Reasoning-based retrieval

Document retrieval forthe reasoning era.

No chunking. No embeddings. No vector DB. Vectorless parses documents into structured maps any LLM can navigate — precision retrieval with citations you can trust.

quickstart · sdk
// 1. Install
npm i @vectorless/sdk

// 2. Ask with citations
const { answer, citations } = await vl.ask(doc, "what changed?")
The shape of an answer

One document, parsed into structure, walked to the node.

We don't guess similarity. We reason through your document's actual hierarchy to find the exact branch where the truth lives.

100%
Citation accuracy
~40ms
Retrieval speed
DOCUMENTTREEWALKCITED ANSWER
The Primitive

Retrieval, rebuilt around reasoning

Explore the core concepts
01

Tree, not chunks

Vectorless parses a document into a hierarchical tree that preserves its real structure — sections, sub-sections, tables. No fixed-size chunking, no lost context.

02

treewalk navigation

An LLM agent walks the tree node by node, reasoning about where the answer lives. Retrieval becomes a navigation problem, not a nearest-neighbor lottery.

03

Citations by construction

Every answer traces back to the exact nodes it came from. Path-correct citations are a property of the engine, not a bolt-on afterthought.

04

No vector DB to run

No embeddings to compute, no index to maintain, no similarity threshold to tune. Point Vectorless at a document and ask.

Retrieval stopped being a search problem. It became a reasoning problem.
The Vectorless thesis