How do you handle agent spend and runaway tool calls?

Quick question for people running CrewAI agents in real workflows:

How do you control costs when agents can call tools or paid APIs?

Especially:

  • per-agent budgets
  • hard stops if something loops
  • visibility into what each agent spent

Right now is it mostly manual monitoring, provider limits, or do you have actual controls set up?

Want to hear what’s working in practice.