The premise (as described by @pelaseyed) is that traditional RAG sucks because it promises "relevant chunks" but in fact returns "similar chunks" and that actua

The premise (as described by @pelaseyed) is that traditional RAG sucks because it promises "relevant chunks" but in fact returns "similar chunks" and that actual relevancy evaluations requires reasoning. No chunking, splitting or vectorizing Stateless, no vector DBs etc. Supports any model (deepseek, o3-mini et al) Reasoning traces Metadata filtering Typescript, Python support