Typly Anonymizer · B2B

Safe AI for Polish documents

ChatGPT, Claude, Gemini, Grok — without leaking personal data of your customers, residents or employees. We anonymize before the AI call, and restore the data in the response.

  • Validated on 1,000,000+ documents in INFOSTRATEG IV (NCBR)
  • 14 categories of Polish identifiers — from PESEL to court case numbers
  • On-premise inside your infrastructure or hosted in the EEA

15 minutes. Demo on your documents. No sales slides.

1,000,000+ documents anonymized in INFOSTRATEG IV (NCBR) · 14 Polish PII categories · GDPR + AI Act compliant
The problem

Your organization uses ChatGPT or Claude. Do you know what's happening with personal data?

In most Polish companies — nobody knows.

Employees use AI off the books

A lawyer pastes a contract into ChatGPT to speed up drafting. HR asks Claude to evaluate a CV. Accounting summarizes an invoice via Gemini. Polish PESELs, NIPs and customer names end up in US clouds. The compliance officer is the last to find out.

Manual redaction costs hours

A clerk spends 15+ minutes anonymizing a single letter before publishing it in BIP (the Polish public information bulletin). Hundreds of letters per month equal a full-time job. A law firm pays lawyer rates for work that can be automated.

AI Act 2026 demands accountability

AI adoption inside an organization without compliance controls is a regulatory risk. GDPR, the AI Act and sector-specific rules each require an audit trail. Existing processes don't cover AI use cases.

The solution

Safe AI in three steps

One product. Three use cases. Full compliance control.

Safe LLM

An employee pastes a document. The Anonymizer removes personal data. The document goes to ChatGPT, Claude, Gemini or Grok. After the response — reversible pseudonymization restores the data in the final version.

The employee gets the value of AI. The compliance officer sleeps well. Customer data does not leak.

BIP publication

Public information requests require redaction before publication. Our product detects 14 categories of Polish data — PESEL, NIP, IBAN, names, addresses, case references — and removes them destructively while preserving formatting.

PDF in → PDF out. DOCX → DOCX. Layout, tables, fonts — all preserved.

Compliance and archival

Pseudonymization with mapping for logs, archives and external analytics. Audit trail with detected entity positions, aligned with GDPR Art. 5(2). Deterministic hashing for consistency across documents.

Capabilities

What sets Typly Anonymizer apart

Polish PII — comprehensive

14 categories of Polish identifiers: PESEL, NIP, REGON, national ID, IBAN, KRS, administrative case number, land register number, court case reference, postal code, phone, date, e-mail, company names with legal form.

Each identifier is verified by a dedicated check — not just "some 11-digit number", but a real PESEL after validation. Plus names, addresses and companies detected in Polish linguistic context.

PESELNIPREGONNational IDIBANKRSCase numberLand registerCourt case refPostal codePhoneDateE-mailCompany name
1 000 000+
INFOSTRATEG IV · NCBR

Validated on a real corpus

Over 1 million documents from Polish public administration anonymized as part of INFOSTRATEG IV (NCBR — the Polish National Centre for Research and Development) — a strategic technology programme.

Document mix: resident inquiries, invoices, administrative decisions, official correspondence, summons, certificates. A realistic mix — not a synthetic test set.

Format-preserving redaction

A government-office user does not want raw text — they want a PDF returned as a PDF, a DOCX as a DOCX. Same layout, tables, fonts, headers — but without personal data.

We support: PDF (text and scanned), DOCX, PPTX, ODT, JPG, PNG, TIFF, plus plain text and e-mails. Document metadata sanitization (author, title). File names containing PII are automatically neutralized.

InputOutput
PDF (text and scan)PDF with preserved layout
DOCXDOCX with preserved styles
PPTXPPTX
ODTODT
JPG / PNG / TIFFImage with OCR redaction
TXT, EMLPlain text
redact

Redact

Placeholder with no reverse path. For BIP publication.

index

Index

Stable [PERSON_1]. For LLM round-trip.

hash

Hash

Deterministic pseudonym from a salt. For logs.

keep_format

Keep format

Structure preserved, value gone. For case numbers.

Reversible pseudonymization compliant with GDPR

Four anonymization strategies, decided per document:

  • Redact — placeholder with no reverse path. For publication.
  • Index — stable placeholder with a number ([PERSON_1]). For LLMs.
  • Hash — deterministic pseudonym from a salt. For logs.
  • Keep_format — structure preserved without leaking the value. For case numbers.

The same API serves a BIP request (redact) and a GPT prompt (index). Policy per request, not global.

Demo

Try it on your own document

No registration. 10 anonymizations per day (text or file, shared counter). Your content never lands on our disks.

Remaining: 5/5 anonymizations in the next 24 hours
0 / 5000
We do not log content. Only: IP, timestamp, entity count, processing time.

Or upload a document

You get back a clean PDF or DOCX containing only the anonymized text. No original images, signatures or page headers. 25 MB max.

Click or drop a file here
PDF · DOCX · PPTX · ODT · JPG · PNG · TIF · TXT

Want more?

Self-service API

1000 anonymizations/day. Free tier with API key.

Get an API key →
Live demo with our team

On-premise or hosted in the EEA. Custom policies. Format-preserving redaction. Full support.

Book a live demo →
Privacy

Privacy in one paragraph

Most privacy policies are legal jargon. Ours in three sentences: we do not store documents, we process in the EEA, we track only usage counters.

What we do NOT keep

The content of documents you paste or send via the API. Text enters server memory only for the duration of anonymization (milliseconds), gets processed and is immediately discarded.

We do NOT write it to disk. We do NOT log content. We do NOT use it to train models. We do NOT pass it to anyone.

