I am building a custom GPT and observe the following behaviour: When my action (calling a REST API endpoint) is performed, it will instantly fail if there are params attached to the request.
Examples
(note: the domain is changed for privacy reasons)
This URL is structured correctly and should definitely work. My server logs do not show any errors (or even access), so I believe ChatGPT does not even send any request.
I deleted my action several times and added it again, but the result is the same.
Use https://www.postman.com/ (or something similar) to make sure that URL format is correct. I it is not, you will see the exact message being returned by the server.
If at all possible to access, review the umami.example.com server logs for the requests to see what error, if any, is being thrown.
This is how I’ve been able to troubleshoot GPT actions.
Thank you. But no, server logs are completely empty - nothing in the error log, and nothing in the access log. This is why I believe that the requests do not get executed on the OpenAI/ChatGPT side.
In the OpenAPI schema, there is this “server” object:
"servers": [
{
"url": "https://umami.example.com/jsonapi",
"description": "The current environment"
}
],
Every path in the schema is therefore relative to /jsonapi. So /node/recipe becomes /jsonapi/node/recipe and so on. This, however, works, ChatGPT simply does not include the server url in the debug message and just prints the domain name and the action’s path.