Retrieval-augmented generation (RAG) has become the de facto standard for grounding large language models (LLMs) in private ...