I already know that retrieval storage is priced. However, for the assistant api, there’s no way to see what prompts actually go into gpt-4 when using RAG, so there’s no way to know the actual token consumption, so I can’t decide if it’s worth it. Is there a way to do this? The run-steps log doesn’t give me the information I need.
yup, it doesn’t seem to be documented. There’s a conversation about this here.