Basically I want my Assistant to use a Tool but directly provide the output of the tool to the user without further processing.
Here’s the use case:
The assistant has a tool where he can make a query and receive a response with citations. This response should be directly given to the user. The assistant could copy-paste it but that would be a waste of time and resources.
How can I do this?
How about just asking as part of the prompt to output, verbatim the responses from the function call? What did you try? I created a new Asisstant added the example temperatur function and simple instructions about function calling:
The issue here (I believe) is the
which I have run into as well. For my use-case I have function-calling create/modify a very lengthy transcript by calling a separate GPT model. I don’t need the Assistant to spend tokens in reading and repeating the transcript verbatim.
Realistically (in my world) the answer is to just give us more control of the Assistants framework and let us send Assistant messages.
How about just cancelling the run after you got your tool response? Nothing will be added to the message in the thread I think.
Hey, did you manage to find a solution to your problem? Were you able to get the assistant to directly return the tool output to the user without involving the LLM afterward?
