Yogurt: Open-Source Local AI Meeting Notes and Transcripts
GitHub Open Source
Record private AI meeting notes entirely on your machine without sending audio to the cloud.
Repository snapshot
- Repository: FrigadeHQ/yogurt
- Language: TypeScript
- License: MIT
- Stars: 1 493
- Forks: 81
Daily open-source picks: GitHub Open Source on Telegram
What it is
Yogurt is an open-source, local-first meeting assistant designed to record and summarise your calls without routing data through third-party cloud services. Built as an alternative to tools like Granola, it captures your microphone and system audio directly, ensuring that your raw audio, transcripts, and notes remain on your machine.
Why it matters
Most meeting assistants rely on sending sensitive audio streams to external servers, creating privacy and compliance hurdles for engineering teams. Yogurt shifts the entire pipeline to your local hardware. By leveraging local transcription and open-weight models, it gives you searchable transcripts and action items without compromising data privacy.
Key features
- Captures system audio and microphone feeds directly without adding a bot participant to your calls.
- Transcribes speech locally using whisper.cpp and Apple SpeechAnalyzer on supported hardware.
- Summarises conversations offline using local open-weight models via llama.cpp.
- Integrates with local assistant apps through an optional Model Context Protocol server.
Quick start
git clone https://github.com/FrigadeHQ/yogurt.gitWho should try it
Developers, remote workers, and privacy advocates who want automated meeting transcripts and summaries but cannot send proprietary discussions to external cloud platforms.
Verdict
Yogurt delivers a practical local-first alternative for meeting notes. While some platform-specific integrations like calendar syncing and speaker detection are currently limited to macOS, its core offline capabilities make it a strong addition to a privacy-focused developer setup.
Continue exploring
Get the next pick in Telegram: Follow GitHub Open Source
Curated from the repository README and metadata. No hands-on test is claimed; evaluate it in a sandbox before production use.