Product

Read the document.
Not the people in it.

AnoVeil takes a document you cannot share, and gives you one you can — without taking the sense out of it on the way.

What it does

One pass, four moves.

AnoVeil is a desktop application for Windows. You give it a document, it finds the values that identify the people and organisations inside, and it swaps each one for a consistent label. What comes out keeps the shape and the argument of the original and loses the identities. When you are finished with the safe copy — a summary, a redraft, an analysis — AnoVeil puts the real values back into the result.

All four moves happen on your machine. There is no upload step, no queue, and no copy of your document anywhere but your own disk. A fast mode trades some accuracy on names for roughly thirty times the speed on large documents.

It opens Word (.docx), Excel (.xlsx), PowerPoint (.pptx), Markdown (.md) and plain text (.txt), and it takes text pasted straight into the application. PDF is not supported. If your source is a PDF, convert it to Word first.

See exactly what leaves your device
Product screenshot Main window — document open, findings in the sidebar
How it works

Three things it has to get right.

01

Detect

Find what identifies someone, wherever it sits in the document.

AnoVeil reads the whole document and marks the values that point at a real person or organisation: names, organisations and places, email addresses, phone numbers, street addresses with their postcodes, account, card and national identifiers, IP and web addresses, project codenames and credentials. Dates, money amounts and internal system names are optional categories, off until you switch them on. Structured values have a shape to match and are caught reliably. Names are the hard case, and the review step exists because of it.

Runs against the document text on your own machine
Bundled detection models, no network call to classify anything
Your own vocabulary and whitelist steer what it looks for and what it leaves alone
02

Anonymize

Replace each value with a stable, readable token.

Every detected value becomes a labelled placeholder that says what it replaced without saying who it was. The same person is the same token everywhere they appear, across every document in the set, so relationships in the text survive the substitution. A contract still reads as a contract between two parties, not as a page of black boxes.

Consistent tokens across a whole document set
Labels carry the category, so a model still knows a date is a date
Or realistic fake names instead of labels, where placeholders would read badly
The mapping back to real values is written only to your local codebook
03

Preserve context

Keep the structure, the argument and the numbers.

Redaction removes meaning along with the name. AnoVeil does not. Headings, clause numbering, tables, emphasis and document order come through unchanged, and every non-identifying figure stays exactly as written. The anonymised copy is a document you can still reason about, which was the entire reason for producing it.

Formatting and document structure come through intact
Non-identifying values are left alone
The result reads as prose, not as a form with the fields cut out
Review and export

Detection is a safety net,
not a guarantee.

Which is why nothing is exported until you have looked at it. The review step is the part of AnoVeil that matters most, and the part we are least willing to automate away.

01
Scan
Open a document and AnoVeil marks everything it believes identifies someone, grouped by category with a count for each.
02
Review
Work through the findings. Accept what is right, reject what is not, and add anything the scan missed. A page-layout view shows each finding marked on the real laid-out document. Nothing is decided for you.
03
Export
Write out the anonymised copy in the original format. The codebook stays on your disk, never in the exported file, and a local audit log records every document processed — exportable as CSV.
04
Restore
Bring the answer back from wherever you used it and AnoVeil turns the tokens into the real values again, from the codebook only you hold.
Product screenshot Review screen — findings grouped by category
Product screenshot Restore — tokens resolved back to real values
What we don’t claim

Names are the hardest category, and we publish the per-category figure rather than a blended average.

Structured identifiers behave far better than names. Read the anonymised copy before you send it anywhere, and treat it as a strong first pass made by a careful assistant rather than a guarantee.

Parts of a file are not covered yet, and you should know which before you rely on it. In Word, hyperlink display text, text boxes, footnotes and tracked changes are left untouched. In Excel, charts, images and formulas are left as they are — and because Excel recalculates formula results and chart caches when it opens the copy, a cached value there could still show an original. A cell whose text was replaced loses mixed formatting inside that one cell.

The numbers, in full
Early access

Not public yet.

AnoVeil is in private testing on Windows. Leave your email and we’ll tell you when it’s ready to download.

Get early access See where it fits