Too much tokens spent on OpenAI agent

Hi ! I'm surrently building hierarchical agent teams for a small project of mine, I'm using LangGraph as framework and OpenAI's API for the LLM.

But i've noticed that when creating agents the number of tokens used (as input) increased drastically, going from 15000 to 19000 tokens per ask, which then increases the costs. So i was wondering if anyone has tips or ressources as to how we can optimize the agents to decrease the number of tokens per ask while maintaining a good accuracy within their reasoning and action.

Oh and if anyone is currently building agent as well, feel free to dm me i'd love to connect with y'all !

Thanks a lot and have a great day 😊

0 comments