Docs / Scanning

AI Guardienne

A Chrome Manifest V3 extension that scans pasted, typed, and submitted text for secrets, PII, PCI, and PHI — entirely on-device, zero network calls. V1 loads unpacked; Web Store publication is Coming Soon.

Build & install

# Node 20+ required
cd backend/radar_scanner/products/ai_shield
npm install
npm run build
# chrome://extensions → Developer mode → Load unpacked → dist/

Architecture

Strict split: core/ is a pure-TypeScript detection/policy/redaction engine with zero Chrome or DOM dependency; extension/ holds thin browser adapters. Detection math — Shannon entropy formula, thresholds, context scoring, confidence composition — is ported line-for-line from the backend. Worst-case scan ~90 ms in the isolated background process; the page's UI thread is never blocked. 245 tests, 97%+ coverage.

Rules

~62 rules across four JSON rule packs — secrets · PII · PCI · PHI — with Luhn / Verhoeff / IBAN checksum validators, placeholder and human-sentence false-positive filters, an entropy floor, and a high-entropy fallback (extractHighEntropyStrings). Known gap: NER-grade PERSON/ORG/LOCATION entities are server-side only in V1.

Policy

Actions per category and domain: allow · warn · redact · block
Category toggles, domain allow/deny lists, severity thresholds
Redaction styles: replace, mask, partial-mask
Prompt overlay UI · popup with stats/history · options page with import/export
Works on CSP-strict (Google-family) pages

Enterprise sync, org-managed policies, and the VS Code port are phased roadmap items Coming Soon — see the roadmap.

Source: products/ai_shield/ (core/, extension/, docs/ROADMAP.md)

← Privacy scanning CLI →