Interactive Playground

AI Prompt Sanitization Playground

Paste any prompt containing sensitive data and watch how Shield's redaction engine catches it. Toggle filter categories to see exactly what gets detected — and why.

Try a sample:

Your Prompt

0 chars

How Shield's Detection Works

Regex Pattern Matching

Fast, precise detection of known formats — emails, SSNs, credit cards, API key prefixes. Catches structured data instantly with near-zero false positives.

Entropy Analysis

Measures the randomness of text segments using Shannon entropy. High-entropy strings (tokens, keys, encoded data) are flagged even when they don't match any known pattern — catching custom API keys and novel credential formats that regex would miss.

Layered Defense

Shield runs both regex AND entropy analysis simultaneously, then cross-validates. A string that matches a regex pattern but has low entropy (like a placeholder) is deprioritized. A string that fails regex but has very high entropy gets escalated.

Local-Only Processing

All detection happens on your machine — before the prompt ever leaves your device. Shield redacts sensitive content, sends the sanitized prompt to the LLM provider, then rehydrates the response on the way back. No data leaves unredacted.

This is what Shield does — automatically, on every prompt

Purfect Shield is a local desktop app that runs this exact detection engine on every LLM API call — redacting PII, secrets, PHI, and payment data before it reaches any model provider. One binary, one config file, zero telemetry.

See Purfect Shield