I build Valta, a proxy that denies calls once an agent run exceeds a cap. CrewAI goes through LiteLLM, so pointing OPENAI_API_BASE at https://valta.co/v1 should work, but I haven’t tested CrewAI yet, which is why I’m posting. If you run a crew with a loop-prone task, would you try it and tell me what breaks?
Limits: OpenAI models only, non-streaming, chat completions only. Guide: Set up the OpenAI proxy (8 steps to your first deny) | Valta Docs