For most use cases, a Large Language Model (LLM) drives the interactive avatar.