[esafak]

That's only true if you can train or fine tune the LLM. If you are merely a user augmenting it with RAG then GraphRAG is perfectly viable.