Glossary
In-process guardrail enforcement
Checking an agent's requests and responses against its rules inside your own service, with no proxy and no network call in the request path.
In-process guardrail enforcement means the check that decides whether an agent's action or reply is allowed runs inside your own application process, not in a separate service your request has to travel to.
The usual alternative is a proxy: every model call goes over the network to a gateway, which checks it and forwards it. That works, but it adds a round trip to every request and makes the gateway part of your critical path.
Why it matters
- No added latency. The rule is evaluated locally, so there's no extra hop. That's what makes guardrails usable on real-time voice agents, where a round trip turns into audible dead air.
- No new single point of failure. If a remote service is slow or down, your agent doesn't stall waiting for it.
- Data stays put. The content being checked doesn't have to leave your service to be checked.
In Zespan, rules compile into a bundle the SDK evaluates in-process, and Zespan syncs the rules ahead of time rather than being called per request. See guardrails without added latency.