Reversible privacy for AI

Your AI never sees
real data.

Format-preserving encryption that lets AI models reason about your data without ever seeing it. The model works. Your data stays home.

query pipeline
You send a query with real data
Show me Elvis Magagula's accounts (ID: 9201015800087)
Veil encrypts before it reaches the model
What the AI model sees
Show me Kgosi Molefe's accounts (ID: 8805120300045)
AI responds with fake data
Kgosi Molefe has 3 accounts. Cheque: R24,500...
Veil decrypts before it reaches you
What you see — real data restored
Elvis Magagula has 3 accounts. Cheque: R24,500...

How it works

Three steps. Zero real data exposed.

01

Encrypt

Real data encrypted to format-preserving fakes. Names become names. Numbers stay numbers. Structure preserved.

02

Process

The AI model works with structurally valid data it thinks is real. It reasons, decides, responds. All with fake data.

03

Decrypt

Response decrypted back to real values before reaching the user. The model never knew. The user sees real data.

Features

Privacy that preserves utility

Format-preserving

Account numbers become valid-looking account numbers. Names become real names. IDs stay the right length. The AI cannot tell the difference.

Per-company salt

Each company sets a secret salt. Same data, same salt, same encrypted output every time. Consistent avatars across every conversation.

Fully reversible

Unlike masking, encryption is two-way. The response comes back with real values. Nothing is lost. Nothing is destroyed.

Stateless

No data stored. No mapping tables. The salt does all the work. Veil is a proxy, not a database.

Get early access

Be among the first to use Veil. Free during early access.