FrigadeHQ/yogurt

What it does

Yogurt is a desktop app for recording calls without joining the meeting as a bot, then turning the recording into transcripts and AI-generated meeting notes. It runs speech-to-text on the device with llama.cpp and Whisper, so meeting audio and notes stay local instead of being uploaded to a service. The README describes the output as a summary with the plan, next steps by person, and personal reminders.

It targets people who take calls regularly and want private, local notes without another subscription, including those moving off hosted meeting-note products. The desktop build is Electron-based TypeScript, and the project is MIT licensed.

Getting started

1. Check your platform: macOS 14.4+ on Apple Silicon, Windows 10 2004+ on x64, or Ubuntu 22.04+ or similar with an x64 AppImage. The README lists 8 GB of memory as a requirement.
2. Download the build for your system: Mac Apple Silicon, Windows x64 or Linux x86_64 AppImage.
3. Start the app, grant the recording/microphone permissions it asks for, and begin a call. Yogurt records the call locally rather than sending a bot into the meeting.
4. After the call, let it transcribe and summarize the audio into notes.

There is no documented CLI or scripted first-run flow. To build it yourself, the README points to the build-from-source guide in docs/building.md (linked from the README); the README excerpt does not include those steps, so follow that guide rather than guessing commands. The project site is yogurt.so.

Constraints

• The download builds are precompiled desktop apps, not a hosted service; there is no web version described for taking notes.
• Platform support is narrow: Apple Silicon only on macOS, x64 only on Windows and the Linux AppImage; 8 GB memory is listed as the minimum.
• Transcription and summarization run locally, so output speed and quality depend on the machine's CPU/GPU and the local models shipped with the app; the README does not give benchmark timings.
• Recording calls may require consent from other participants depending on your jurisdiction; the project only says it does not add a bot, not how it handles consent.

HotRepo's take

The project appeared in HotRepo's new-project snapshot on 2026-11-11 with 114 stars gained and 2 forks in the recorded window, placing it in the TypeScript new list. A Show HN thread the same day drew 50 points and 28 comments, with one commenter asking how it differs from Muesli and OpenOats (discussion) and another questioning why so many similar transcription-and-summary apps exist (comment). That discussion suggests the category is crowded and the comparison question is still open; the repository's own materials do not answer it, so evaluate the local-only recording design against those alternatives yourself.

The strongest concrete differentiator in the supplied material is the bot-free local recording with on-device transcription and summarization, but the binaries and their exact relation to the source are not documented here, which is why the repository is flagged for review.