About

The documents that need AI most
are the ones you can’t paste.

AnoVeil exists because professionals were left with two bad options: paste confidential work into a tool they don’t control, or do without the tool entirely.

The problem

AI is most useful on the documents you can least afford to share.

Summarising a contract, restructuring a report, drafting from rough notes — these are the things language models are genuinely good at. They are also the things people do with material that carries client names, patient details, salaries and account numbers. So the work either goes into a tool nobody in the room controls, or it doesn’t get the help it needs. Most people quietly choose the first and hope.


Why the usual answers don’t fit

Enterprise agreements

They cover a company, after months of negotiation and legal review. They don’t cover the consultant working from a laptop, and they don’t arrive in time for the document that is due on Friday.

Manual redaction

Black boxes take the meaning out along with the name. A contract full of blanks is a contract no model can reason about — which was the entire reason for asking.

Simply not using AI

The safest option and the most expensive one. The work still has to be done, only slower, and the people who skip the rule keep their advantage.


What AnoVeil does instead

Keep the structure. Replace the identifiers.

On your own machine, AnoVeil swaps identifying values for consistent labels — the same person becomes the same token every time they appear. The document keeps its shape, its argument and its numbers, so a model can still work with it. You paste the safe copy wherever you like, and AnoVeil puts the real values back into the answer when it comes home.


What we don’t claim

Detection is not perfect, and pretending otherwise would be the most dangerous thing we could do.

Names are the hardest category. In our own testing AnoVeil finds roughly 76% of the person names in a document in its high accuracy mode — across runs that has ranged from 76% to 85%, and we quote the most recent. It is a property of the problem, not a gap we expect to close next month.
Structured identifiers behave far better. Email addresses, phone numbers, account and card numbers, and the postcodes inside street addresses have a shape to match, and are caught much more reliably than names.
So AnoVeil reduces exposure — it does not eliminate it. Read the anonymised copy before you send it anywhere. Treat it as a strong first pass made by a careful assistant, not as a guarantee.
Where to look next

Don’t take any of this on faith.

The security page sets out exactly what leaves your machine and what never does. The use cases page shows where this actually fits in a working day.