Guardrails.
Block bad outputs at the request layer, not after the fact.
Guardrails enforce output quality and safety at the request layer. PII redaction, JSON-schema validation, regex blocks, content moderation, and custom JS evaluators all run on the response before it returns to the caller. Pairs with Gateway, so the guard runs once per call at proxy latency, not as a post-hoc audit job that catches violations after they've already shipped.
Configure, enforce, audit.
Pick PII redaction, JSON-schema validation, regex blocks, content moderation, or custom JS guards per route.
Guards run on the response at Gateway latency, before output ships to the caller.
Every block, redaction, and override is logged with the policy that triggered it.
What it controls.
Emails, phone numbers, SSNs, custom patterns scrubbed from the response before output ships.
Require the response to parse as JSON and carry the top-level keys you name. Violations warn or block at the Gateway.
Match-and-block patterns at request time, not after the fact in dashboards.
Prompt-injection, jailbreak, and abuse patterns matched on the response, with a per-rule match threshold. Deterministic, so a flag always explains itself.
Write arbitrary validation in JS or TS. Runs at Gateway latency, not a separate service.
Who reaches for Guardrails.
- ·Security and compliance teams
- ·Platforms serving regulated industries
- ·Anyone shipping AI to external users
Pairs with the rest of Treasury.
Want Guardrails in your stack?
We're onboarding design partners now. Join the waitlist to be in the Guardrails cohort.
Just email is required. One email when Guardrails goes live. Nothing else.