budgeted-agent

Overview ¶

Command budgeted-agent demonstrates LoopKit's budget enforcement.

Overview ¶

This example creates a loop with a $0.50 USD budget and a scripted fake provider whose responses carry realistic token usage. The custom pricing table makes each model call cost $0.20, so the loop is forced to exit with a BudgetExceeded outcome before completing its intended work.

Key concepts demonstrated ¶

• loopkit.WithBudget: set USD (and optionally token/iteration/wallclock) caps.
• Custom model.Table: supply your own per-model pricing so spend accumulates predictably in tests and examples.
• outcome.Match: exhaustive dispatch over all six terminal outcome types, used here to print a human-readable summary.
• Scripted fake provider: inject deterministic usage without a live API.

Expected output ¶

BudgetExceeded: dimension=usd spent=$0.40 (2 model calls × $0.20 each)

The loop requests 3 model calls but the $0.50 budget is exhausted after the second call charges $0.40 total — the third call's pre-authorization fails.