Vectorless
Getting Started

Overview

What Vectorless is, who it's for, and how the pieces fit together.

Updated 2026

This section gets you from zero to your first cited answer. If you just want the commands, jump to the Quickstart.

The mental model

Working with Vectorless is three steps:

Ingest a document

You hand Vectorless a document (PDF, HTML, Markdown, …). The engine parses it into a tree — preserving sections, sub-sections, and tables. There is no chunking step and nothing to embed.

Ask a question

You send a natural-language question. An LLM agent runs the treewalk strategy, navigating the tree to the nodes most likely to contain the answer.

Get an answer with citations

You receive an answer plus the exact tree nodes it was derived from — so every claim is traceable back to the source.

What you'll need

  • An LLM provider key (the agent does the reasoning).
  • One of the official SDKs, or direct access to the HTTP API.
  • A document to ask questions about.

The retired name "pageindex" referred to an earlier version of this navigation idea. The engine's canonical strategy is now called treewalk — use that name everywhere.

Next steps

On this page