Currently you can configure the prompt, and they'll use the default chat model.
We should also be able to configure the model (although leaving the ability to use the default in place makes sense) and set the temperature like you can in the AI explorer.