeline.
Sources
- GitHub — FrigadeHQ/yogurt: Private AI meeting notes. Free and open source. github.com
- Yogurt — AI meeting notes that stay on your computer yogurt.so
- ScriptByAI — Yogurt: Free, Open-Source AI Meeting Notes (Granola Alternative) scriptbyai.com
- LavX News — FrigadeHQ releases Yogurt, an open-source app for local meeting notes news.lavx.hu