Your voice stays yours.

Whisper is designed to process speech locally on your Mac. It does not require an account and does not include an analytics or advertising SDK.

Microphone audio and transcripts

Whisper captures microphone audio only while dictation is active. Speech recognition runs through models stored on your Mac. The app does not upload dictation audio or resulting transcripts to a Whisper-operated server.

Model downloads

During setup, Whisper downloads the speech models you select from their model distribution source. Once cached, the default NVIDIA Parakeet transcription model and Supertonic 3 voice model can run offline.

Accessibility permission

Whisper uses macOS Accessibility permission to paste a completed transcript into the active app and, if enabled, press Return. If you do not grant that permission, the transcript is copied to the clipboard instead.

Claude Code integration

The optional Claude Code hook sends a short response summary from Claude Code to a local HTTP endpoint on your own Mac. Whisper can read that summary aloud using its local text-to-speech model. The hook is not exposed as a cloud service.

Local files

Preferences and diagnostic logs are stored in the app's macOS application-support directory. Downloaded model files are cached locally by the QVAC SDK. Removing the app does not automatically remove model caches.

Questions

For privacy questions or deletion guidance, visit Whisper Support.

Last updated September 19, 2026.