Check every message before it reaches your agent, and every reply before it leaves.

ProofAgent Guardrail masks private data and blocks attacks in one call. Powered by ProofAgent Nano.

Four checkpoints on every turn

The user message, the tool call, the tool result and the reply are each checked before they move on.

Private data is masked

Twenty kinds of personal data, from names and cards to health data (PHI) and secrets. Your agent only sees placeholders.

Risky actions wait for a person

Send, delete, pay, publish or change access: each can be allowed, held for review or blocked.

Your rules, as code

Policies are versioned files, with starting points for healthcare, banking, insurance, airlines and legal.

One API call, or MCP

Call POST /v1/scan from your server, or add it as a tool in Claude Code, Cursor, n8n and any MCP client. 1,000 calls a month free. Text is checked in memory and never stored.