🕶️
Anonymizer · reversible LLM round-trips
← lab

Detect sensitive entities locally, swap them for opaque placeholders, send the redacted text to the LLM, then restore the originals on the way back. The model never sees a real name, IBAN, or address.

1

Original input

your machine
2

Anonymized — sent to LLM

leaves your machine
Run a round-trip to see the redacted version the LLM actually receives.
3

LLM response — placeholders intact

returns from OpenRouter
The model's reply, still containing the placeholders it saw.
4

De-anonymized — final output

your machine
Originals restored locally from the mapping vault.

Mapping vault (stays on your machine)

Placeholder Original value Entity type
No mappings yet — run a round-trip.