Currently, valid models are hardcoded in the adapter(s) and checked by exit.lua#set_model.

This requires us to release a new version of exit.nvim every time a provider (eg, openai) releases a new model.

One way to fix this would be to check only the validity of the adapter, allow any model, and check in the API response if an invalid model was used.

Example:

• 0p3n4i:gpt-3.5-turbo should error and list the valid adapters (eg, openai, ollama)
• openai:gp1-3.5-1urb0 should be allowed, but the request would fail with something like "model 'openai:gp1-3.5-1urb0' is not valid, please select a valid model"; a list of valid models could be returned or a URL to the docs where to find them
• openai:gpt-3.5-turbo would work as usual

Another solution would be to contact an API endpoint when setting the model to check what models are available for that adapter.