GPT Builder says that the request to API was sent, but API logs show that it was not

GPT builders Plugins / Actions builders
, , , ,
1

I’m experiencing an odd issue in GPT Builder’s test mode. After making unpublished changes to instructions and API schema, GPT Builder shows it’s sending requests (“[debug] Calling HTTP endpoint”), but these aren’t logged in my API’s access logs. Despite indications from the GPT Builder (showing “[debug] Calling HTTP endpoint”) that the request was properly sent, the access log of my API does not record any incoming request, suggesting it was never received.

Additional context:

  • The API works correctly when tested independently.
  • Another GPT, when tested with a different server URL in the API schema, functions without issues.
  • The production environment, using the old API schema but the same server URL, operates smoothly.

This issue might be relevant to the test mode when changes are unpublished. I am hesitant to risk moving these changes to production due to the ongoing errors, so I do not know if those errors still going to be present if I publish those changes.

I suspect either an hallucinogenic indication from GPT Builder about sending the request through openai.com/bot, or a mistake in how it provides task to openai.com/bot, preventing it from reaching my API.

Has anyone experienced a similar issue or have any suggestions for resolution?

Thank you for any insights or help.

1 Like
2

Same here. The test mode only says “error when communicating with [blank here]”, but not mentioning to whom it’s communicating.

2 Likes
3

@FlyinDeath
I just was able to reproduce this bug in another GPT that I created for test purposes. This bug happens when there are significant changes of instructions and API schema and when those changes are unpublished.
So for example I changes instructions #1 and API schema #1 to instructions #2 and API schema #2, this bug was reproduced. Then I published this changes, and the bug disappeared. Then I changed back instructions #2 and API schema #2 to instructions #1 and API schema #1, and did not publish changes and this bug came back. Now I hit the usage cap and need to wait for couple of hours to check if publishing the last changes again will treat this bug :slight_smile:

1 Like
4

Thanks a lot for the tip. I’ve confirmed that publishing the changes can work around this problem. At least I get a way to continue testing now. :slightly_smiling_face: Hope that OpenAI fix this problem soon.

2 Likes
5

@FlyinDeath you are welcome :slight_smile:

6

Continuing the story:
When I published changes again, the bug was fixed. So, according to those experiments, this bug only occurs when changes are unpublished.