Five steps. Sub-millisecond pattern detection, fully on-device. Enterprise AI scanning uses Redactr's secure server — never stored, discarded immediately. Here's exactly what happens when an employee types a sensitive prompt.
The employee opens ChatGPT, Claude, or Gemini in their browser and begins typing a prompt — perhaps asking an AI to help write an email to a customer, debug code, or generate a report.
Redactr runs as a Chrome extension, silently watching the text input field. The employee doesn't notice anything — until a problem is detected.
The content/scanner.js file, injected into the chatbot page, passes the text to the detection engine lib/detector.js on every keystroke (debounced for performance).
The detector runs pattern checks simultaneously — all on-device using regular expressions and the Luhn checksum for card validation. On Enterprise plans, an additional AI pass is sent to Redactr's secure server for name and address detection; it is never stored.
When the risk score crosses the red threshold (61+), Redactr intercepts the send event and displays a modal overlay showing exactly what was found, the risk score, and the severity level.
The employee sees a clear warning with colour-coded severity. They cannot send until they take action.
The redaction engine rewrites the prompt, replacing each sensitive item with a numbered placeholder tag. The employee gets a usable, safe prompt — not just a blocked message.
This is the feature that makes Redactr helpful rather than annoying. Work continues; leaks do not.
[EMAIL_1] → sarah.jones@acme.com |
[PHONE_1] → +61 412 345 678 |
[SECRET_1] → sk-live-abc123
If the employee genuinely needs to send the original data (e.g. to a vetted AI system), they tap "Request Exception". An alert is pushed to the manager's phone via the Redactr mobile app.
The manager sees the employee name, the type of data, the risk score, and the full prompt context. They tap Approve or Deny and the extension acts immediately.
Plain HTML/CSS/JS. Markets and sells the subscription, then hands off to the backend to create the company and admin invite on purchase. PayPal Sandbox for payment.
Manifest V3. Content script injected into ChatGPT, Claude, Gemini. Pattern detection runs entirely in-browser. On Enterprise plans, text is also sent to Redactr's secure server for AI entity detection — never stored, discarded immediately.
Manager's control centre. Shows real leak alerts, risk scores, and employee names from Firestore. Approve or Deny with one tap, synced live.