Checked on your device
The extension, the Windows app, the keyboard and the Outlook add-in all match text, files and pages locally. The optional AI model runs locally too.
A privacy tool that doesn't phone home.
VeilGuard is 100% offline by default. Detection runs on your device, and nothing leaves it without your approval or redaction.
Where your data goes
Never uploaded
Your prompts, pastes and attachments are never uploaded, logged remotely or shared. No personal accounts, no analytics, no telemetry.
You decide the rest
Four optional lookups are off by default. Turning one on shows you exactly what would be sent before anything is.
What leaves, and when.
The complete list. Anything not on it stays on your device.
- Your prompts, pastes & files
- Never sent
- Scan history & counts
- On-device only
- Evidence log, reports & exports
- Generated locally
- Password breach check
- Off by defaulthash prefix only
- RDAP domain age, DNS-over-HTTPS
- Off by defaultdomain name only
- Community blocklist
- Off by defaultdownload only
- Team plans: link to your company
- Off unless linkedcounts only, never text
How the private parts work.
Placeholders stay in the tab
Each sensitive value is swapped for a token such as [EMAIL_1] before the message is sent, so the AI only ever sees the token. The link between tokens and real values stays in that browser tab's memory, and VeilGuard draws your real value over the answer in its own protected layer that the website can't read. Closing the tab forgets the tokens.
Screenshots read on your PC
The Windows app reads the text in an image with the OCR engine built into Windows. The browser extension asks the Windows app on the same PC to do this, so images are never sent to a server. Without the Windows app, image checks are skipped and VeilGuard tells you so.
A log you can't quietly edit
Each entry in the local evidence log carries the SHA-256 hash of the one before, so any edit or deletion breaks the chain. One click re-verifies it, and exports are generated on your device.
Passwords never stored
The password-reuse alarm keeps only a local salted hash. The optional breach check sends the first few characters of a password hash, using k-anonymity, and only after you approve it.
Teams see counts, not content
A linked device sends counts of what was caught to the company console, never the text. A linked Outlook add-in sends daily counts only: date, detector, category, risk level and action.
Read it before you run it
Every extension build is reviewed by the browser stores and also published on GitHub, where its code can be read before you install it, and every detection pattern is published.
See for yourself.
Load it, open the test page, and watch what it catches without a single request leaving your machine.