euRedact

European PII detection and redaction SDK -- available for Python and Node.js.

euRedact Core is the open-source SDK: it runs on your machine and detects and redacts personally identifiable information across 31 European countries using a two-pass rule engine: liberal pattern matching followed by checksum validation, context-aware suppression, and priority-based deduplication. Zero required dependencies in both languages.

Quick Start

Python

pip install euredact

import euredact

result = euredact.redact("Mijn BSN is 111222333 en IBAN NL91ABNA0417164300.")
print(result.redacted_text)
# "Mijn BSN is [NATIONAL_ID] en IBAN [BANK_ACCOUNT]."

Node.js / TypeScript

npm install euredact

import { redact } from "euredact";

const result = redact("Mijn BSN is 123456782 en email [email protected]", {
  countries: ["NL"],
});
console.log(result.redactedText);
// "Mijn BSN is [NATIONAL_ID] en email [EMAIL]"

Features

• 31 European countries with country-specific patterns
• 41 PII entity types: national IDs, IBANs, phone numbers, email, VAT numbers, license plates, credit cards, BIC/SWIFT, VIN, IMEI, GPS coordinates, UUIDs, social handles, IP/IPv6, MAC addresses, secrets/API keys, and more
• Optional euRedact Cloud (private alpha): mode="cloud" adds a fine-tuned model for the categories that have no shape to match on — person names, organisations, job titles, diagnoses
• Checksum validation: IBAN (mod-97), Luhn (credit cards/IMEI), and 30+ country-specific validators (Dutch BSN 11-proof, Belgian national number, French NIR, etc.)
• Secret/API key detection: known-prefix patterns for AWS, GitHub, Stripe, OpenAI, Slack, JWT, SendGrid + Shannon entropy-based fallback for generic secrets near context keywords
• Context-aware: keyword proximity checks and structural detection (JSON field names, CSV headers) for ambiguous patterns like dates of birth
• Custom patterns: register your own regex patterns for domain-specific PII
• Priority-aware deduplication: validated matches (checksum, corroborated by the document's country) > custom patterns > regex-only; a failed checksum demotes same-type matches rather than deleting them, so it can never silence a detection
• Country self-detection: infers a document's countries from the entities that carry one — IBAN prefixes, +CC dialling codes, VAT prefixes, BIC, email ccTLDs — so an ambiguous value resolves without the caller naming a country. countries never gates what is looked for; it scores and flags
• Referential integrity: consistent label mapping within a session (same PII value always gets the same label)
• Reversible tokenization: tokenize=True swaps values for EMAIL_K7Q2-style tokens and restore() puts them back — for prompts that go to an LLM and come back with the tokens in them
• Allowlist: values that are never redacted — your own organisation's name and addresses — per call or per instance
• Fast: see Performance
• Zero required dependencies in both Python and Node.js
• Thread-safe (Python), immutable detection objects

Supported Countries

Region	Countries
Western Europe	AT, BE, CH, DE, FR, LU, NL
Southern Europe	CY, EL, ES, IT, MT, PT
Northern Europe	DK, EE, FI, IS, LT, LV, NO, SE
Eastern Europe	BG, CZ, HR, HU, PL, RO, SI, SK
British Isles	IE, UK

Detected PII Types

EntityType has 41 distinct values. The rule engine emits the 29 that have a shape to match on:

NATIONAL_ID       BANK_ACCOUNT      PHONE             EMAIL
CREDIT_CARD       VAT               TAX_ID            SSN
PASSPORT          DRIVERS_LICENSE   LICENSE_PLATE     POSTAL_CODE
BIC               VIN               IMEI              SECRET
IP_ADDRESS        IPV6_ADDRESS      MAC_ADDRESS       UUID
GPS_COORDINATES   SOCIAL_HANDLE     DOB               DATE_OF_DEATH
HEALTH_INSURANCE  HEALTHCARE_PROVIDER  CHAMBER_OF_COMMERCE
RESIDENCE_PERMIT  INTERNAL_ID

11 are the cloud tier's. The rule engine never emits them — there is no shape to match on, which is precisely why the model exists:

PERSON_NAME       ADDRESS           ORGANISATION_NAME JOB_TITLE
MEDICAL_CONDITION SENSITIVE_ATTRIBUTE  BIOMETRIC_REF  FINANCIAL_AMOUNT
QUASI_IDENTIFIER  URL               LOCATION

The last is OTHER: the cloud client's fallback for a label the service returns and the SDK does not know. Neither layer detects anything as OTHER.

Three names are legacy aliases kept so existing code keeps working: IBAN → BANK_ACCOUNT, NAME → PERSON_NAME and CREDENTIAL → SECRET (the cloud model's label for prose passwords and connection-string secrets). The alias resolves to the canonical member, so the placeholder written into redacted text is [BANK_ACCOUNT], [PERSON_NAME] and [SECRET]. LEGACY_TYPE_ALIASES publishes the full mapping, which also recognises the string forms STREET_ADDRESS → ADDRESS and NATIONALITY_ETHNICITY → SENSITIVE_ATTRIBUTE.

Each detection carries a confidence of "high" (a pattern matched and its checksum passed), "medium" (the type comes from a label touching the span, because no pattern of that type claimed it) or "low" (a pattern matched, its checksum failed, and the document labels the span as that very type). Every detection is masked regardless; filter on confidence == "high" if you need only checksum-backed types.

Cloud tier

Status: private alpha. The cloud tier is in closed testing — it is not in public beta and is not generally available. Keys are issued to alpha participants only, and the request/response surface may still change between releases. The rules engine below is unaffected and is the supported path: mode="rules" is the default and carries no alpha caveat.

여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