Meetily review: local transcription and its limits

Learn how Meetily records and summarizes meetings locally, why live transcripts lag, and which Whisper models the hosts recommend for stronger hardware.

Player not loading? Watch on YouTube

Andrew Warner and Nick McDonald review Meetily, an open source meeting note taker that they describe as keeping audio and transcripts on the user's computer. It captures meeting audio without a bot joining the call. Nick shows a timestamped transcript and summary from an hour-long team meeting, plus settings for choosing where recordings and text are saved.

The review compares Parakeet with Whisper. Nick runs Parakeet on a laptop without a GPU, but the hosts report missed words during fast speech. Andrew cites the project's requirements for Whisper medium or large V3: at least 16 GB of RAM and a discrete GPU or Apple Silicon. Summary generation uses separate models, including Gemma 3 in 4B and 1B sizes and Qwen. Optional API connections provide another processing route, so the local workflow depends on the settings chosen.

Meetily also transcribes existing video files through a feature described as beta. In the meeting example, the transcript has timestamps but no speaker labels, although the summary includes names. The live demonstration exposes a processing delay that the hosts find unsuitable for immediate captions.

The hosts discuss capturing audio from FaceTime, Slack and Discord. They also suggest pointing an AI agent at saved transcripts to extract tasks or retrieve past discussions. Their assessment favors meeting notes for later reference over live transcription.