Pattern detection
Names, SSNs, phone numbers, account numbers, and addresses are tokenized before a prompt can leave the machine.
Veilpath masks sensitive client information locally before every prompt reaches an AI provider, then restores real values only on your screen. Your machine, your API keys, no migration.
Windows pilot installed from a jumpdrive. Nothing to migrate or upload.
Protected before send
Local tokenization sits upstream of Anthropic, OpenAI, and Google.
What to expect from every pilot
The proof is architectural.
We show where masking occurs, what leaves the device, and how real values return locally.
Your keys, your machine
No new data custodian and no corpus to migrate.
A useful audit artifact
Monthly counts and categories, never the sensitive values themselves.
Three layers, one guarded send path
Veilpath combines broad pattern matching, entity detection, and a whitelist configured for your real client list. If the protection layer cannot complete, the prompt does not send.
Names, SSNs, phone numbers, account numbers, and addresses are tokenized before a prompt can leave the machine.
A second layer identifies names and sensitive entities that rigid patterns can miss.
Veilpath is configured around your actual client list, so protection reflects the work your firm performs.
After protection
Staff still get a model dropdown, file attachments, and the providers they already use. Real values are restored locally so the response remains useful on-screen.
Managed pilot
A focused engagement for small firms that need a low-consequence way to evaluate protected AI workflows. Final pricing is confirmed after a short scoping call.
Firm Pilot
Installation, configuration, and proof in your environment.
Custom quote
If the pilot fits, ongoing service includes managed configuration, human support, and a monthly compliance report with counts and categories only.
Scope a pilotSecurity review, answered plainly
A safer first step
See Veilpath on one Windows machine using your keys and a workflow your firm already understands.