Rampart

Rampart detects 17 entity types using a 35-label BIO schema. A deterministic recognizer layer handles structured identifiers (SSN, credit cards, emails, URLs, IPs); the neural model handles contextual PII (names, phones, addresses, document numbers). City, state, and ZIP are retained by default for context.

⚠️ Known limitations of these demo inputs

• Phone numbers must not start with 555. The 555-01XX exchange is reserved for fiction and is effectively absent from real-world training data, so the model does not recognize it as a phone number. Use a realistic exchange (e.g. 212-867-5309) to see PHONE detection fire.
• Credit cards that fail the Luhn checksum pass through unredacted. CREDIT_CARD is caught by the deterministic layer, which validates the Luhn check digit. A made-up number that does not satisfy Luhn is treated as ordinary digits and is not redacted. Use a Luhn-valid test number (e.g. 4532 1234 5678 9014) to see it caught.

See how Rampart protects you before sending text to an AI assistant. Your message is redacted client-side, the "assistant" sees only placeholders, and the response is rehydrated with your real values — all in your browser.

Rampart replaces PII with stable, typed placeholders like [GIVEN_NAME_1]. The same value always maps to the same placeholder, so multi-turn conversations stay coherent. The real values live only in the client-side entity table — they never leave your device. Type in the demo above, then check here to see the mapping.