I was experimenting with the agent tool and uploaded a 6000 word document to ask it some questions. I asked less than 12 questions but my usage which is hovering at 0 shot up to 100 calls, and 1 million tokens, and billed me for a $1.
May be an issue with me not understanding what an
assistant is, but Im sure I didnt make 100 requests. Can anyone shed some light on this?
Is an assistant making dozens of API requests at once?
Hi and welcome to the Developer Forum!
What was your agent built to do? is there any kind of looping or logic tree following? As agents can finish early if they detect that they need to make a function call or get new data, then it’s possible to have a bug in that side of your code and run calls many times.
Appreciate that; it was setup to read the document and provide summaries and answer questions about the doc. Nothing fancy since it was my first time using it. Experimenting with the models and switching modes, I turned off its access to file, it said it couldnt summarize anything. Turned it back on, it said it couldnt find any file. I reloaded the page and kept giving me errors on how it could not process my request. So after that set of interactions I’m confused as to how that came up with exactly 100 api requests. EXACTLY 100. lol
What is “agent tool”? In the OpenAI API, there are “assistants”. Assistants can have files attached, which they then use for retrieval and code interpreter if either feature is enabled. They are autonomous and do what they want with your API credits. Assume also that OpenAI is taking your “instructions” only as part of the prompt messages that they are writing.
Essentially: learn to use the standard API and model interactions before you embark on assistants. Then after you understand, read all the documentation about assistants and decide why you would ever use this.
That sort of condescending tone is entirely un useful. If you’ve got nothing helpful to add just keep it yourself. NO WHERE in the
assistant documentation does it explain usage.
One wants to find out of you are even referring to a service that OpenAI offers, or if you are talking about a different site or software that you are using causing woes.
If we are indeed talking about assistants (as has now been edited in) - you can read the experience of others on the forum that have encountered such huge token usages, iterations, and resulting unpredictable billings. It is not condescension: it is a warning.
yeah, you can just hop on over to another thread and forget this one existed. Sounds like youre just bored and looking for validation.