Getting started

Download & install

Download MiniWhisper, open the DMG, and drag the app into your Applications folder. Or install with Homebrew:

brew install --cask andyhtran/tap/miniwhisper

MiniWhisper requires macOS 14 (Sonoma) or later.

First launch

Launch MiniWhisper from Applications or Spotlight. It asks for two permissions — grant both:

• Microphone — to capture your voice
• Accessibility — to paste transcribed text into other apps

Look for the waveform icon in your menu bar, at the top-right of your screen. The default Parakeet model loads on first use — it downloads once, then stays cached on your Mac.

Your first transcription

1. Click into any text field — a browser, an editor, a chat app
2. Press ⌥ W to start recording
3. Speak naturally
4. Press ⌥ W again to stop
5. Your words are pasted where your cursor is

Shortcuts

Change the toggle shortcut from the menu bar panel — click the shortcut field and press a new combination. The bare Fn key works too.

Features

Auto-paste

Transcriptions go straight into the app you're using. MiniWhisper saves your clipboard, copies the text, sends a paste keystroke, and then restores whatever was on your clipboard before — so a copied link or image is not lost to a dictation.

Models

Switch models from the Model button in the menu bar panel:

• Parakeet (default) — very fast, on-device, English plus 20 European languages. Best for most people.
• Whisper — multilingual with automatic language detection, on-device via whisper.cpp. Use it for languages Parakeet doesn't cover.
• Custom — an OpenAI-compatible transcription endpoint you supply, with your own API key. For advanced setups only; this is the one mode where audio leaves your Mac.

Text replacements

Auto-correct words or phrases after transcription — fix a name the model keeps mishearing, or expand a shorthand. Add find → replace rules from the Replace button in the menu bar panel. Spoken symbols are built in: saying "comma" or "new line" produces the symbol.

History & stats

The History panel keeps your recent transcriptions so you can re-copy one at any time. The panel header tracks your usage: recordings, total speaking time, words transcribed, and average words per minute. Everything is stored locally.

Edit mode

An optional second shortcut that rewrites a transcription with an AI model before pasting — useful for cleaning up rambling dictation into tidy prose. It can use your existing Claude Code or Codex CLI login, or a custom OpenAI-compatible endpoint. Entirely opt-in and off by default.

CLI

MiniWhisper ships with miniwhispercli, a command-line companion that transcribes audio and video files with the same on-device models. Install it from MiniWhisper's settings (the MiniWhisper CLI row) — it lands in ~/.local/bin.

miniwhispercli transcribe recording.mp3   # transcribe a file
miniwhispercli diarize meeting.wav        # label who spoke when
miniwhispercli models status              # show local model readiness

Watch it in action:

Help

Can't find the menu bar icon?

macOS hides menu bar icons that don't fit near the clock. If you run many menu bar apps, MiniWhisper's icon may be pushed out of view.

• Hold ⌘ and drag the MiniWhisper waveform icon closer to the clock to keep it visible
• Open System Settings → Control Center → Menu Bar Only and hide icons you don't need
• In MiniWhisper's settings, click Open Menu Bar Settings to jump there directly

Something else? Open an issue on GitHub.