Guard.
Spend policies that enforce themselves.
Guard is enforcement on top of Treasury's tracking. Budget caps that actually cap. Kill switches for runaway loops. Approval flows for unusual spend. Everything opt-in, everything logged, nobody silently blocked.
Arm, enforce, override.
Set budget caps, kill switches, and approval thresholds.
Policies act on their own, with every action logged.
Humans can always step in. Nothing is silently blocked.
What it controls.
Hard and soft caps per team, project, or developer. Warn, throttle, or block, opt-in only.
Loop detection that pauses an agent mid-spiral and pages a human.
Spend over a threshold queues for approval instead of executing.
Every policy change, every enforcement action, every override. Immutable.
Who reaches for Guard.
- ·Platform teams running shared AI infra
- ·Finance owning the AI budget
- ·Security reviewing agent permissions
Pairs with the rest of Treasury.
Want Guard in your stack?
We're onboarding design partners now. Join the waitlist to be in the Guard cohort.
Just email is required. One email when Guard goes live. Nothing else.