I think with RAG it's pretty reasonable. Put your corpus in pinecone or some other vector store and relevant sections get injected along with your prompt which lessens the burden on context window.