Suggestion: Faster written responses when using values "as-is" from an endpoint

GPT builders Plugins / Actions builders
1

I’m currently developing a plugin that allows users to generate charts using Chat-GPT.

We have an API endpoint that accepts JSON data and returns two URLs, one for a static image, and one for an interactive version of the chart. For privacy reasons, we don’t store the data on our side, we embed it in the URLs which has the side-effect of making them quite long.

The long URLs mean a poor user experience when using the plugin. Chat-GPT does a great job of generating the correct markdown in the response, however the time taken to “type out” the markdown including those long URLS can be upwards of 30 seconds.

It would be great if this could be sped up somehow. Chat-GPT already knows the URLs as they are returned in the response, so it feels like unnecessary work for it to do to just display them “as is”.

My naive suggestion would be whether it might be possible for the plugin system to prompt Chat-GPT to generate markdown with placeholders like:

![Image of a chart]({{response.data.imageUrl}})

And then replace these on the fly with values from the response. This could potentially be useful for speeding up the time to a useable response in situations where the data that is returned from an API doesn’t need to be rewritten in any way.

1 Like
2

Could you store the data (encrypted and for a limited time), then send the user the key in the response and delete it on your end? Obviously not actually very secure, but they’re already trusting you with the data, and there’s some ways you could make it more trustworthy if you had some kind of user authentification?

Then the links would just be the decryption key+a hash, and you could send the data with the page to be decrypted on the users end.

1 Like
3

They’d have to actually visit their page on your site though, or you’d need to figure out some way to run js within the chatgpt interface. Unsure if that’s supported or not, now that I think about it, but the interface is a bit on the fragile side so I wouldnt be surprised if there’s a way.

Edit, not that I’m recommending you go against the policies. If that’s forbidden dont do it.

4

Thanks for the suggestion @chrstfer. I think your solution would work well to resolve our specific issue in the short term, so I’m going to check it out!

1 Like
5

No problem, feel free to ping me here if you need anything or want to brainstorm.