What we keep (and why)

To enforce usage limits and to invoice, we keep minimal technical metadata:

  • Your IP address (hashed) — to enforce limits
  • Number of documents anonymized per month — for billing
  • Timestamp and processing time — for performance monitoring

We do NOT keep: document content, detected entities, mappings, file names, document metadata.

Where and how

Two deployment options:

On-premise — full product inside your infrastructure. Your documents never leave your data centre. Ideal for the public sector and regulated industries.

Hosted in the EEA — TYPLY SP Z O O servers inside the European Economic Area. Zero customer data transfer outside the EEA. Zero US sub-processors for content you anonymize.

Your rights (GDPR): access to metadata, account deletion, data portability (JSON export), withdrawal of consent, complaint to the Polish DPA (UODO). Contact: [email protected] · Data Controller: TYPLY SP Z O O

Deployment

Pick a deployment that fits your organization

On-premise

Inside your network. Full control.

  • Full product inside your infrastructure
  • Your documents never leave your network
  • Custom entity policies tailored to your processes
  • Configuration aligned with your security policy
  • Hands-on support for your IT team during deployment

Recommended for: public sector, banks, insurance, healthcare, regulated industries.

Hosted in the EEA

No setup. Ready to use.

  • REST API and Web UI from your account
  • TYPLY SP Z O O servers inside the European Economic Area
  • Zero customer-data transfer outside the EEA
  • Scaling managed by us
  • Updates and new features rolled out automatically

Recommended for: law firms, NGOs, mid-size compliance teams, organizations without a dedicated DevOps function.

Deployment details discussed in the live demo →
Industries

Safe AI for different industries

🏛️

Public administration

Redaction before BIP publication. Responses to public information requests. AI-assistant adoption for clerks without leaking resident data. GDPR + Polish freedom-of-information act compliance.

⚖️

Law firms

Masking client data in court filings before sending to ChatGPT/Claude. Detection of court case references, KRS, land register numbers. Format-preserving PDF→PDF redaction for procedural documents.

🏢

Corporate compliance and IT

Control over shadow AI tools. Employees use AI officially and safely. Polish PESELs do not leak to US clouds. Audit trail for the compliance officer. Pipeline-level masking automation.

🏥

Regulated sectors

Banks, insurance, healthcare. High compliance bar plus pressure to adopt AI. On-premise deployment + DPA + custom entity policies. Compliance with sector regulator recommendations.

Track record

1 million documents. INFOSTRATEG IV. NCBR.

Typly Anonymizer was validated as part of the INFOSTRATEG IV research programme run by NCBR (the Polish National Centre for Research and Development) — a strategic programme advancing modern technologies for the Polish public sector.

More than 1,000,000 documents from Polish municipalities were anonymized in the project: resident inquiries, invoices, administrative decisions, official correspondence with companies and contractors. A realistic mix — not a synthetic test set.

We do not disclose the names of partner offices in line with our partnership agreement. For your compliance officer, what matters more than a logo is a real-world benchmark on a corpus that resembles your documents.

1,000,000+
Polish public-administration documents
Compliance

Compliance as a foundation, not an afterthought

Pseudonymization compliant with GDPR Art. 4(5)

The mapping enables reversal. We meet the formal definition of pseudonymization.

Anonymization compliant with GDPR Recital 26

Redact strategy without a map → data outside GDPR scope.

Audit trail for accountability (Art. 5(2))

Every anonymization returns a list of entities with positions. Ready for decision_log.

AI Act compliance

PII anonymization is a foundation for high-risk AI system compliance. Audit trail with entity categories and strategies.

No leaks via metadata

PDF metadata, DOCX core props, file names containing PII — all neutralized automatically.

Configurable per-tenant salt

The hash strategy uses your salt. Pseudonyms do not collide between organizations.

See Typly Anonymizer on your documents

A 15-minute live demo with our technical team. We'll show the product on a document from your organization (anonymized in advance for the test). No sales slides. No follow-up spam.

Pick a slot

The demo is run by our technical team, not sales. We sign an NDA before any demo on your documents. Standard DPA and contract templates available on request.

FAQ

Frequently asked questions

Does it work offline?
Yes. In an on-premise deployment the full product runs inside your infrastructure with zero outbound connections. In the hosted deployment — operations run on our servers in the EEA.
Are my documents stored?
No. Document content enters server memory only for the duration of anonymization (milliseconds), gets processed and is immediately discarded. We don't write to disk, we don't log, we don't use it for training.
What about scanned documents or PDFs with handwriting?
Built-in OCR with a Polish language model. Scanned PDFs, JPG, PNG, TIFF — anonymized while preserving the original file format.
Can I add custom entity categories?
Yes, on the Enterprise tier. Custom policy plus optional custom training on your data. The standard 14 categories already cover most Polish public administration and law firm use cases.
What does AI Act 2026 compliance look like?
PII anonymization is a foundation for high-risk AI system compliance. We provide an audit trail with entity positions, category and strategy — ready for reporting.
Can I test without talking to sales?
Yes — the "Try it on your own document" section above gives you 10 free anonymizations per day (text or file, one shared counter), no registration required. Self-service signup at anon.typly.app for 1000 anonymizations per day.
What about pricing?
It depends on scale, deployment choice and compliance requirements. We discuss it in the demo based on your use case — we have models from self-service to enterprise on-premise.
What documents are needed to start?
For most customers: NDA + a standard licence agreement. For public sector and regulated industries: additionally a GDPR Art. 28-compliant DPA and a scope-of-work document. Templates available on request.
Contact

Let's talk

Live demo
Pick a slot
Business contact
[email protected]
Data Protection Officer
[email protected]