PII Redactor for GDPR & LLM Pipelines avatar

PII Redactor for GDPR & LLM Pipelines

Pricing

from $2.00 / 1,000 item redacteds

Go to Apify Store
PII Redactor for GDPR & LLM Pipelines

PII Redactor for GDPR & LLM Pipelines

Detect, mask or anonymize emails, phones, IBANs, cards, national ID numbers of 22 countries (PESEL, CNP, codice fiscale, DNI, EGN, Steuer-ID, SSN) and API secrets in text, PDF and DOCX. Checksum-validated, no external API.

Pricing

from $2.00 / 1,000 item redacteds

Rating

0.0

(0)

Developer

Maine Wayne

Maine Wayne

Maintained by Community

Actor stats

0

Bookmarked

2

Total users

1

Monthly active users

4 days ago

Last modified

Features

Limited permissions

Share

PII Redactor for GDPR & LLM Pipelines (EU national IDs, IBAN, secrets)

Find and mask personal data (PII) and secrets in text, PDF and DOCX before you share a document, store it, or send it to an LLM. Three modes cover the usual needs: numbered placeholders that can be mapped back (pseudonymisation), data masking with blocks, or removal. The Actor runs inside Apify: no external AI or redaction API, no API key, and nothing is sent to a third party.

Unlike simple regex masks, identifiers with a check digit are validated, so random numbers are mostly left alone: a 10-digit order number is not mistaken for a Bulgarian personal number (EGN), an 11-digit one must pass the Polish PESEL or another national rule, and a long number is not masked as a card unless it passes the Luhn check.

What it detects

TypeHow it is recognised
EMAILE-mail addresses
PHONEInternational numbers (+359 888 123 456, 00359...) and Bulgarian national formats (0888 123 456, 02 987 6543)
IBANAny IBAN of 80+ countries: length per country and the ISO 13616 mod-97 checksum must be valid
PAYMENT_CARD13 to 19 digit card numbers that pass the Luhn check
BG_EGNBulgarian personal number (EGN): valid birth date (all three century encodings) and valid check digit
NATIONAL_IDPersonal identification numbers of 22 countries, each checked against its own rule (check digit, check letter or embedded birth date): PL PESEL, RO CNP, IT codice fiscale, ES DNI/NIE, FR NIR, DE Steuer-ID, SE personnummer, FI henkilötunnus, BE national number, CZ/SK rodné číslo, GR AMKA, EE/LT personal code, LV personas kods, UK National Insurance number, US SSN. Numbers with a weak check are only taken next to their label: NL BSN, PT NIF, HR OIB, AT SVNR, IE PPS number, BG LNCh. With includeDetails each finding names its scheme (subtype, for example PL_PESEL)
EU_VATEU VAT numbers (format per country; Bulgarian numbers checksum-validated). Off by default
IP_ADDRESSIPv4 and IPv6 addresses
SECRETAWS access keys, GitHub and Slack tokens, Google API keys, sk- style keys, Stripe secret keys (sk_live_, rk_live_, whsec_), JWTs, private key blocks
custom termsAny words or phrases you list (names, project code names), case-insensitive

Names are not detected automatically. The Actor does not use a name-recognition model. Add the names you know to customTerms, or review the output before relying on it.

Input

FieldDescription
textsPlain texts (up to 100, each up to 1,000,000 characters).
urlsPublic links to PDF, DOCX, HTML or text files (up to 50). robots.txt is respected (also after redirects). PDFs need a text layer. Older encodings such as windows-1251 are read correctly.
inlineDocumentsSmall files as base64 ([{"filename": "a.pdf", "contentBase64": "..."}]).
entitiesWhich types to redact (see above).
customTermsExtra exact terms to redact.
modeplaceholder ([EMAIL_1], same value = same placeholder), mask, or remove.
includeDetailsAdd positions of findings (without original values).
includeOriginalValuesAdd a mapping placeholder to original, to restore text after an LLM step. This puts the personal data back into your dataset.

Example

Input:

{ "texts": ["Contact Ivan Petrov at ivan.petrov@example.com or +359 888 123 456. IBAN BG80 BNBG 9661 1020 3456 78."], "customTerms": ["Ivan Petrov"] }

Output item:

{
"source": "text #1",
"status": "ok",
"totalFindings": 4,
"findings": { "EMAIL": 1, "PHONE": 1, "IBAN": 1, "CUSTOM": 1 },
"redactedText": "Contact [CUSTOM_1] at [EMAIL_1] or [PHONE_1]. IBAN [IBAN_1]."
}

Ready-made examples

Saved examples you can open, run and copy:

Pricing

Pay per event: a small fee per item processed (item-redacted) plus a small fee per page unit (page-processed: one PDF page, or about 3,000 characters). Failed items are free. Set Max cost per run to cap spending; the Actor never processes or charges more than your limit allows. See the pricing panel for current prices.

Limits and honest notes

  • Rule-based detection: it finds well-defined identifiers, not everything that is personal. Names, addresses, free-text descriptions, dates of birth and other context are not detected unless you list them. Always review output that must be compliant.
  • National ID numbers are recognised by their format and check rule, not by context. A random number of the same length can pass a check by chance (measured: about 3% of random 11-digit numbers and 0.2% of random 13-digit numbers; none of 9 or 10 digits), so an order number is occasionally masked too. Switch NATIONAL_ID off in entities if that matters more to you than missing one.
  • Phone detection is pattern-based. A phone-like number can be missed or, rarely, a long number masked; check important documents.
  • PDFs must have a text layer; scanned PDFs need OCR first (run PDF & DOCX to Markdown, then pass the text here).
  • This is a tool, not legal advice, and no guarantee of GDPR compliance.
  • The Actor processes the personal data you give it only to redact it; results are stored in your own dataset (default retention applies).

FAQ

Is my data sent to another service? No. All detection runs inside the Actor.

Can I get the original values back? Yes, switch on includeOriginalValues with placeholder mode; you get a mapping to restore the text after an LLM step. Without it the originals are not returned.

Why is my 10-digit number not redacted as EGN? Because it is not a valid EGN (birth date and check digit must match). That is intentional.

Which countries' ID numbers are covered? Bulgaria (EGN, own type BG_EGN) plus Poland, Romania, Italy, Spain, France, Germany, Sweden, Finland, Belgium, Czechia, Slovakia, Greece, Estonia, Lithuania, Latvia, the Netherlands, Portugal, Croatia, Austria, Ireland, the UK and the US under NATIONAL_ID. Danish CPR numbers are not covered (they have no check digit and look like many other numbers).

Does it work with Bulgarian? Yes: Cyrillic text, EGN, Bulgarian phone formats and BG IBANs are supported.

Can an AI agent call it? Yes, via the Apify API or MCP server; send texts and use the redactedText field.