Security · no external AI

Your documents stay on your device.

The online account service is separated from the document engine. Privacy relies on observable technical controls, not a marketing promise.

Local architecture

A clear trust boundary.

Internet-facing features cannot access the document in plain text.

No external AI

Detection and replacement run without sending content to a remote AI model.

Offline-capable engine

The local engine and its libraries ship with the application, with no CDN dependency.

Accounts separated from files

The service knows your account and licence, not the contents being processed.

Encrypted identity vault

Mappings are encrypted in the browser. The password is never stored.

Verifiable network policy

The processing area applies a security policy that blocks outgoing connections.

Flattened PDF output

Final PDF pages contain no hidden text layer or recoverable annotation.

A precise promise

What the tool protects. What you still need to check.

Kept on your device

Document content is handled locally. The processing area blocks outgoing network connections.

Review before sharing

OCR can miss text in a damaged scan or an unusual layout. Inspect the protected copy, particularly photos and graphics.

No zero-risk claim

The final check targets known leaks; it cannot replace human review. We do not claim a certification we have not obtained.

After working with AI

What can be restored?

You need the same vault and unchanged codes.

Names and codes

Supported text, Office and OpenDocument content can recover the identities saved in your vault.

PDF

An unchanged COSMO copy uses saved positions; otherwise restoration relies on local OCR. The result remains flattened.

Not recoverable

Replaced photos, solid masks and deleted information cannot be recreated from the protected copy.

Practical answers

Questions about privacy

Check the result before you share — not after a leak.

Processing starts on your device. The original stays unchanged.

Try with a fictional file