Every usage-priced platform says it has budgets. The test is simple: what happens on the call that would cross the line? If the answer is "an alert fires", the budget is decorative.
In AstroFabric, the plan's data envelope is written onto the workspace and enforced by a cost ledger. Every paid call reserves its estimated cost before executing and settles or releases after. When the reservation would cross the ceiling, the call returns a structured 402 your code can catch. Nothing about that path is advisory.
Degrade in the right order
A hard stop at 100% is necessary, but the road to it matters. As the month fills, the platform drops freshness first (optional enrichment stops, sampling slows), then fidelity (classification moves to a cheaper model), and only at the end coverage. A customer would rather have two models daily than four models weekly, so the ladder is ordered to protect the daily habit. The ordering is pinned by a monotonicity test in CI.
The same ledger meters revenue: tool calls and agent runs land on usage meters that flow to billing, so what you are charged and what the ledger governed are one number, not two systems drifting apart.
Every playbook on this blog ships as a runnable mission.
Open a free workspace and the playbook library is waiting - describe the outcome and the agents carry it end to end, with 1,000 trial credits on us.