YapToText privacy policy

YapToText is private by design. The short version: your voice, your words, and your data never leave your Mac.

What the app processes

  • Audio from your microphone is captured only while you are dictating, or while the mic route is kept warm for instant starts. Warm audio is analyzed for levels only and never stored. Recordings are transcribed on your Mac by on-device models, so dictation works with Wi-Fi off; the offline dictation page shows how to test that.
  • Transcripts are produced on device by the speech model bundled with the app, or by a model you downloaded into it.
  • Cleanup runs on device with the bundled language model. Your text is never sent to any server for processing.

What the app stores

  • History, meaning transcripts and optionally the audio of each dictation for playback, lives in the app's own sandboxed container on your Mac. You choose how much history is kept, can turn audio saving off, and can delete any entry. Deleting an entry deletes its audio. The History help page lists every retention setting. Turning Save dictation history off stops new entries; anything already saved stays until you delete it or it passes your auto-delete limit.
  • Audio during a dictation is also written to a crash-recovery file in the same container, so a dictation survives a crash. With Also save the audio recording on, which is the default, the file stays with its History entry; with it off, the file is deleted when the dictation ends.
  • Settings, modes, dictionaries, and commands are stored in the same container.
  • Nothing is synced, uploaded, or backed up by the app itself.

Network access

The app makes no network calls in normal use: no account, no analytics, no telemetry, no crash reporting of its own, and no tracking of any kind. If the app crashes, macOS may send Apple a crash log under your own analytics settings in System Settings; the app adds no crash SDK. The single exception to the rule is a model download that you start yourself on the AI Models page, which fetches the model file over HTTPS from Hugging Face. Nothing about you goes with that request. The App Store build ships both models inside, so even that download never happens unless you choose another model, and the AI Models help page covers those downloads. The app never checks for updates; updates come from the App Store.

Permissions

  • Microphone, to hear your dictation. Requested on first use.
  • Accessibility, optional. Used to type your words into other apps, to read the text around your cursor for Intelligent Insert, to read the text you have selected for Quick Edit, and for global shortcut keys. Without it, results are copied to the clipboard instead.

Tips

The tip jar is optional, and a tip changes nothing in the app. Apple processes the payment. The App Store listing's privacy answer is Data Not Collected.

Your data, your machine

Everything described above is inspectable: history is a plain JSON file inside the app's container, and the application's complete source code is public on GitHub. The privacy promise is not something you have to take on faith. It is checkable.

Contact

Questions or concerns: open an issue on the GitHub repository, or use the contact page at ryleighnewman.com. The person responsible for the app and this site is Ryleigh Newman.

Last updated: October 5, 2026.