Preserve: PII detection demo

A live, private demo

A hands-on demo of what the Preserve library and API do: detect and scrub personally identifiable information into reversible placeholders like [EMAIL_1], using the same detectors. Type or paste on the right and watch it work.

Your text never leaves this browser. No network requests, no server, no logging; everything runs in client-side JavaScript on your device. (Open the network tab and watch: nothing is sent.)

How detection works

1 Normalcy scanner flags regions that look unusual, to focus the search library
2a 53 regex patterns across 15+ countries live
2c 9 checksum validators (Luhn, IBAN, CPF, DNI/NIE, BSN, NHS, HETU, RRN) live
2d context scoring weights each match's confidence live
2h name gazetteer, bare names across 55 countries live

3 Local LLM review reads leftover free text for PII the rules miss precomputed

Layers 2a to 2h run live in your browser. Layer 1 (normalcy) and Layer 3 (LLM) belong to the full Python library; the browser can't run a model, so the Layer 3 panel shows results precomputed by the real pipeline on the built-in examples.

regex / checksum name scorer local LLM

Coverage

53regex patterns

9checksums

55countries (names)

Sensitivity

Examples

Input

Detected (highlighted)

Scrubbed output

With a local LLM (Layer 3)

Reversible: each placeholder like [EMAIL_1] maps back to the original value locally. Part of Preserve, MIT License. Patterns are exported verbatim from the Python library, so this demo stays in sync with the real detector.