How to interact with Custom GPT using API edpoints

API
1

How can I interact with custom GPT models hosted on OpenAI using their API endpoints in a way that is compatible with Atlassian Forge’s runtime and authentication model?

2

Generally: GPTs are not “available” on the OpenAI API. It is a feature of the consumer ChatGPT product.

First, we must define what you are discussing here.

I suspect that by the term “custom GPT”, you are referring to what are simply "GPT"s - created in ChatGPT.

That is not creating a special model when you build one (building either talking to the GPT builder AI about what you want, or filling in the varying text boxes). In fact, it is powered exclusively by instructions, ChatGPT-exclusive internal tools such as 4o-image-gen or canvas, and files that you may upload. It is paid for by the subscription of the person using it in ChatGPT, for example, if you were to share to “store” for others.

Your GPT is likely just instructions. You can review the fields of the GPT, and see what is there to take with you to a new platform.

This image shows a configuration screen for creating a custom GPT, with a cosmic and cryptic oracle-themed description and instructions, along with selectable capabilities like web search, canvas, image generation, and code interpreter. (Captioned by AI)
This image shows a configuration screen for creating a custom GPT, with a cosmic and cryptic oracle-themed description and instructions, along with selectable capabilities like web search, canvas, image generation, and code interpreter. (Captioned by AI)629×580 12.5 KB

Are you familiar with the OpenAI API at all? It is pay per use, for the data consumed, and requires prepayment.

You have a wide selection of AI models of varying quality and cost to use.

And then, to customize the behavior, you would start with “system messages”, a guidance to the AI that is similar to the textual instructions of a GPT, but oriented towards developer control of a model.

Also necessary is customer-oriented databases for their authentication and payment/billing, tracking sessions of conversations to have a “chatbot” experience, developing a secure backend, etc. There is no “paste a plugin window into your web page for a free ChatGPT experience”.

For Atlassian, here is a demonstration of the complexity needed just for API AI to perform a single task for a web page developer: https://developer.atlassian.com/platform/forge/build-jira-comments-summarizer-with-openai/#build-a-jira-comments-summarizer-app-with-openai

Hope that’s enough information to let you explore more and see if your vision is within the realm of possibilities for you.

1 Like
3

This image shows a list of AI model rate limits, displaying tokens per minute (TPM) and requests per minute (RPM) for various models in a project settings interface. (Captioned by AI)
This image shows a list of AI model rate limits, displaying tokens per minute (TPM) and requests per minute (RPM) for various models in a project settings interface. (Captioned by AI)1343×830 35.7 KB

You actually do get access to all major OpenAI models — including GPT-3.5, GPT-4.1, GPT-4o, embeddings, and Whisper — directly through the API.

and more as you can see in the screenshot, tho if you refer to gpts built by users that would be different.

1 Like