This is for LLMs. In general RAG takes a user prompt and uses it to find potentially relevant documents in the database. It then enriches the original prompt with those documents so that the LLM has context that wasn't in its training dataset.