Redact PII from text, free and without signup
Paste anything. It runs through the same detector the API uses and shows you every span it found, with the exact character offsets. Nothing is stored.
What you are seeing
The output is produced by the same detector that serves the API, running on the same build. The playground is deliberately narrower than the production endpoint: input is capped at 2,000 characters, the confidence threshold is fixed at 0.5, and there is a per-IP budget of 20 runs per 10 minutes so a public page cannot be used as free compute.
Nothing you paste is written to storage. The text is redacted in memory and returned in the response.
Styles
- replace swaps each span for its entity type in brackets, for example
[contact.email]. The text stays readable and the shape of the sentence survives. - mask replaces every character with a block character, which keeps line lengths stable in logs.
- remove deletes the span outright, for fields that must come back empty.
When you need more than the playground
- Inputs longer than 2,000 characters, or more than 20 runs.
- JSON and CSV files, redacted cell by cell and returned as a download.
- Entity filters and a confidence threshold you control per request.
- Webhook delivery and an API key you can revoke.
Questions
Is the text I paste stored or logged?
No. The demo endpoint redacts in memory and returns the result. Request bodies are not written to storage, and the original text is not returned to the page beyond what your browser already holds.
Why is the input capped at 2,000 characters?
Inference runs on a shared CPU container. The cap keeps a single anonymous request from monopolising it, and keeps the free page responsive for everyone.
How accurate is it?
Format-bound types such as emails, cards, IBANs, and SSNs are matched by pattern and, where applicable, a checksum, so they are deterministic. Names, addresses, and health conditions depend on the model and are probabilistic. Treat the output as a strong first pass, not a guarantee.
Can I get the original values back?
Not from the output. Redaction is irreversible by design. If you need reversibility, keep your own mapping between the placeholder and the original value.
Related
- Entity typesAll 40 types, with the rules behind each one.
- Use casesWhere to put the call in a real pipeline.
- API referenceThe full endpoint, schema, and error reference.
- IntegrationsPython, Node, curl, n8n, Postman, LangChain.
Ready for real volume?
Get an API key for file uploads, entity filters, webhooks, and no rate limit. Plans start at $24/month.
See pricing