What you say stays on your Mac

Summary

• No accounts, no sign-up, no analytics, no telemetry, no ads and no trackers, in the app or on this website.
• Speech recognition runs on your Mac. Audio is never uploaded and never kept.
• Everything SayWrite saves stays on your Mac, in folders only your user account can read.
• The app only connects to the internet to download its speech model when you ask, to check GitHub for updates (you can turn this off), and, if you enable it, to reach your own AI server on your Mac.

What SayWrite processes on your Mac

Your voice

While you dictate, audio is written to a temporary file in the app's temporary folder so it can be transcribed. It's deleted as soon as the text is ready, and also when you cancel or an error happens. Leftover temporary files are cleared when SayWrite starts. Audio is never saved anywhere else and never leaves your Mac.

Your text

Transcribed text is typed into the app you were using. If History is on, a copy is saved on your Mac with the time, the app it went to and its length, so you can search and reuse it. You choose how long History is kept and can clear it at any time. Text dictated into password managers is never saved to History, and spoken AI instructions (like “make this more concise”) aren't saved either.

Typing speed and stats

To compare your speaking and typing speed, SayWrite counts keystrokes and measures their timing in short bursts, along with the app you were typing in. It never records which keys you press or what you type. Filler-word counts are stored as numbers, without the text.

Everything else

Your settings, snippets, personal dictionary, dictionary suggestions, per-app rules, optional “How I write” note and saved file transcripts are stored on your Mac. File transcription keeps the text and timestamps, never the audio.

Where your data is stored

SayWrite stores its data in your user Library on this Mac: an Application Support folder for History, stats, file transcripts and dictionary suggestions (readable only by your user account), the app's preferences for settings, a Caches folder for the Neural Engine's compiled model, and the system temporary folder for recordings in progress. The speech model itself is kept in Application Support too. Nothing is synced to iCloud or any other service.

When SayWrite connects to the internet

SayWrite makes exactly three kinds of network connection, and none of them include anything you say or write:

1. Speech model download. When you click Download Model, SayWrite downloads NVIDIA Parakeet TDT 0.6B v2 (about 465 MB) from Hugging Face. Like any download, Hugging Face sees a normal web request, including your IP address. This happens once.
2. Update checks. While “Automatically check for updates” is on (the default), SayWrite checks for a new version once a day by fetching a small file from GitHub. It sends no system profile. If you accept an update, it downloads from GitHub. You can turn automatic checks off in Settings › General and check by hand instead.
3. Your own local AI (optional, off by default). If you turn on AI and choose Ollama or LM Studio, SayWrite sends the text to be rewritten to that server at 127.0.0.1, which is your own Mac. It refuses any other address, even if you type one in. Apple Intelligence's on-device model makes no network request at all.

Third parties

SayWrite doesn't share or sell data, because it doesn't collect any. Downloads come from GitHub and the speech model from Hugging Face, which handle those requests under their own privacy policies: GitHub and Hugging Face. GitHub shows us a total download count for each release, with no information about who downloaded it.

This website

saywrite.app is a static website. It sets no cookies and runs no analytics, tracking pixels or third-party scripts, and its fonts are served from this site. Our hosting provider may process standard request logs, such as IP address and browser type, to deliver the site and protect it from abuse.

Deleting your data

In the app, use History › Clear All (this also clears filler-word counts), clear typing data in Settings › History & Stats, and delete saved file transcripts in Files. Removing the app and its folders in your user Library deletes everything, including the speech model.

Children

SayWrite doesn't knowingly collect information from anyone, including children, because it collects no personal information at all.

Changes to this policy

If SayWrite ever changes how it handles data, this page will be updated before the change ships, with a new date at the top. Release notes are on the changelog.

Contact

Questions about privacy? Email hello@saywrite.app.