No-frills offline meeting transcripts with context | Trace

Hey Product Hunt 👋 I'm the maker. I built Trace, a Mac app that records and transcribes your meetings entirely on-device. I built Trace for myself. I'd been using MacWhisper, but there was enough fiddling before each call that I'd forget to start it and walk out of an hour-long meeting with nothing written down. So the things I cared about most were that it's quick to activate and stays out of the way. You hit one global shortcut and a small bar appears at the bottom of your screen (there's also a keystroke to hide it entirely if you'd rather not see it while it's working). As it records your meeting you can flag anything important, with an optional note, as a "key moment". I built that because the "wait, that part matters" thought never survives to the end of a call. The key moments sit inline in the markdown at the right timestamp, so any AI you later paste the output into can see what mattered. It records your mic and the system audio as two separate tracks, then runs the system side through on-device diarization so you get who said what rather than one blended wall of text. Your mic lines are labelled as you, and the other voices come out as Speaker 1, Speaker 2 and so on (speaker renaming is planned for a future update). The transcription, and the live recap while you're recording, both run entirely on your machine, and you can choose between two engines. One is fast, using NVIDIA's Parakeet-TDT (Core ML, on the Neural Engine). The other is more accurate, using WhisperKit and Whisper large-v3-turbo, and holds up better on accents, jargon and quiet rooms. Both download their weights once and then run locally. There are no meeting bots to join the call. It just captures the audio your Mac is already playing out loud. Trace doesn't do any of the summarising itself, it just hands you clean markdown. On privacy, the app is sandboxed and your audio never leaves the Mac. The only network call it has to make is on first run, downloading the speech and speaker models (around 500MB) from Hugging Face, and after that you can go fully offline and it keeps working. The one other thing that ever touches the network is an optional Google Calendar connection for auto-naming sessions, which is read-only and stays off until you turn it on. You don't need to create an account, and we don't collect any telemetry. Trace is £9.99 once on the App Store, no subscription, and it needs macOS 14 or later on Apple Silicon since the models run on the Neural Engine. If you've used Granola, Otter or Fathom, the difference is they put a bot in your call or do the work in the cloud, sometimes both, and I wanted neither. I've been using it every day for months now and it's genuinely fixed the problem for me. Feedback very welcome, roasts included, and I'd love to hear whether it helps anyone else the way it helped me.