Antrum

Local-first meeting transcription for macOS. Record or import audio, process it on-device, repair speaker labels, then export clean Markdown.

Status

Source-only project. No signed or notarized macOS binaries are published yet. Build locally from source.

What it does

• Imports .mp3, .wav, .m4a, .mp4, and .mov files.
• Records microphone and meeting audio as separate local tracks on macOS 14.2 or later.
• Transcribes locally with FluidAudio / Parakeet.
• Labels remote speakers with NVIDIA Sortformer v2.1, with Pyannote Core ML fallback.
• Lets you rename or merge speakers before copying or saving Markdown.
• Provides optional local dictation from the menu bar.

Requirements

• macOS 14.2 or later for live system-audio capture.
• Xcode and Xcode Command Line Tools.
• XcodeGen: brew install xcodegen.

Build and run

make doctor
make test
make run

make build creates an unsigned Debug build. make test generates the Xcode project from project.yml before running tests.

Models and privacy

Antrum asks before downloading local ASR and diarization models from FluidAudio/Hugging Face. Audio, transcript text, speaker labels, and calendar details are not uploaded by Antrum.

Model weights are not included in this repository. Review their upstream terms before downloading. See THIRD_PARTY_AND_MODEL_NOTICES.md.

Recording is manual and consent-forward. Antrum does not join calls, inspect browser tabs, or start recording automatically. You are responsible for obtaining permission from all meeting participants before recording.

See PRIVACY.md for storage and permission details.

Contributing

Read CONTRIBUTING.md. Run these checks before opening a pull request:

make lint
make secret-scan
make test

Report vulnerabilities privately through GitHub private vulnerability reporting. Use Issues for bugs, support, and feature requests.

License

Licensed under the Apache License 2.0.