Local transcription
A warm Parakeet model runs on your Mac through Swift and CoreML, keeping speech recognition on your machine.
Khuluma
Local dictation for macOS
Press a hotkey, speak and paste into your Mac apps. Khuluma transcribes locally with Parakeet, with a menu-bar app for dictation and a CLI for audio files.
Why it exists
Khuluma lets me dictate into the apps I already use. Press a hotkey, speak and let a local model turn the recording into text. The menu-bar app handles recording and pasting; the CLI also transcribes audio files.
One vocabulary file keeps names and corrections consistent across dictation. Local history lets you find recent takes, while Stats shows delivered words and timings without using transcript text for its counts.
Install
Khuluma is built from source for macOS 14 or later. The app and CLI need to come from the same checkout. Setup uses Python, ffmpeg, Apple’s Swift toolchain and a local Parakeet model.
The repository README has the current CLI setup and manually signed app instructions. Existing Tmux Whisper or Lowkey installations should follow its migration guidance before replacing a working setup. The CLI is available through Homebrew with brew install ricardo-nth/tap/khuluma; the menu-bar app is built from source.
Khuluma is free and open source under the MIT licence.
What it does
Hotkey dictation
Press the hotkey to record, then press it again to transcribe and paste. Sending afterwards is optional.
Audio file transcription
Transcribe voice memos and other audio files with the CLI, or add a Finder Quick Action for selected files.
One shared vocabulary
Correct names and spellings across your dictation, with local filler and repeat cleanup.
Menu-bar controls
Check recording and model status, reload settings and choose Start at Login from Khuluma's menu.
Health checks
Inspect the CLI, configuration and local daemon with built-in doctor and status commands.