Actions test fine, fail on first attempt in chat interface

GPT builders
1

I have an OpenAPI spec that I’ve had in place for some time now and it was working flawlessly. Recently, the chat interface would error out on calls. When I open up the builder to test the calls, they all work every time. Additionally, if I use the debug chat interface, they work without fail. It’s only when I use the user-facing chat interface that the calls end in error the first time in that chat session. Also, I’ve confirmed that no call to my API is actually happening via the logs. Subsequent calls to any action succeed.

The error in the chat interface simply says “error taking to” after I confirm access.

“It seems there was an issue retrieving the registration data for N472TW. You may try again later, or if you need further assistance, feel free to ask!”

I’ve double checked using postman and all is well.

Anyone else experiencing this? I’ve had to add content to my GPT instructions to advise users to try the call more than once, which is not great UX.

I’m using a bearer token for auth.

1 Like
2

I have had the same experience as you. Working with the custom gpt functionality for a while now, I have come to the conclusion, that it is very inconsistent.

Your gpt might work perfectly as intended one day, and the next - without any changes - errors start to occur.

It has been frustrating developing, and I have advised my employer not to invest more resources into it for now, or at least be prepared that the product is far from being in a mature stage.

1 Like