What's this about?
OpenAI's dashboards now break down usage and spend by individual API key, so you can see which app or workload is driving cost, and you can set monthly organisation or project limits — including hard limits that stop traffic when reached.
What exactly is a "hard limit"?
A soft limit warns you; a hard limit stops the traffic. This matters more for agents than for ordinary software, because an agent stuck in a retry loop can burn a very large amount of money overnight without anything appearing broken. The ability to attribute that spend to a specific key is the difference between finding the culprit in minutes or days.
What's the community saying?
Reaction is straightforwardly positive from developers, and the recurring note is that this should have existed from the start — runaway agent spend is a common enough experience that the feature reads as overdue rather than innovative.