PII Redactor
Strip personal and sensitive data — emails, phone numbers, credit cards, SSNs, IP addresses, API keys, JWTs and your own custom terms — out of text before you paste it into an AI assistant, then swap the real values back into the reply. Reversible placeholders, and everything runs locally in your browser.
New to this? Read the PII Redactor guide →
Paste text above to detect and mask personal data.
Re-identify a response
Paste the assistant’s reply (which will contain the placeholders) and the map to swap the real values back in.
Paste the model’s response to restore the original values.
Detection, masking and re-identification all run locally in your browser — the text you paste is never uploaded.
How to use the PII redactor
Paste the log, ticket, stack trace or document you want an assistant to help with, and the redactor replaces the personal and sensitive data it finds with numbered placeholders — [EMAIL_1], [PHONE_1], [API_KEY_1] — while keeping the surrounding text intact so the model still has the context it needs. Copy the redacted version into the assistant. When the reply comes back still referring to [EMAIL_1], paste it (with the saved map) into the re-identify box to swap the real values back in.
What it detects
Out of the box: email addresses, phone numbers, credit-card numbers (validated with the Luhn checksum to cut false positives), US Social Security numbers, IPv4 and IPv6 addresses, AWS access-key IDs, common API-key and token shapes (OpenAI, GitHub, Slack, GitLab, Google), and JWTs. Add your own custom terms — customer names, project codenames, internal hostnames — and those are masked too, which is usually the most important part, because names are the one thing no pattern can catch reliably.
The reversible map
Redaction here is reversible by design. Each unique value gets one stable placeholder (the same email always becomes [EMAIL_1]), and the mapping is written to the placeholder-map box. Keep that map for the length of your conversation: it’s what lets you paste the model’s answer back and restore real values, so a reply like "email [EMAIL_1] about order [CUSTOM_2]" becomes actionable again. The map contains the sensitive values, so treat it like the data it protects — it never leaves your browser, and you can discard it when you’re done.
Why local matters here
A redactor that uploads your text to scrub it has defeated its own purpose. Everything on this page — detection, masking and re-identification — runs in your browser with no network calls, which you can confirm in the network tab. That’s what makes it safe to run on exactly the material you’re worried about: production logs, support tickets, HR documents. Pair it with a quick read-through, since no automated detector is perfect (see the guide on the residual risks).
Frequently asked questions
What does the PII redactor detect?
Emails, phone numbers, credit-card numbers (validated with the Luhn checksum), US Social Security numbers, IPv4 and IPv6 addresses, AWS access keys, common API-key and token shapes, JWTs, and any custom terms you add such as names or project codenames. Detection runs entirely in your browser.
Is the redaction reversible?
Yes. Each unique value is replaced with a stable numbered placeholder like [EMAIL_1], and a map of placeholder to original value is produced. Keep the map to swap the real values back into the assistant's reply using the re-identify step.
Will it catch every piece of personal data?
No automated redactor is complete. Pattern detectors miss unusual formats and can over-match, and names in free text need a custom-terms list. Use it as a strong first pass and still read through the result before sending.
Is my text uploaded?
No. Detection, masking and re-identification all run locally in your browser with no network calls — which is the whole point, since the text is exactly the sensitive material you don't want to send anywhere.