Datablur AI — PII Redaction for AI Chats

Datablur AI catches sensitive data in your prompts before you hit send. It watches the prompt box on major AI chat sites — underlines anything sensitive as you type, and steps in before a flagged prompt goes out. WHAT IT CATCHES • Contacts — email addresses, phone numbers, postal addresses • Financial — credit card numbers (Luhn-checked), IBANs, US Social Security numbers • Secrets — API keys, JWTs, private keys, database connection strings, and high-entropy tokens that match no known format • Identity — personal names, dates of birth, ages • Crypto — Bitcoin, Ethereum and Solana wallet addresses • Money — currency amounts Each category can be switched off independently, and a strict / balanced / lenient dial controls how aggressive the detection is. THREE MODES • Warn before send (default) — the send pauses and you choose: edit, redact & send, or send anyway • Auto-redact on send — masks everything automatically, with one-click Undo • Highlight only — never blocks or alters anything, just flags HOW IT LOOKS Sensitive values are underlined right in the prompt box, colour-coded by severity, with a hover tooltip explaining what was found. A small indicator in the corner of the screen shows the severity mix at a glance and expands into a panel where you can redact or ignore each finding individually. PRIVACY The whole scan runs inside your browser. Your prompt text, the values it detects, and the sites you visit are never transmitted — not to us, not to anyone. This is enforced by the build itself: the release pipeline scans every shipped file for networking code and fails if it finds any outside a single, optional analytics call. That one call sends anonymous counters only — how many values were redacted, in which mode, which settings you changed. Never content, never hostnames. One toggle in the popup switches it off entirely. No account. No login. No sync. No ads. Works offline.