Speakmend

Speakmend lives in the menu bar. Hold Control + Option + Space, speak, and let go: the words appear in a small panel while you talk and are typed at the cursor when you release. Or tap the shortcut once to talk hands-free and tap it again to finish; Esc cancels.

Then fix what you said without touching the keyboard. Say an edit on its own: “change Tuesday to Thursday”, “delete last word”, “scratch that”, or “undo that”. Anything else is typed as text. With Apple Intelligence ready on your Mac, you can turn on rewriting and say “rewrite that more formally” or “make that shorter”; “undo that” brings the original back.

Where your voice goes

Nowhere off your Mac. Speech is transcribed on the Mac, by Apple’s on-device speech model (the default, already installed on most Macs), by NVIDIA Parakeet, or by Whisper. Rewriting runs through Apple’s on-device model, never a cloud one. Audio is recorded only while you hold the shortcut, or until you tap it again, and is kept in memory and discarded after it is transcribed. Speakmend refuses to type into password fields.

It goes online for two things only. A Parakeet or Whisper model is downloaded when you choose one in Settings, from fixed addresses, and checked by size and SHA-256 before it is used. And it checks for updates once a day, which you can turn off in Settings; an update installs only when it carries the signature built into the app. There is no account, no telemetry and no cloud AI service.

What it needs

A Mac with Apple Silicon and macOS 26 or later. macOS asks for two permissions: the microphone, to hear you while you hold the shortcut, and Accessibility, to type into the app you are using and to read back what you dictated so it can be edited. Rewriting needs Apple Intelligence turned on and ready; without it, that switch stays off and says why.