PII Unmasking Proxy
Personal data is replaced with placeholders before it reaches an AI model, then put back afterwards. Try it with the sample text or your own.
New here? Read the short guide for this demo, with steps to follow and what to look for.
This is a demo, and it runs only in your browser. Nothing you type is sent to our servers or to any AI model, and nothing is stored when you leave the page. The "model reply" in step 3 is text you can edit yourself, not a real AI response.
1 Paste text with personal data
2 What the model would see
Every sensitive value becomes a placeholder. The same value always gets the same placeholder, so the model can still follow who is who.
The vault: what each placeholder stands for (stays in this page only)
| Placeholder | Type | Original value |
|---|
3 Restore the reply
A model replies using the placeholders. The proxy swaps the real values back in before the reply reaches the user. Edit the reply to see it work.
4 Audit log
Every mask and restore is logged with counts and types only, never the original values. You can download it as JSON.
- No events yet.
How it works
A proxy between your app and the model.
- Detect sensitive values with patterns, with checks like Luhn for card numbers.
- Replace each one with a stable placeholder and keep the originals in a vault that never goes to the model.
- Send only the masked text, then restore the placeholders in the reply.
- Log each step by type and count, so you can show what was protected.
Honest limits
What this demo does not do.
- It matches formats, not meaning. An Aadhaar-shaped number is masked even if it is not an Aadhaar number, and a card number is only caught when it passes the Luhn checksum.
- Names and addresses need a trained model or a list, so a production system would add one.
- The vault here lives in the page. A real system keeps it encrypted, with access controls and retention rules.
Have an idea? Let's build it properly.
Tell us what you want to build or automate. We'll reply with practical next steps.
