See what Lucairn
would redact.
Paste a prompt — the way you'd send it to a model — and we run the L1 + L2 sanitizer pass against it. Presidio NER, heuristic rules, and known-entity matching. You get the redacted version, span locations, and per-recognizer confidence. Stateless: nothing is stored, nothing is sent to a model.
What runs.
The same L1 (Presidio NER + heuristics) and L2 (known-entity matching) layers Lucairn applies in production sanitization.
What does NOT run.
No bridge call, no LLM, no audit witness, no signed receipt. No L3 LLM-classifier pass — that's reserved for the full pipeline.
What you get back.
Redacted text with placeholders, a span-level findings list (placeholder, type, position, confidence), and which recognizers fired.
What we keep.
Nothing. Stateless: input and redactions are processed in-memory, not stored, not logged. Server logs only record status, latency, redaction count, and a hashed IP.
Paste a prompt,
see what L1+L2 catches.
Looking for batch scanning of files, S3, SQL, or full corpora? See the Scanner product page.
L1+L2 is one slice.
The full pipeline does more.
The full Lucairn pipeline adds the ID Bridge token mapping, L3 context-sensitive LLM scan, network-isolated inference, audit witness, and a signed certificate per response. Sign up free to get an API key and see all of it end-to-end.