Document anonymization: detect, review, restore.
A desktop application that does one thing and does it rigorously: it removes personal data from documents before you use them with AI, and puts it back into the answer. No servers, no cloud accounts, no compromises.
Find personal data. Even what a tired eye would miss.
The detection engine combines patterns, name dictionaries in four languages, field labels and mathematical validators. Every finding gets a confidence score.
- Italian tax code (codice fiscale) with check-character verification, omocodia included
- VAT number with check-digit verification
- IBAN with mod-97 check and length for more than 60 countries
- Payment cards with the Luhn algorithm, network and length
17 types of data detected
RSSMRA85T10A562S passes the check and is anonymized. RSSMRA85T10A562X has the same shape but is not a valid tax code: it is almost always an order reference, and it stays as it is.
The last word always belongs to a person.
Certain findings, such as a validated IBAN, are already confirmed. Uncertain ones, such as a name appearing without a title, wait for your decision. Nothing is replaced without the reviewer knowing.
- Context and confidence score for every finding
- Confirm or reject one by one, by multiple selection or all at once, keyboard included
- Residual-risk indicator before export
- Custom lists of terms to always redact or never touch
- Six built-in profiles, adaptable to the type of document
Recall from heuristics, precision from human review: a surname the dictionary does not know still reaches the reviewer, without the document ever being altered silently.
Consistent placeholders that AI understands.
Every value becomes a stable placeholder. The same person stays [PERSONA_001] throughout the document, even when later referred to by surname only. The model reasons about relationships without knowing identities.
- Alias propagation: “Mario Rossi”, “Rossi” and “Mr Rossi” remain the same person
- Protection against ambiguity when a surname is shared by several people
- Metadata removal: author, company, comment authors
The safe document, in the format you need.
Save the anonymized version or copy the safe text straight to the clipboard, ready to paste into the AI tool you already use.
- Word (.docx): styles, tables and layout preserved
- PDF: detection on the text layer and a new anonymized version
- Text: any plain-text file
- “Copy safe version for AI”: one click, and the text is on the clipboard
Supported formats
| Format | Read | Export |
|---|---|---|
| Word .docx | Formatting preserved | |
| PDF with text | Regenerated document | |
| Text .txt | ||
| Scanned PDFs | Flagged as such OCR coming | |
| Excel and PowerPoint | Coming soon | |
The AI's answer, with the real data back in place.
Paste the model's answer: AnonyAI replaces each placeholder with the original value using the encrypted map. If the AI has invented a placeholder that does not exist, you are told rather than it being ignored.
.anonymapmap encrypted with ChaCha20-Poly1305 and scrypt- Unknown placeholders flagged
- Restore stays available even after the trial expires: your data is always yours
[PERSONA_001] may withdraw within 30 days…”Easy to deploy. Easy to verify.
No administrator rights
Extract a folder and run it. It even works from a USB stick. Windows 10 and 11, 64-bit.
Command line included
The same functions as the interface are available from the terminal, to build anonymization into existing procedures.
Four languages
Interface and detection in Italian, English, Spanish and German, for international documents and teams.
Audit log
A local log of operations supporting GDPR accountability, which never contains the data processed.
No external dependencies
No third-party libraries, no models to download, no cloud service for processing. A smaller attack surface.
Isolated workstations
Offline activation with two files carried by hand. The licence is verified locally, for ever.
What comes next.
Every new feature will respect the same constraint as version 1.0: no data leaves the workstation.
OCR
Scanned PDFs and images, with a bundled offline OCR engine. Never a cloud service.
Batch processing
A whole folder with a single map, so the same person has the same placeholder across the entire case file.
Excel and PowerPoint
Spreadsheets and presentations, without breaking formulas and references.
Central policies
With the Business edition the DPO defines profiles and lists once for every workstation.
See it at work on your own documents.
Seven-day full trial. No credit card, no document sent anywhere.