Prompt guard for the AI era
VeilGuard catches API keys, passwords, personal data and seed phrases in flight — the moment you paste or press Enter in ChatGPT, Claude, Copilot, Gemini or a search bar — and lets you mask them before they ever leave your machine.
Desktop Claude desktop · VS Code · Microsoft 365 Copilot · Word · Outlook · Teams · Cursor · ChatGPT desktop
Product names are trademarks of their respective owners. VeilGuard is not affiliated with or endorsed by them.
How it works
On paste, on Enter, on attach — VeilGuard scans what you're about to send. PII, API keys, passwords, private keys, seed phrases, and the contents of attached documents are matched locally against its rules.
You get a clear, categorized warning — what was found, which category it belongs to, and how risky it is — before anything is submitted.
Choose Mask & send to swap secrets for safe placeholders, Edit to fix it yourself, or Send anyway. Admins can force redaction so the risky option disappears.
The halo
VeilGuard draws a thin border around the browser tab or desktop app it's protecting. Green means it's watching. Red means that surface isn't protected — so you never assume you're safe when you aren't.
Features
Stops sensitive data before it's sent to AI chats and search.
Blocks the tricks that steal clicks, clipboards and credentials.
Keeps keys, seeds and wallets out of the wrong hands.
Stops adversary-in-the-middle phishing that walks straight past MFA.
See what was caught and show your working. Generated on your device, in the extension and the Windows app.
Built for individuals, ready for teams.
Industry protection packs
Switch on the packs that match your work. Each adds detectors tuned to that field's sensitive data and scams, and still runs entirely on your device.
Keeps payment, banking and payroll data out of prompts, and flags business-email-compromise lures.
BEC Warns on "updated bank details" and urgent wire requests.
Helps protect patient data, and supports privacy programs for medical staff with local evidence logs.
Evidence A hash-chained record of what was caught, kept on the device.
Spots privileged and confidential material before it's pasted into an AI tool.
Blocks fake exchanges and wallet look-alikes impersonating Coinbase, Binance, Kraken, OKX, Uniswap, OpenSea, MetaMask, Phantom and others.
Before Risky sites are checked before a wallet is ever connected.
VeilGuard helps support data-protection and privacy programs. It does not by itself make an organisation compliant with any regulation or standard. Brand names identify impersonation targets only.
Privacy promise
VeilGuard is 100% offline by default. All detection runs on your device, and nothing leaves it without your approval or redaction. The four optional lookups are off by default, and turning one on shows you exactly what would be sent.
Read the privacy policy →Install
chrome://extensions (or edge://extensions).Protects Claude desktop, VS Code, Microsoft 365 Copilot, Word, Outlook, Teams, Cursor and ChatGPT desktop by intercepting paste and Enter.
Download for Windows (.exe)IT roll-out: VeilGuard-Setup.exe --install --silent · build from source
FAQ
Not by default. VeilGuard is 100% offline out of the box: all detection runs locally in the extension or desktop app, and your prompts, pastes and attachments are never uploaded, logged remotely or shared. Nothing leaves without your approval, or until you've masked it.
Only four optional lookups, all off by default: the password breach check (the first few characters of a password hash, using k-anonymity), RDAP domain age and DNS-over-HTTPS checks (a domain name), and the community blocklist download. Each needs your explicit approval, and the prompt shows exactly what would be sent.
Optional detector sets for Finance & Accounting, Healthcare & MedTech, Contracts & Legal, and Web3 & Crypto: checksum-validated card, bank and routing numbers, NPI and NHS numbers, privilege markings, wallet-drainer signatures and more. They run locally like everything else.
No tool can do that on its own. VeilGuard helps support your data-protection and privacy program by catching sensitive data before it leaves and keeping a local, tamper-evident record of what it caught. It isn't certified against any standard.
Adversary-in-the-middle kits proxy a real sign-in page to steal your password and one-time code. VeilGuard warns and blocks when a password you use on your real sign-in domains is typed on any other site (it keeps only a local salted hash), flags MFA codes entered on look-alike pages, catches device-code lures, and detects Microsoft, Google and Okta sign-in pages on the wrong domain.
A local record of detections in which each entry carries the SHA-256 hash of the one before, so any edit or deletion breaks the chain. One click re-verifies it. You can export an evidence pack, CSV, JSON or a printable HTML/PDF report, all generated on your device, in both the extension and the Windows app.
Sending is paused and you see a categorized warning with a risk level. You can Mask & send (secrets are replaced with placeholders), Edit the message yourself, or Send anyway. If an admin has enabled forced redaction, masking is required.
Out of the box: ChatGPT, Claude, Copilot, Gemini, Grok, ElevenLabs, Cursor, Perplexity, Amazon and Google, plus searches typed into the URL bar. You can add any other site yourself. The Windows app covers Claude desktop, VS Code, Microsoft 365 Copilot, Word, Outlook, Teams, Cursor and ChatGPT desktop.
Green: VeilGuard is actively protecting that tab or app. Red: it isn't — for example protection is paused or the surface isn't supported — so you know to be careful.
Yes. The Windows app reads admin policy from the registry and ships ADMX templates for Group Policy, including forced redaction.
For now it's installed as an unpacked extension from the release download, so you can read exactly what it does before you load it.
English, Spanish, French, German and Japanese.
Read the source, load it, and see the green border in minutes.