AgentGateUntrusted input guardrail for AI agents

Direct API

POST /api/sanitize

Send trusted task, source type, and untrusted content directly to the existing scanner endpoint.

JSON bodyNormalized decisionSaved run support
Open API docs
Safety notes

Private URLs are blocked, file uploads are capped at 1 MB, and provider keys stay server-side.

Webhook

External system POST

Load demo

Use these examples to test the guardrail without entering real data.

No result yet.

Run this ingestion test to see a verdict and saved-run link.