Yes you can do this, I had a problem where my GPT was far too eager to send POST requests to my API. I gave it a little bit of instruction telling it to not be too eager to create things via the endpoints, and to wait until it had a good understanding of what the user wanted. That did the job, although you’d probably want to define what a ‘good understanding’ means. Generally speaking if your API is documented well it should mostly just work, other than the eager-posting, which you deal with via instructions.
Just curious @callum.bradbury do you use a Markdown format for your instructions? I have been using unordered and ordered lists, headers and other “markdown like” syntax in the instructions to properly gate POST requests but haven’t yet reached true reliability. Wondering if there is a specific syntax that does the trick.