Ruja Guard

Protect sensitive text and create reviewed plain-text copies locally before sharing with AI.

Ruja Guard is a private checkpoint for text moving into and out of AI tools. Before sharing, review a prompt, email, note, contract excerpt or code snippet for sensitive values. When text comes back, inspect unusual invisible characters and removable tracking parameters before creating a reviewed plain-text copy. Current capabilities: - local detection for email, validated IBAN and payment-card numbers; - bounded detection for bearer tokens, JWT-shaped values, assigned secrets and PEM private keys; - structural detection for Swedish personnummer and coordination numbers, Danish CPR candidates, and Lithuanian personal codes; - an encrypted private dictionary for client names, employees, codenames, cases, projects and product names; - stable, session-scoped placeholders and local restoration; - explicit Select all and Clear controls, plus a local exact-match action for values detection missed; - deterministic Clean Copy review for selected Unicode/control characters and an exact allowlist of URL tracking parameters; - an exact before/after preview before Clean Copy changes anything; - explicit clipboard write only after you choose Protect & copy, Copy sensitive result or Copy clean text. Ruja Guard does not read pages, inject controls into AI sites, request general website access, read the clipboard, create an account, or send reviewed text to a Ruja server. Clean Copy is not an AI-watermark detector or remover and does not rewrite text. Findings require your review; detection can miss sensitive values, and unusual characters can be legitimate, so check every result before sharing it.