Protect Overview

Where Protect's dashboard guardrails and SDK checks live

What is Protect?

Protect is the safety layer over your AI traffic. It runs named guardrail checks, like PII detection or prompt injection, on each request and returns one of four actions: block, warn, mask, or log.

You turn guardrails on in the dashboard under Gateway → Guardrails, where they apply to traffic passing through a gateway. If your app isn’t pointed at Agent Command Center yet, see the quickstart for swapping in the base URL and API key. You can also call protect() directly from your code to run checks inline on text, image, and audio inputs. Use the dashboard if traffic goes through a gateway; call protect() directly if it doesn’t.

Protect is available on Future AGI Cloud and Enterprise plans.

Start here

Understanding Protect

How guardrail checks, actions, and the two surfaces fit together

Turn on a guardrail

Enable a check on your gateway traffic from the dashboard

Run Protect from the SDK

Call protect() directly from your code instead of the dashboard

Test a guardrail

Try a check against sample input before it goes live

Guardrail checks

Browse the first-party and external-provider checks available

Review guardrail activity

Check trigger volume and correct a verdict from the request log

Evaluation

Score the same kinds of risks across a dataset instead of live traffic

Observe

See what your AI app actually did, request by request
Was this page helpful?

Questions & Discussion