How do plugin updates work, in terms of OpenAI approval?

As far as I can tell, a plugin gets “updated” (i.e. the API spec and ai-plugin.json are re-fetched) every time a user activates the plugin.

If a plugin is approved by OpenAI, is there some obligation to keep these files static? If a plugin developer wanted to make a substantial change to a plugin to add some new functionality, would they have to re-apply for approval? Or is it okay to push updates to an existing plugin on the fly.


I’m not OpenAI but this is the judgement I’d use

  • publishing updates is a fine and normal part of software
  • maintaining backwards compatibility may be necessary for ongoing sessions
  • don’t write plugins with an intent to deceive on future release (OpenAI can always pull your plugin, and it wouldn’t be good for your reputation)

That being said, I’d love to have more dialogue with OpenAI to make sure I’m able to stretch within boundaries and partner well with them.


I’m also interested in finding out how this will work.
But my assumptions align with yours. I think it makes sense to allow frequent plugin updates (I imagine there’ll be an approval process for any update) and updates are fetched when users activate the plugin.

I imagine when it gets closer to general availability of plugins OpenAI will supply us all the procedure for that. @logankilpatrick keeps us all well informed!

I surely hope not. While this may not affect plugins that are simply wrapping existing APIs, it would be hugely detrimental to novel AI plugin experiences where iteration is necessary.

It would make sense to approve at the developer/organization level and re-review if their apps are flagged/reported.

1 Like