Every LLM call gets logged with input tokens, output tokens, and the price-per-million snapshot from the moment it ran. A user blasting through your free tier? You can see it in real time and shut them off.

This is the pattern Petal Pro ships with. The Req plugin attaches to your LLM client and writes an ai_call_logs row on every response. Pricing is snapshotted per-call, so an OpenAI price drop next month doesn’t retroactively rewrite last week’s history.

Pair this with a daily budget check before kicking off any expensive job.