Best Tools to Anonymize Documents Before ChatGPT or Claude (2026)

You want to ask an AI about a contract, a medical report, a client file or a 40-page insurance quote. You cannot paste it as it is: names, addresses, account numbers and amounts would leave your machine and land on someone else's server. Redacting them with black boxes does not work either, because the model then cannot tell who is who or which figure is which.

The answer is anonymization that keeps the meaning: every person becomes a consistent label such as [PERSON_1], every amount an [AMOUNT_n], and the document stays readable. Lawyers call it pseudonymisation; GDPR defines it in Article 4(5). A small crop of tools now does exactly this for documents. We compared nine of them.

We build one of them, BlurData, so read the ranking with that in mind. Prices and features are what each vendor published at the time of writing, September 2026.

What matters when the output goes to an AI

• Where it runs. If the tool itself is a cloud service, your document is uploaded before it is anonymized. That may be fine with a data processing agreement, and useless without one.
• Consistent labels. "John Smith" on page 1, "Smith" on page 12 and "J. Smith" in the footer must all become the same label, otherwise the AI sees three people.
• Human review. Automatic detection misses things and merges things. You need to see what was found and fix it before export, not after the AI has answered.
• The key. A mapping from labels back to the original values, kept separately, lets you read the AI's answer. Who holds it decides whether the process is reversible and by whom.
• Scans and images. Many real documents are scanned. Without OCR, nothing in them gets detected.

1. BlurData

Best for: reviewing a document on your Mac before it goes to an AI, fully offline.

BlurData is a native macOS app for redaction and pseudonymization. Drop a PDF, a scan or a folder; it detects names, emails, addresses, amounts, account numbers, IBANs, plates, IPs and URLs, plus country presets (SSN and EIN, Steuer-ID, Codice Fiscale, invoices, bank statements, court documents) and your own regex patterns. In Pseudonymize style every entity gets a typed label, and names are linked across the whole document: the surname alone, the OCR misread and the full name all resolve to one [PERSON_n].

The part that sets it apart is the Entities map that opens at the end of the scan: one card per entity with its spellings, count and pages. Tick two cards and merge them, split a wrong merge, drop what should stay visible. Export writes a new PDF or images, and "Export key" saves a CSV that maps each label to the original value. Nothing is uploaded at any point.

• Runs: on the Mac, offline. OCR on every page, so scans and screenshots work.
• Price: $79 per year, $149 lifetime, team licenses for 5 Macs. 14-day money-back guarantee.
• Limits: Mac only. Multi-line addresses are still detected line by line, so check them in the map. No API or Shortcuts action yet.

2. anonym.plus Desktop

Best for: offline desktop anonymization with an API and an MCP server for automation.

anonym.plus is a desktop application that processes everything locally and advertises more than 340 types of personal data across many languages. What makes it interesting for AI workflows is the built-in local API and MCP server, so an AI agent or a script can call the anonymizer directly instead of going through the interface. There is a free tier; paid plans are one-time lifetime licenses from €149 plus VAT, up to €599 for the top tier, with updates within the major version.

• Runs: on the desktop, offline.
• Price: free tier; lifetime licenses €149 to €599 plus VAT.
• Limits: broad entity coverage is a strength and a source of false positives; expect to review the output. The interface is built for throughput rather than for visual page review.

3. PII Anomalyzer

Best for: a flat-price desktop tool, popular with US legal teams.

PII Anomalyzer is a desktop application that detects, anonymizes and redacts personally identifiable information without sending anything to the cloud, using what the vendor describes as a dual-model AI. It is listed in legal software directories such as LawNext and Capterra, which tells you who it is for. One plan, $249 per year, no per-document metering, 7-day free trial.

• Runs: on the desktop, offline.
• Price: $249 per year.
• Limits: the highest yearly price among the desktop tools here; check the platform support before buying.

4. Stript

Best for: on-device anonymization with a local language model, if your machine can carry it.

Stript runs on macOS, Windows and Linux and uses a local language model to find personal data in documents before they go to ChatGPT or another AI. The model runs on your machine, which is the point, and also the cost: the vendor asks for 8 GB of RAM (16 GB recommended) and around 33 GB of free disk space. The app is free to use; Pro is €470 per year or €49 per month.

• Runs: on the desktop, offline, with a local LLM.
• Price: free; Pro €470 per year or €49 per month.
• Limits: hardware requirements, and a local model is slower than pattern-based detection on long documents.

5. anymize.ai

Best for: teams that want anonymization and the AI chat in one GDPR-oriented workspace.

anymize is a German cloud workspace: you upload the document, it is anonymized, and the anonymized text is sent to GPT, Claude, Gemini or Mistral from inside the same interface, with the answer de-anonymized on the way back. It targets law firms and companies that want GDPR-compliant AI usage without setting anything up. Plans start at €20 per month (450,000 credits, roughly 600 pages), Professional at €49 per month, with a free trial.

• Runs: in the cloud, EU-hosted.
• Price: from €20 per user per month.
• Limits: the document is uploaded to anymize before anonymization, so you rely on their processing agreement. Per-credit pricing on heavy use.

6. Docuflair Redact & Mask

Best for: organizations that need a server with projects, users and pseudonymization that can be reversed after the AI step.

Docuflair is a German server solution with a web interface. Mask, its pseudonymization product, replaces around 259 categories of personal data with consistent pseudonyms and can automatically restore the original values in the AI's output, the key idea of GDPR Article 4(5) turned into a product. Pricing starts at €299 per month for 5 users, billed annually, with larger tiers above.

• Runs: on a server (self-hosted or hosted), web interface.
• Price: from €299 per month.
• Limits: sized for organizations; overkill for one professional with a Mac.

7. Cleardox

Best for: legal and public-sector teams that need redaction and pseudonymisation at volume, with approval workflows.

Cleardox, from Denmark, detects personal data in PDFs and lets you choose between black redaction and pseudonymisation, with team review and approval and a fully automated module for thousands of documents. It is enterprise software: pricing depends on the modules and the volume, on request.

• Runs: cloud or on-premise deployment.
• Price: on request.
• Limits: procurement-grade pricing and onboarding.

8. Microsoft Presidio

Best for: developers building their own anonymization pipeline.

Presidio is an open-source library from Microsoft for detecting and anonymizing personal data in text and images. It is free, well documented and used inside many of the products above. It is not an app: you need Python, a place to run it and someone to tune the recognizers for your documents and languages.

• Runs: wherever you deploy it.
• Price: free, open source.
• Limits: engineering time; no review interface, no PDF handling out of the box.

9. Caviard.ai

Best for: a prompt, not a document.

Caviard is a free browser extension that redacts names, emails, phone numbers and IDs from a ChatGPT prompt before it is sent. Handy for a paragraph you are about to paste; not built for PDFs, scans or long files.

• Runs: in the browser.
• Price: free.
• Limits: prompts only.

Comparison table

How to choose

• You are one person or a small office on Macs, and you read every document anyway. BlurData: the review step is built in, the key stays with you, and nothing leaves the machine.
• You want to script it or plug an AI agent into it. anonym.plus, for the local API and MCP server.
• You want one flat yearly price on a PC. PII Anomalyzer.
• Your team lives in a chat interface and is fine with an EU cloud processor. anymize.ai.
• You are buying for an organization. Docuflair or Cleardox, and get the processing agreement in writing.
• You have developers. Presidio, and budget the time.

Whichever you pick, keep two habits: look at what was detected before exporting, and keep the key apart from the document you send. The second is what makes the difference between pseudonymised data and personal data with extra steps.