Veil

Mask before inference. Restore before delivery.

Private AI inference for MCP and OpenAI-compatible clients. Detected values are replaced with request-local tokens before inference, then restored exactly. Detection is probabilistic. This is not guaranteed anonymization and not HIPAA compliance.

Open chat

Endpoints

GET https://veil.nextautomatica.com/health
GET https://veil.nextautomatica.com/ui — website chat (paste a Veil API key; Claude/ChatGPT use the MCP App instead)
GET https://veil.nextautomatica.com/v1/models — authenticated catalog with live prices
POST https://veil.nextautomatica.com/v1/chat/completions — non-streaming text chat
https://veil.nextautomatica.com/mcp — Streamable HTTP MCP (private_chat, open_chat, list_models)

Limits

Text-only system / developer / user / assistant messages
No streaming, tools, attachments, or provider overrides
Clients never receive the server inference credential
Presidio, the worker, and the client sit inside the sensitive-data boundary