QuickTranscript is a small macOS app for recording meeting audio from your microphone and transcribing it locally with MLX Whisper.
It was built for the pragmatic case: you join a Zoom/Meet call on laptop speakers, start QuickTranscript, and get a transcript saved to disk while the meeting runs.
- Lives in the macOS menu bar.
- Records microphone audio in 30-second chunks.
- Transcribes completed chunks locally with
mlx-whisper. - Appends text to
transcript.txtduring the meeting. - Keeps raw
.cafaudio chunks so a session can be re-transcribed later. - Shows recent recordings in a History menu.
brew tap rohitjavvadi/tap
brew install --cask quicktranscriptThe cask installs QuickTranscript.app and runs the local runtime setup script, so the MLX Whisper dependencies are installed as part of installation.
The app uses a local Python/MLX Whisper runtime stored at:
~/Library/Application Support/QuickTranscript/.venv
If the runtime is missing, QuickTranscript installs it automatically the first time you start recording. You can also install or repair it manually:
./scripts/setup_runtime.shThis downloads mlx-whisper and its dependencies.
If you installed the app directly, you can also run:
/Applications/QuickTranscript.app/Contents/Resources/setup_runtime.sh./package_app.shThe built app is created at:
dist/QuickTranscript.app
Install it locally:
ditto dist/QuickTranscript.app /Applications/QuickTranscript.app
open /Applications/QuickTranscript.appMaintainer release steps are in RELEASING.md.
- Join your meeting using laptop speakers.
- Open QuickTranscript.
- Click the menu bar icon.
- Choose
Start Recording. - Choose
Stop Recordingwhen done.
Output is written to:
~/Desktop/MeetingTranscripts/meeting-YYYY-MM-DD-HH-MM-SS/
The main file is:
transcript.txt
Start RecordingStop RecordingOpen TranscriptOpen Session FolderCopy Transcript PathHistory
History shows recent sessions with start time, audio chunk count, transcript size, and quick actions.
- QuickTranscript captures the microphone, not privileged system audio.
- For meeting transcription, use laptop speakers so the microphone hears the call audio.
- The default model is
mlx-community/whisper-tinyfor low latency. - To use another model, set
QUICK_TRANSCRIPT_MODELbefore launching from a shell.
The original terminal workflow still exists:
./start.shMIT