Using a LLM from huggingface, I send it the prompt and it returns a configuration for how the character should behave.
The configuration itself is just a nested JSON that tells which preconfigured visuals and event handlers to use :)
