I’m using a function with an assistant in my OpenAI API application, and can trigger function execution successfully with a user prompt (e.g. “tell me about products in CATEGORY_NAME”). The function then hits a url in our e-commerce backend and returns JSON, which the assistant then renders verbatim in the chat reply to the user.
I was instead hoping that it would convert the JSON to human readable text, describing the products and perhaps selecting those that appear the best match for the user’s past expressed preferences. This is the behavior we get in our Plugin when hitting the same endpoint, but in the API Assistant application it just renders the raw JSON response for the user.
Any way to fix this? Guessing there is an additional configuration for the function or assistant which may help. Many thanks for any guidance.