redacta 1.1.0

Pseudonymise patient identifiers and PII in text (and restore them) — a local, dependency-free pattern engine.

redacta

Pseudonymise patient identifiers and PII in text — and restore them. A local, dependency-free Python pattern engine.

pip install redacta

from redacta import redact, reinstate

redacted, report, token_map = redact(
    "Dear patient, NHS Number: 943 476 5919, tel [phone]."
)
# redacted -> "Dear patient, NHS Number: [NHS_NUMBER_1], tel [PHONE_1]."

original = reinstate(redacted, token_map)
# original -> "Dear patient, NHS Number: 943 476 5919, tel [phone]."

What it detects

Deterministic, checksum-validated patterns: NHS numbers (Modulus-11), UK National Insurance numbers, dates of birth (keyword-anchored; appointment dates left intact), UK postcodes, US SSNs and ZIP codes, hospital/MRN numbers, emails, and phone numbers. Same value → same token; a token_map lets you reverse it.

Scope: this library is the deterministic layer only. Names, postal addresses and identifying ages need contextual judgement and are not covered here — the Redacta agent skill and the MCP server add those via reasoning. Stdlib only, no network calls; review output before sharing.

CLI

redacta letter.txt                 # prints JSON: redacted_text, report, token_map
redacta letter.txt --text-only     # just the redacted text
redacta-reinstate redacted.txt --map token_map.json

License

MIT-0 (MIT No Attribution). Built by PharmaTools.AI.