AI Agent Budget Controls

An agent that is permitted to call a tool is not permitted to call it a thousand times. EVE enforces deterministic budget and resource limits: a permitted call executes until the configured limit is reached, then subsequent calls are blocked. Counters are deterministic, so the limit behaves predictably.

What budget controls cover

Deterministic limits — a configured budget or resource limit is enforced in the decision path; the same sequence of calls hits the limit at the same point every time.
Block on exceed — once the limit is reached, further permitted calls resolve to BLOCK, with a finding that records the limit, the consumed amount, and that it was exceeded.
Signed findings — budget findings can be captured in signed evidence for review.
Composable with other controls — budget limits combine with sequence governance and anomaly detection.

Evidence example

budget finding { limit: 100, consumed: 100, exceeded: true }
# the 101st permitted call resolves to BLOCK

SDK context

Budget enforcement runs in the same governance path as tool-call decisions:

context = {"tenant_id": "org_abc", "principal_id": "agent_worker", "session_id": "sess_8"}
# repeated eve.govern_tool_call(...) with this context is counted against the configured budget

Readiness

Deterministic budget/resource limits are PILOT_READY, part of the pilot-validated core governance architecture and covered by the agent-governance test suite (Python; embedded-service mode).

Limitations

Budget counters are deterministic per instance; durable cross-instance budgets require a shared store.
Budget enforcement is available in the Python embedded-service mode; wiring it into your agent loop is an integration step.
Budget controls limit call volume and resource use; they do not judge the correctness of an individual permitted call.

Next step

Configure budgets for your agents in a pilot and confirm the block point with signed budget findings.