Create an ai-plugin.json file for ChatGPT to interact with the API
batpad opened this issue · 0 comments
(technically something we'd implement in the API code, but creating the issue here for more visibility)
ChatGPT has introduced a concept of "Plugins": https://platform.openai.com/docs/plugins/introduction
To create a "plugin", one needs to create an ai-plugin.json file and serve it at https://yourdomain.com/.well-known/ai-plugin.json . This JSON file will contain an "AI-readable" representation of your API, essentially telling an LLM how to query your API, etc. One can then connect ChatGPT to your API and it would attempt to use the API where appropriate to answer questions, etc.
A simple example of an ai-plugin.json file: https://github.com/openai/plugins-quickstart/blob/main/.well-known/ai-plugin.json
We should be able to mostly point it at our OpenAPI API spec definition, and have it mostly infer usage from there. It would greatly benefit from improved documentation describing each endpoint and property returned, etc. - perhaps this would be good incentive for us to also improve our API docs.
This seems like a reasonably low-lift thing to do to be able to start experimenting with using ChatGPT to "query GO". Would love to do this before / during our meet-up, to be able to do some initial experiments and have some discussions around this.