VeriPrompt sits between your people and the AI providers they use. It routes each prompt to the right model, strips sensitive data before it leaves the building, and records what was asked, what it cost, and who approved it.
| Person | Area | Model | Cost | Shield |
|---|---|---|---|---|
| M. Falk | Guru | claude-opus-4 | €0.14 | 3 protected |
| S. Ortmann | Studio | gpt-4.1 | €0.02 | nothing found |
| A. Weber | Guru | mistral-large | €0.01 | 7 protected |
| T. Brandt | Ranger | claude-sonnet-4 | €0.31 | nothing found |
| J. Lindqvist | Architect | gpt-4.1-mini | €0.05 | 2 protected |
Before a prompt leaves your tenant, Shield detects personal and confidential data and replaces each value with a format-preserving token. The model sees the structure of your text, never the substance. Originals are restored on the way back.
Bitte prüfe den Vertragsentwurf für [PERSON_1] von [ORGANIZATION_1]. Rückfragen an [EMAIL_ADDRESS_1]. Abrechnung über [IBAN_CODE_1].
Presidio and spaCy scan the prompt and its attachments for names, contact details, bank details, credentials and your own redlisted terms.
Each hit becomes a readable surrogate token. The mapping is encrypted with AES-256-GCM and stays on your side.
The sanitized text goes to whichever model the routing policy selected.
Tokens in the answer are swapped back for the real values — visible to your user, never to the provider.