Last updated: 11 May 2026
Dictator runs entirely on your Mac. The short version: your audio, your transcripts, and your settings never leave the device.
Audio captured from the microphone is held in memory for the duration of a recording and then discarded. The transcript is produced locally by WhisperKit or Parakeet, optionally tidied by a local language model running through MLX, and pasted into the focused app. Dictation history, conversations, vocabulary, custom prompts, and input-device preferences are stored in plain JSON files under ~/Library/Application Support/Dictator/ — only on your machine.
Two things, both initiated by you:
Model downloads. When you pick a Whisper or LLM model from Settings → Models, Dictator downloads the weights from Hugging Face. Hugging Face will see those requests; Dictator doesn't add anything to them.
Update checks. Dictator periodically asks dictator.robgough.net whether a newer version is available, via the Sparkle updater. The request includes your operating system version and the running app version — what any auto-updater needs to compare. The site is hosted on GitHub Pages; standard GitHub access logs apply. You can disable update checks in Settings.
Dictator does not send analytics, crash reports, or usage data to the author or anyone else. There is no account, no sign-in, and no server-side component beyond the static update feed.
macOS asks you to grant Microphone (to record) and Accessibility (to paste into the focused app via synthetic ⌘V) the first time you use the relevant feature. Both can be revoked at any time from System Settings → Privacy & Security.
Questions: hello@robgough.net.