AgentGateUntrusted input guardrail for AI agents

Guardrail scanner

Guard untrusted content before it reaches your agent.

Paste a trusted task and untrusted content. AgentGate returns an allow, sanitize, or block decision with the exact output to use.

Run a checkSource tests
01Untrusted content

Ticket, page, file, webhook, or tool output.

02Guardrail decision

Classify injection risk and extract hostile instructions.

03Safe output

Allow, sanitize, or block before the agent sees it.

Input

Scan content

Do not submit secrets, credentials, or sensitive customer content.

Load demo

Start with a malicious or benign sample.

Task

What the agent is supposed to do.

53/2,000
Policy
Content

Ticket, email, webpage, document, or tool output.

164 / 5,000 characters

Guardrail checks run server-side. API keys stay in `.env.local`.