GPT oauth callback URL keeps changing

Not sure what I am doing to trigger a change, but it seems like any time you add or remove an action the callback URL completely changes. This makes it quite painful to develop with if you continuously need to be updating that. Anybody else seeing the same thing?

7 Likes

I’m seeing this as well. It seems to randomly add/duplicate actions as well which updates the callback URL.

1 Like

:neutral_face: Yes very annoying since my backend server uses global caching and changing the callback URI in OAuth2 settings triggers a 15 minute (minimum) wait time to propagate. So annoying.

Hey @scott420 can you add a “feedback” tag to your topic post? That way the product team at OpenAI might pickup on the feature request here. I guess in some ways it might even be a bug.

Can you elaborate—are you hard-coding the callback URL from the GPT Editor into your app somewhere? In case you are, OpenAI will call your /oauth/authorize (or other authorization URL) with redirect_uri in the query string. You should be returning the user to that URL programatically.

Most oauth services like Auth0 require the callback urls to be whitelisted. So each time it changes in the OpenAI interface you have to copy it and paste it in your app configuration… gets to be a little repetitive.

2 Likes

In my case I am accessing services inside Salesforce. To allow any outside entity like OpenAI to authenticate, you create a “Connected App” definition in Salesforce whereby you cut/paste the callback URi and this must match EXACTLY the URI specified by OpenAI on the very first authorize HTTP call into Salesforce. It gets better – the authorize URL is a central load-balanced global cluster with DNS routing so in other words my URI and Oauth2 config is replicated internal to Salesforce and hence takes 15 min or more to sync and be usable. So by definition of be in R&D development mode the OpenAPI spec is changing all the time in the Action editor and hence (seemingly) the callback unique id inside the OpenAI URI is regenerated and changed constantly.

This makes development take hours and hours due to global cache update wait times.

2 Likes

I thought no one shared my struggle and the time it takes to commit the callback URL! I was trying to demo to my team and looked like a moron because the URL changed and I couldn’t authenticate anymore. Working in SFDC also and the changing callback URL is a challenge, @Emilio it might make more sense to create an API user? Would that still require the call back URL?

1 Like

Not sure about the API user. There is another compounding issue I am now seeing which is when the auth token is expired and/or the session expired overnight, and GPT is asked to do something that requires an action, it fails auth and just gives up and tells the business user “go login again to Salesforce and come back here”. Which I am like “ummmm really?”.

The struggle is real.

Been struggling with this too and it’s really making POC / dev of GPT’s very tricky.

The callback URI seems to remain static as long as any updates you make to the schema are valid. As soon as an invalid schema is detected (automatically) the callback URI changes. To replicate simply delete the servers: URL from your schema and observe the callback URI will have changed. This feels like a bug to me @logankilpatrick any chance of getting eyes on this?

1 Like