Debugging Actions in create a GPT

Documentation
, ,
1

I have been trying to get my GPT instance to execute an action correctly, but the lack of debugging tools has made it extremely challenging to pinpoint what’s wrong with my action configuration. I have two pieces of feedback:

  1. The natural language configuration capabilities of GPT are impressive; however, it seems to lack the functionality to configure actions. I would appreciate guidance through the setup process or, alternatively, I simply provide a direct link to the API documentation and it sets it up for me.
  2. When my custom GPT attempts to invoke my action, it responds with “Error talking to…” without providing further details. This lack of information makes debugging quite difficult. Even when inspecting the Network tab in Google Chrome, I’m unable to locate the endpoint it attempted to call. Does anyone have any suggestions?
12 Likes
2

Is there a way to look at the logs of the GPT? I think that would help a ton.

2 Likes
3

That would certainly help. I didn’t see log options like I do in the playground.

4

I’m having the same problem. Additionally, it doesn’t seem like actions are saving for me.

5

I have the same issue where the actions fails in the conversation but no way to debug what’s going on.

1 Like
6

I have the same issue: “error talking to…” and no traces of an attempt. I scanned the destination weblogs, and there are no records. I assume that error happens before an attempt to contact the API server.

1 Like
7

Same issue here as well… seems to not be hitting the endpoint at all.

8

I’m having this same issue. I’m trying to migrate an existing (and working) ChatGPT plugin to be a custom GPT.

It just keeps saying it is having issues communicating.

9

I finally was able to debug what was going on a bit by actually chatting with the GPT Builder. It told me my API response was too large, once I scoped the query down, the agent started to work correctly. To do that I instructed the builder to add query parameters to my API call.

1 Like
10

I have managed to request my Api. But only without the api key. As soon as I set up authentication, I always get the message “Error saving draft”.
Does anyone know why?
My scheme is:

{
   "openapi": "3.1.0",
   "info": {
       "title": "User Data API",
       "description": "API to retrieve user data",
       "version": "v1.0.0"
   },
   "servers": [
       {
           "url": "API:URL"
       }
   ],
   "paths": {
       "\/API\/v2\/users\/": {
           "get": {
               "summary": "Get User Data",
               "operationId": "GetUserData",
               "parameters": [
                   {
                       "name": "location",
                       "in": "query",
                       "description": "...",
                       "schema": {
                           "type": "string"
                       }
                   }
               ]
           }
       }
   }
}
11

Also getting the "“Error talking to…” issue

Anybody find a fix? Quite tricky to debug since I’m not seeing any calls hit our dev server and we don’t have access to logs on the OpenAI side

12

+1 I hope there are tools to check what happened when we got an issue.

13

+1 really hard to know what is going on for the API communication.

14

When I use https://catfact.ninja/fact as api, it works.
But when i build my own server and the same api,https://gpts.idealeap.cn/fact,it failed.
Anyone know what’s the key problem?

17

I finally make it working, here is what I have found.

The editor for the action schema IS NOT UPDATE your recent changes!!! So for every update, I remove previous action schema and re-add a new schema with latest change.

2 Likes
18

well. I resolve my problem. it seems my SSL cert is R3, it can’t work correctly.

19

I’ve had the same experience. Hopefully this is fixed soon as it is a huge pain not being able to save changes to schema.

20

What was the size of your API response? I’m hitting the same issue, and it’s breaking on a 200kB response with about 22k words and 206k characters.

No way to limit the response in this particular API unfortunately, but I haven’t been able to suss out what the response limit for the system is yet.

Edit: Found this in the Plugins documentation, and I imagine actions adhere to this as well

We also impose a 100,000 character maximum for the API response body which may also change over time.

That explains my 206k character limit response being too large for sure!

21

I am also trying the actions, but somehow it does work as expected. Either it is not triggered, or if I need to change the HTTP get/post parameters, I am not sure how to write the schema. Where I can find schema specifications?

22

I am having the same problem here.

When I try my endpoint from my computer it works fine:
curl -H “Content-Type: application/json” http…com/ping
Also when I try it like so:
curl http…com/ping

But it did not work within a “Create my GPT”, and I don’t know how to debug it.