January 14, 2026
RAGtag rivalry
Ask HN: How are you doing RAG locally?
DIY AI that 'looks stuff up' splits the crowd: hacks vs toolkits
TLDR: Hacker News asked how to run “AI that looks up your files” locally, and the crowd split. Minimalists vouched for simple setups like SQLite and faiss-cpu, while others favored bundled tools like LibreChat, AnythingLLM, and LightRAG—everyone chasing private, fast answers without cloud complexity.
Hacker News lit up over a simple question: how do you run RAG—“Retrieval-Augmented Generation,” aka AI that looks up your own files—on your laptop without a tech circus? The thread instantly split into two camps. Team Minimalist came in hot with “just use faiss-cpu and pickle,” which one commenter translated into plain speak as: keep it tiny, keep it fast, keep it local. Another went even leaner with SQLite + FTS5, old-school text search that looks for words instead of meaning. They basically said, why build a rocket when a bike gets you there.
Meanwhile, Team Toolkit flexed the all-in-one setups: LibreChat bundling a vector database (think “search by meaning” instead of exact words), AnythingLLM as the shiny new contender, and LightRAG with an Archestra UI for those who like buttons, dashboards, and vibes. The drama? Minimalists rolled their eyes at “vector cult” complexity; tool fans clapped back that one-click systems save time and sanity. The jokes kept coming: “pickle jar method” vs “Big Tool Energy.” Underneath the memes, the mood was clear—people want private, local AI that understands their docs without summoning the cloud. The thread felt like a kitchen showdown: spice rack vs meal kit, both trying to serve answers fast and fresh.
Key Points
- •The post asks how to implement RAG locally with minimal dependencies.
- •It targets use cases involving internal codebases and complex documents.
- •It inquires about the use of vector databases in local RAG setups.
- •It asks whether semantic search is employed as an alternative or complement.
- •It explores whether knowledge graphs or hypergraphs are part of local RAG architectures.