AI GuardienneShipped · V1Web Store: Coming Soon

Stop sensitive data before it reaches the prompt.

A Chrome extension that scans pasted, typed, and submitted text for secrets, PII, PCI, and PHI — entirely on-device, with zero network calls. Detection math ported line-for-line from the server engine, so browser and backend agree on what counts as a secret.

Try the demo Read the docs

Interactive demo — everything below runs in this page, exactly like the extension: on-device, no network.offline

Paste something you'd type into an AI chat tool

Insert sample:

1 · detection — waiting

2 · classification — waiting

3 · virtual redaction — waiting

Click Protect to see what would actually be sent.

In the real extension this happens in ~90 ms, before the request leaves the machine.

The boundary problem

AI tools sit entirely outside your security perimeter.

Every day, employees paste code, credentials, and customer data into ChatGPT, Claude, and similar tools. Your firewall doesn't see it. Your DLP suite wasn't built for it. The only place that data can still be caught is the browser itself — before the request leaves the machine.

0

Network calls

Fully offline by design — zero backend dependency at runtime. Nothing is logged out, nothing phones home. Verifiable under security review.

~62

Rules, four categories

Secrets, PII, PCI, and PHI rule packs with Luhn / Verhoeff / IBAN checksum validators, placeholder and human-sentence false-positive filters, and an entropy floor.

~90ms

Worst-case scan

Measured in the isolated background process — detection never blocks the page's UI thread, by construction.

Policy decides, per category and domain

allowwarnredactblock

Category toggles, domain allow/deny lists, severity thresholds, and redaction styles (replace, mask, partial-mask) — configured in the options page with import/export.

Same detection math as the server

The Shannon entropy formula, thresholds, context scoring, and confidence composition are line-for-line ports of the backend engine — so browser and server agree on what a secret is.

Strict core/extension split: the detection engine is pure TypeScript with zero Chrome or DOM dependency — which is what makes the VS Code port mechanical. 245 tests, 97%+ coverage.

Deploy today

Loadable as an unpacked extension.

AI Guardienne V1 builds with Node 20+ and Vite, and loads unpacked in Chrome today. Chrome Web Store publication is Coming Soon.

$ cd backend/radar_scanner/products/ai_shield
$ npm install && npm run build
› chrome://extensions → Load unpacked → dist/

Where it's going

One core, every surface.

Enterprise sync — org policies, masked-metadata telemetry Coming Soon
VS Code extension from the chrome-free core Coming Soon
Desktop agent — clipboard hooks, OCR Coming Soon
Slack / Teams / Outlook / Gmail scanning Coming Soon

See the full roadmap.

Close the AI gap.

Protection at the exact boundary where your data meets external AI tools.

Request a demo Read the docs

Part of the Perryscope platform — see also Secret Security and Privacy Security, or explore the architecture.