Budget
This is one policy kind in the ADR-014 governance engine. Policies resolve most-specific-wins with deny-wins at a tie, across scopes that run agent → user/role → workspace → team → org. Rules of the same kind combine rather than replace, so what matters is which rules exist, not which policy row holds them.
What it does
A budget policy caps spend over a period. Its scope picks the basis — the whole org, a team's members, or one user — and the winning policy's scope is the one that is measured against the usage ledger.
Soft and hard
| Mode | Effect |
|---|---|
| Soft | The run proceeds; the owner is notified once per period. |
| Hard | A chat send is refused before any run exists (HTTP 402); non-interactive paths fail the run cleanly with a budget reason. |
As a floor
A budget declared a floor binds every narrower scope. A team may set a
smaller amount, never a larger one; on_hard may only get stricter;
and a floor at a different period stays in force beside the narrower
budget, so a team's daily cap under an org monthly floor means both are
checked and either stops the run.
Where it is measured
Every run writes a usage-ledger row through the one execution path, so a budget covers chat, automations, channels, orchestration and A2A alike — there is no path that spends without being counted. Cross-provider routing and failover happen underneath, so cost is tracked whichever provider served the call.
