Yogurt is a free, open-source alternative to Granola for private meeting notes

What happened

Frigade has released Yogurt, a free and open-source application for AI meeting notes, presenting it as a private alternative to Granola. The project was featured on Hacker News on November 11, and its source code is publicly available on GitHub under the MIT license.

According to the project's README, the app captures your conversations directly on your device rather than joining the call as an automated participant. That distinction matters: many competing notetakers send a bot into the meeting that appears in the participant list and can be blocked by meeting software or policies.

Platform support and requirements

Yogurt ships as a desktop application for all three major operating systems. The documented requirements are macOS 14.4 or newer on Apple Silicon, Windows 10 version 2004 or newer on x64, and Ubuntu 22.04 or a comparable distribution via a 64-bit AppImage. The repository lists 8 GB of memory as the baseline. Prebuilt downloads are available through the project's website, yogurt.so, and the README links to instructions for building and running the app locally.

A local-first design

The repository's topic tags give a clear picture of the technical approach. Alongside "meeting-notes" and "privacy," the project lists "electron," "whisper," "llama-cpp," "local-ai," "transcription" and "mcp." Those tags indicate the app is an Electron build that leans on Whisper for speech-to-text and llama.cpp for running a language model on the user's own hardware, with support for the Model Context Protocol, the emerging standard for connecting AI tools to external data sources.

While the README itself is terse, the combination of local transcription, local model inference and a stated focus on privacy suggests the full pipeline — audio capture, transcription and summarisation — is intended to run on the user's machine rather than in a hosted cloud service. The 8 GB memory requirement is consistent with on-device model inference.

Project maturity

The repository snapshot shows an early-stage project: 228 stars, 4 forks and 74 commits, with Yogurt developed by Frigade. Anyone evaluating it for sensitive calls should treat it as a young codebase, though the MIT license means it can be audited, forked and self-maintained in a way proprietary notetakers cannot.

Why it matters

AI meeting notetakers have become popular, but the dominant products raise two recurring concerns. First, they typically route audio and transcripts through a vendor's cloud, which is a problem for confidential sales calls, legal discussions or internal strategy sessions. Second, the bot-based approach inserts a visible participant into every meeting, which some organisations prohibit outright.

Yogurt addresses both objections by design: no bot joins the call, and the evidence in the repository points to processing that stays on the local machine. Add the MIT license and a zero-cost price tag, and it becomes a credible option for developers and teams who want meeting notes without a subscription or a data-handling agreement. It also joins a growing set of tools showing that useful AI transcription and summarisation no longer require a round trip to someone else's servers — a trend with implications well beyond meeting notes.