Data controls

Pressay uses your microphone, and when you confirm delivery it writes text into the app you are working in. This page states the boundary precisely: what Pressay can see, what it cannot, where your data is used, and where it goes afterwards. In one line: Pressay sees nothing until you start it, and does nothing until you confirm.

The current Pressay boundary: write, never read

By default Pressay writes text only when you confirm, and never reads the body text of the target app. Concretely, three negatives:

• It does not read field contents. Pressay neither reads nor alters text you have already written in other apps; the composer holds only what you handed to it.
• It does not read screen text. No screenshots, no reading of anything on screen, and no keystroke logging.
• It does not monitor edits after insertion. Once the text is in your app, Pressay stops reading it and has no idea what you changed afterwards.

There are only explicit paths on the write side: tapping the shortcut again confirms direct insert, and composer content is inserted when you click Insert. Both write text and neither sends it for you; Esc only hides the composer and never deletes the draft. The client collects no context from the target app, and the cloud that performs recognition and refinement sees only the audio and text you handed to the Pressay composer. When the desktop app uses the system accessibility APIs (AX/UIA) to insert, it reads only the metadata delivery requires — whether a focused target exists, whether it is editable, whether it is a secure field — never the text itself.

Context capabilities in the future

As an AI composer, Pressay may support questions and tasks in the future, but “understanding more context” will never be implemented as silent background reading. Any new context must become an input that is visible in the interface, reversible, and bound to a specific task; the current candidate has no such reading path.

When the microphone is on

By default the microphone runs only during voice input that the interface visibly shows. The sole exception is iOS Always Ready, which you enable explicitly: the main app keeps the microphone channel open for up to 12 hours, the system microphone indicator and the Live Activity stay visible, and standby audio is discarded on the device immediately — never saved, never transcribed, never sent over the network. Tapping enters visible continuous dictation until you tap again or press stop. Clicking the capsule only opens the composer and does not start the microphone. Any brief pre-roll audio used to cut first-word latency stays in private memory and enters real processing only after you confirm with a tap.

Ownership and retention

Your content never enters Pressay’s database

Raw audio travels from the client directly to the selected speech-recognition provider under a short-lived, session-scoped grant; it does not pass through DeepFlywheel infrastructure, and the client keeps no complete recording. Transcripts and refined results are never written to Pressay’s content database and exist only for as long as the current request requires. The cloud persists only your email address, account, login sessions, content-free usage and limited product events, any Pro launch-notice registration you submitted, and a request digest used to prevent duplicate processing — that digest holds only a request id, a request hash, and a processing state, and no audio or text. Production logs record no audio, no transcripts, no login codes, no sessions, and no model prompts.

Never used for training

Pressay does not train models on your voice or text. We select and configure only processing services that do not use request content for model training.

No selling, no tracking

We do not sell user data, run advertising, or track you across apps. This website loads no analytics, advertising, or personalization cookies. Pressay’s future revenue will come from software services, not from user information — that is not merely a promise; it follows from the business model itself.

Where processing happens

In the current version both speech recognition and text refinement run in the cloud, so a network connection is required. That is a property of the current implementation, not a comparative claim about on-device processing; we publish when data is sent, how it is processed temporarily, and where it is deleted, and both input and insertion remain triggered by you.

Data on your device

History and dictionary

History and dictionary stay on your own device by default. They are not uploaded, and they are not synced to your other devices. Each history entry stores only the final text you delivered — no intermediate versions, no raw transcript, no recording. You decide how long they are kept; setting retention to Never deletes existing entries permanently and immediately. Whether uninstalling also removes local data depends on the operating system, so uninstalling is not a reliable way to delete data.

How the dictionary learns

Beyond the entries you add explicitly, the dictionary also learns from the edits you make in the composer: when you deliver the text, Pressay sends that same delivered text so the service can judge which words you corrected. The request contains no target app, no surrounding context, and no audio; the service only judges — it creates no entry and keeps no copy, and the dictionary itself always stays on your device. Every automatic addition is announced and can be undone, and a deleted entry never appears in any later request. Pressay does not read content in other apps and does not infer your writing style from your history.

Login credentials

Login credentials are written to the private application-data directory of the current system user. Inside the app, only the main process can read them; the interface layer cannot.

Deleting your account

Deleting your account clears the server-side account, login sessions, verification codes, usage, product events, and any Pro-notice registration under the same email address; necessary Admin audit records are kept with the email address anonymized. Only after the server confirms success does the app clear local login state, history, and dictionary. If that cleanup fails, we keep your local data and tell you plainly rather than pretending it succeeded.

Third parties

Pressay uses third-party speech-recognition and large-language-model services to perform recognition and refinement. We send them only the audio and text you deliberately handed to Pressay, and we configure them not to use it for model training. Providers and configuration change as the service evolves; material changes are published on this page.

Contact us

If you have any question about how we handle data, write to support@deepflywheel.com.