Your voice stays yours.

The short version

• Your audio is captured, transcribed, and discarded entirely on your Mac. It is never uploaded, and we never hear it.
• There are no accounts, no analytics, and no ads, in the app or on this website.
• The only personal data we hold is an email address you choose to give us for the Windows waitlist.
• Diagnostic reports leave your Mac only when you press Send, and they never contain your audio.

Your voice and transcripts

When you press the hotkey, UltraTalk captures audio from your microphone and transcribes it with open-source speech models running on your Mac’s own hardware. The audio exists in memory while it is being transcribed, then it is gone. Nothing is sent to a server, because there is no server to send it to.

Your transcript history is stored locally on your Mac, in the app’s Library. You control how long it is kept (the latest 100 items, a number of days, or forever) and you can delete any entry, or all of them, at any time. We have no access to your transcripts and no way to obtain them. The one exception is a diagnostic report you choose to share: its logs can contain short excerpts of dictated text, as described below.

What stays on your Mac

• Your transcripts and any recordings you choose to keep.
• Your settings, custom dictionary, and modes.
• Vocabulary UltraTalk learns from your dictation and corrections, including pending suggestions.
• The speech models you download.

All of it lives in your user folder, under your control, protected by your Mac’s own security.

Vocabulary learning

UltraTalk can suggest words for your custom dictionary. Vocabulary learning is on by default, including when you update from an earlier version, and you can turn it off in Settings > Privacy. While it is on, UltraTalk keeps a text-only memory of words and short phrases from your local transcripts: names and terms you repeat, corrections you speak, near misses of words already in your dictionary, and edits you make in UltraTalk’s Library. With each one it keeps counts and dates and, where available, how the word was misheard and a link to the transcript it came from. It contains no audio.

This memory stays in its own file in your user folder. UltraTalk never uploads this file and never adds it to a diagnostic report. Nothing is added to your dictionary until you accept a suggestion or add a word yourself. Separately, the application logs in a diagnostic report can contain short excerpts of your dictation and dictionary words that matched it, whether or not a word came from a suggestion.

In Settings > Privacy you can see the words UltraTalk has noticed and the ones you dismissed, delete a noticed word, copy the full memory as JSON, clear it, or turn learning off. Turning learning off keeps what is already there. Clearing it keeps dictionary entries you already accepted, which you can remove in Settings > Output. A word that was noticed but never suggested becomes eligible for removal 90 days after it was last seen. Deleting a transcript, or letting it expire under your Library retention setting, also removes what was learned from it, including suggestions that depended on it; dictionary entries you accepted remain. Other entries stay until you delete or clear them, or until the memory reaches its size limit. UltraTalk does not read your screen or watch your typing to learn vocabulary. If you use the “Add to UltraTalk Dictionary” service from another app’s Services menu, macOS gives UltraTalk only the text you selected.

When the app uses the network

UltraTalk connects to the internet for two routine purposes: downloading, repairing, or updating the speech models you choose, together with the runtime components some models need; and checking for and downloading signed app updates. The app checks for an update automatically at most once a day, and whenever you ask it to. Once a model and anything it needs are set up, transcription works entirely offline; airplane mode changes nothing about dictation. The only other time the app touches the network is when you explicitly send a diagnostic report, in a build where sending is enabled, as described below.

Diagnostics you control

If something goes wrong, you can export a diagnostic report from Settings > Privacy and inspect exactly what it contains. In builds where sending is enabled, you can also send it to us. This never happens automatically. Reports include application logs, your settings, and build metadata to help us reproduce the problem. The logs can contain short excerpts of your dictated text and dictionary words. Reports never include your audio or recordings. Sent reports are processed through Sentry, an error tracking service, solely to diagnose and fix issues.

macOS permissions

UltraTalk requires at most two permissions:

• Microphone, to capture your voice. It is used only while you are recording.
• Accessibility, to place text into the app you are using. No screen content is read.

Setup also offers notifications, which are optional. UltraTalk uses them for confirmations such as a word added to your dictionary. No contacts, no location, no files, no browsing history. If the app ever needs a new permission, macOS will ask you first, and this page will explain why.

This website

ultratalk.app does not use analytics, advertising trackers, or cookies. If you join the Windows waitlist, we store the email address you provide with our database provider and use it for exactly one thing: telling you when UltraTalk for Windows is available. You can unsubscribe at any time, and you can ask us to delete your address entirely by emailing us. Our hosting provider may keep standard, short-lived server logs to operate the site.

Changes to this policy

If this policy changes, we will update the date at the top of this page and note anything material in the changelog. We will never quietly weaken the commitments above.

Contact

Questions, deletion requests, or anything else: hello@ultratalk.app.