Secure anonymization
for AI.
Your superpower.
You want to use the world's best AI models – GPT, Claude, Gemini, Mistral – with your real data. No detours. No compromises. anymize makes that possible: through intelligent bidirectional anonymization. What the AI sees cannot be traced back to you. What you get is the full answer to your real case.
Real anonymization.
Not just EU hosting.
EU servers are not a firewall. Contractual assurances don't protect data. The only solution that truly works: sensitive information is altered before it reaches the AI.
While others promise where your data is stored, anymize solves the actual problem: whether the data is ever transmitted at all.
Three approaches. One difference.
“??? lives in ???”
Confused AI, unusable answer
“[REDACTED] lives in [REDACTED]”
Context lost, generic answer
“[[LASTNAME-a7f39cd]] lives in [[ADDRESS-b67d240]]”
Full context – the answer with your original data
Smart placeholders preserve the semantic context. The AI recognizes that [[LASTNAME-a7f39cd]] is a person and [[ADDRESS-b67d240]] is an address. The answer quality stays at frontier level – just without data leaving.
Four steps.
Zero compromises.
From input to answer: a single pass, fully reversible. You get exactly what you would have needed – with original data in your output, without it ever leaving your building.
Detection
A specialized 3-billion-parameter model scans your document and identifies personal and business-sensitive data. More than 40 categories. >95% detection rate (target on German-language documents). Context-sensitive and multilingual – German, English, French, Spanish, Italian.
Smart placeholders
Every detected element is replaced with a unique placeholder – semantically typed, so the LLM sees the type but not the content. The hash-to-original mappings are kept exclusively within the EU, isolated per workspace.
AI processing
Only the anonymized version leaves the anymize platform. To GPT, Claude, Gemini or Mistral – you pick the model. No personal data is ever transmitted unprotected to an LLM. No matter where the LLM runs.
Automatic de-anonymization
The AI's response contains the same placeholders. anymize replaces them automatically with your original data. The result reads as if you had spoken directly to the AI – except nothing sensitive left your building.
Bidirectional in action
One document.
One pass.
How it looks in practice – with the example of a legal document:
Dr. Maria Schmidt (Hamburg District Court, file no. 123 C 456/24) requests a review of contract no. DE2024-7789 with Müller GmbH, IBAN DE89 3704 0044 0532 0130 00.
Ms. [[LASTNAME-a7f39cd]] ([[ORGANIZATION-b67d240]], file no. [[CASE-c83e4f7]]) requests a review of contract no. [[ID-9e72b15]] with [[ORG-8a91c45]], IBAN [[IBAN-6f81d02]].
[[LASTNAME-a7f39cd]] should review clause § 4 para. 2. [[ORG-8a91c45]] granted a liability limitation in contract [[ID-9e72b15]] that is directly relevant to [[LASTNAME-a7f39cd]].
Dr. Maria Schmidt should review clause § 4 para. 2. Müller GmbH granted a liability limitation in contract no. DE2024-7789 that is directly relevant to Dr. Maria Schmidt.
The result: A fully contextualized answer to your real case – without a single name, address or IBAN ever leaving the anymize system.
Over 40 categories.
Automatically detected.
