Agent Budgets
Daily request, token, and cost limits per agent.
Agent budgets cap what each agent can do in a 24-hour window. Three dimensions run independently — exceed any one and the configured enforcement action fires.
| Dimension | Unit | What it measures |
|---|---|---|
| Request budget | governance checks / day | Verify calls + MCP tool calls + extension events |
| Token budget | tokens / day | Estimated LLM token usage (observability only — not billed) |
| Cost budget | EUR / day | Estimated LLM spend at current model rates |
Enforcement
When a budget is breached, the agent's configured on_exceed action kicks in:
| Action | Behavior |
|---|---|
WARN (default) | Requests still pass; the breach fires a budget-warning notification (webhook/event, deduplicated per day) |
BLOCK | Further requests are rejected — the verify response is 403 with decision BLOCKED and a budget reason |
APPROVAL | Requests are queued for human approval |
Configure per agent under Agents → {agent} → Budgets, via PATCH /api/v1/agents/{id}/budgets (body field on_exceed), or through the budget_* fields on PATCH /api/v1/agents/{id}.
Counter reset
Budget counters are daily (24-hour window). Consumption surfaces in the usage dashboard.
Token and cost figures are estimates — Palveron infers them from the prompt + response payload using the model's published per-token price. For exact billing, reconcile against your LLM provider's invoice.