hark: macOS CLI that captures any app's audio and transcribes it with whisper.cpp

[PhantomYdn]

I built hark (https://github.com/PhantomYdn/hark, MIT, Swift) because every "record system audio on a Mac" answer ends with "install BlackHole". It uses the macOS 14.4 Core Audio process-tap API to capture the mic, all system audio or specific apps (or everything except some), optionally mixed with the mic for meetings, and transcribes with whisper.cpp as the default engine: whisper-cli for files, whisper-server kept resident for live transcription so text streams while you talk.

brew tap PhantomYdn/hark https://github.com/PhantomYdn/hark && brew install hark   # pulls whisper-cpp
hark models download base.en --default
hark --system --mix -t -                      # live captions of a call on stdout
hark -i lecture.m4a -t lecture.srt            # file → subtitles
hark --system --mix -t - | fabric-ai -p summarize_meeting

Other engines (Apple Speech, WhisperKit, Parakeet) are selectable, but whisper.cpp is what it ships with and the only one that also runs on Intel. Speaker labels: You/Others from the mix, Speaker N from CoreML embeddings. Signed + notarized binary, no telemetry, no network except model downloads.

Would appreciate feedback on the live segmentation (VAD-cut segments to whisper-server); true streaming is on the list.