Every piece of sensitive information in your documents is identified and replaced by a unique, semantically typed placeholder. The following table shows an excerpt of the supported categories:
| Category | Examples | Placeholder (example) |
|---|---|---|
| Names | Max Müller, Dr. Schmidt | [[LASTNAME-a7f39cd]] |
| Addresses | Musterstraße 123, 80331 München | [[ADDRESS-b67d240]] |
| Locations | Berlin, Munich, Hamburg | [[LOC-c83e4f7]] |
| Organizations | Müller GmbH, Weber & Partner | [[ORG-8a91c45]] |
| Emails | max@example.com | [[EMAIL-9e72b15]] |
| Phone numbers | +49 176 12345678 | [[PHONE-6f81d02]] |
| IBANs | DE89 3704 0044 0532 0130 00 | [[IBAN-d4a7e1b]] |
| Social security numbers | 12 150385 M 123 | [[SSN-e5b8f2c]] |
| Tax IDs | 36 912 345 678 | [[TAX-f6c9a3d]] |
| ID and passport numbers | L01X00T47 | [[ID-a7b4e1f]] |
| License plates | B-AB 1234 | [[PLATE-b8c5f2a]] |
| Dates | 15.03.1985, 2024-01-15 | [[DATE-c9d6a3b]] |
| Case / contract IDs | 123 C 456/24, VP-2019-445566 | [[CASE-d0e7b4c]][[ID-e1f8c5d]] |
Names
Max Müller, Dr. Schmidt
[[LASTNAME-a7f39cd]]Addresses
Musterstraße 123, 80331 München
[[ADDRESS-b67d240]]Locations
Berlin, Munich, Hamburg
[[LOC-c83e4f7]]Organizations
Müller GmbH, Weber & Partner
[[ORG-8a91c45]]Emails
max@example.com
[[EMAIL-9e72b15]]Phone numbers
+49 176 12345678
[[PHONE-6f81d02]]IBANs
DE89 3704 0044 0532 0130 00
[[IBAN-d4a7e1b]]Social security numbers
12 150385 M 123
[[SSN-e5b8f2c]]Tax IDs
36 912 345 678
[[TAX-f6c9a3d]]ID and passport numbers
L01X00T47
[[ID-a7b4e1f]]License plates
B-AB 1234
[[PLATE-b8c5f2a]]Dates
15.03.1985, 2024-01-15
[[DATE-c9d6a3b]]Case / contract IDs
123 C 456/24, VP-2019-445566
[[CASE-d0e7b4c]][[ID-e1f8c5d]]More categories are continuously added – for example, industry-specific identifiers like insurance numbers or patent registrations.
Note on placeholders
The tokens shown are illustrative. Depending on context, document type and language, the specific placeholders vary – some are German (e.g. [[LASTNAME-…]]), others international (e.g. [[PER-…]]). The semantic typing is preserved in every case, so the AI always recognizes the type.
Note on transparency
The >95% detection rate is our target on German-language documents. For compliance-critical processes, anymize shows you what has been anonymized before sending – you always have four-eyes control.
File formats
What you upload.
What we detect.
Native PDFs and scanned documents (automatic OCR)
Word
.doc, .docx – including tables and formatting
Excel
.xls, .xlsx – cell-by-cell processing
Images
PNG, JPG, JPEG, GIF, BMP, TIFF (OCR)
Text
.txt – directly
Structured data
JSON, CSV
1 page of text in < 2 s · 10-page PDF in < 10 s · scan with OCR in < 15 s.
up to 50 MB per document, 500 pages, 100 requests/minute.
Where you use it
In every
anymize product.
Secure anonymization isn't just a feature – it's the foundation anymize is built on. You'll find it in all three products:
anymize Chat
The multi-model interface for everyday work. Every input, every upload is automatically anonymized – in real time. Admins can enforce anonymization for the entire team (Team Business / Enterprise).
More on anymize Chatanymize API
REST endpoints for anonymization, de-anonymization and chat – for integration into CRM, DMS and your own applications. Bidirectional in every automation workflow: Zapier, Make.com, n8n, Flowise, Langflow, MCP servers.
See integrationsanymize Enterprise
Tailored deployments – all the way to on-premise installation on your own servers or in your private cloud. Full data sovereignty, SSO/SAML, SLA, compliance reports for audits.
On-premise optionsSecurity architecture
More than
a promise.
No data transfer to third countries.
No US Cloud Act.
anymize is hosted entirely within the EU (at Hetzner in Germany) and is subject exclusively to EU data protection law. Our own 3B detection model also runs in the EU – not at a hyperscaler.
Who it's for
For professionals
who don't want to wait.
What you should know about secure anonymization.
Frequently asked questions
Document anonymization is the process of automatically detecting personal and sensitive information in a text, PDF, Word or Excel document and replacing it with placeholders – in a way that keeps the content machine-readable. At anymize it works bidirectionally: after AI processing, the placeholders are automatically replaced back with the original data. You work with your real cases, the AI never sees sensitive content.
We stand behind anymize. And we know – when an AI tool touches client, patient or employee data, a demo video isn't enough. That's why we give you 14 days of full access – all models, all features, no credit card. Enough time to be certain, before you trust us.
Your AI workplace awaits.