RAG commonly implies some sort of vector database to be built and which will then be used for response augmentation. If it operates over the repo, I believe it will index your codebase using those vector embeddings.