Docs
AI Control
Authoring governance policy, testing it, enforcing it, and explaining every decision.
AI Control is where governance is defined and applied.
Policies
A policy binds an effect (allow, deny, warn, audit) to a subject, a target, and optional conditions such as rate limits or token budgets. Policies carry a priority, and priority is how conflicts resolve: when several policies match one invocation, the highest-priority match wins.
Enforcement happens in your cluster
FaburAI is not a proxy in your request path. Policies compile into a bundle; an enforcer running alongside your own proxy pulls that bundle and evaluates locally. Changing a policy changes behaviour without redeploying anything and without adding a network hop to every AI call.
Every decision can explain itself
A decision records more than its verdict. It records which policies were candidates, which triggered, which lost on priority, and which did not apply at all. That is the difference between an audit log and an audit *trail*: you can answer "why was this denied?" without re-deriving the policy set by hand.
Test before you enforce
Gateway Test simulates a single invocation and shows the expected outcome. Policy Impact Simulation asks the broader question — what would this policy change across everything currently in the catalog.