In this mode, the Assistant is equipped with behavior tuning information derived from a JSON-based configuration file, guiding its incorporation of relevant product knowledge graphs when responding to prompts. This process of loosely-coupling a Large Language Model (LLM)-based Assistant and a Knowledge Graph is generally known as Retrieval Augmented Generation (RAG).
Here are examples of the aforementioned JSON-based configuration files that offer powerful templates for configuring the behavior of the Assistant: