Core Concepts
Overview
The ideas behind Vectorless, and how they differ from vector RAG.
Updated 2026
Vectorless rests on four ideas. How it works shows them running together as one pipeline.
The contents tree
A document's own table of contents, placed on its pages, is the index.
The no-chunking model
Why there are no chunks, no embeddings and no vector database.
Pincites
Each claim in an answer pinned to a line on its page, with no generated text in the citation.
treewalk, the generative fallback
The older strategy in which a chat model walks the pages, used when Jev is unavailable.