Our model for personal data. It finds names, tax numbers, accounts and addresses in a document, and covers the ones that belong to people before the text goes to a model.
Contrato de arrendamento
Entre Lusitânia Imobiliária, Lda., NIPC 512 345 678, com sede na Rua Augusta 14, Lisboa, e Maria Inês Carvalho, NIF 234 567 891, residente na Rua do Almada 210, Porto, nascida a 14/03/1988.
A renda de 850 € é paga da conta PT50 0035 0123 0000 4567 8901 2 para a conta do senhorio, PT50 0010 0000 1234 5678 9015 4.
Contacto da arrendatária: m.carvalho@exemplo.pt, +351 912 345 678.
The models.
Veil 1
The general model, for documents in European languages.
Veil 1 Mini
A smaller model that runs offline on a laptop's processor.
Veil 1 PT
Trained further on one country's identifiers, addresses and names.
Veil 1 PT Legal
A country model trained further on contracts and court filings.
Veil Judge
Decides whether each value belongs to a person or to a company.
AI assessment.
Before a team connects an assistant to its documents, Veil assesses every file: whether a model can read it, what it would expose and what it hides. You know which files need work before the rollout, not after.
Ready for AI
6 documents, 33 pages
82.1/100
- employee-handbook.pdf96.0
- master-services-agreement.pdf93.8
- annual-report-2025.pdf93.0
- rechnungen-2026-de.pdf85.6
- vendor-due-diligence.pdf84.5
- supplier-invoices-scanned.pdf40.0
Why 40
supplier-invoices-scanned.pdf
- No text layer on any page
- Scanned at low resolution
A model would get almost nothing from it. Route it through a vision model first.
- Can a model read it?
- A score out of 100 from more than twenty signals: text layer, scans, tables, columns, fonts.
- What would it expose?
- Which files hold names, accounts and tax numbers, and how many, before any of it is shared.
- What is hidden in it?
- Text a person cannot see and instructions written to an AI, file by file.
For the documents your industry runs on
Contracts, referral letters, statements, claims and CVs. Each page shows what Veil covers and what it keeps.
Try Veil on your own documents.
Veil is in training and runs on your own machines. We are looking for teams to test it on real documents.
Ask for early access



