I’ve been reading about Latenode’s RAG setup and one thing keeps standing out: you don’t manage the vector store yourself. And I’m trying to understand what that actually means for how I build and think about RAG.
With traditional RAG, a lot of the complexity is around vector databases—chunking documents right, choosing embeddings, managing retrieval quality, all that. But if Latenode handles that part, what does that actually shift? Does it mean the RAG pipeline is simpler? Does it just move the complexity elsewhere, like now I have to be more careful about prompt engineering instead?
I’m also curious whether not managing the vector store limits what you can do. Like, are there advanced RAG techniques that depend on direct vector store control that you can’t do through Latenode?
Who’s actually built RAG this way and noticed a real difference in how they approach the problem?