Skip to content

Commit 45273b3

Browse files
authored
Merge pull request #106 from vectorlessflow/dev
Dev
2 parents fb28c3a + 9ca2fc3 commit 45273b3

5 files changed

Lines changed: 17 additions & 50 deletions

File tree

README.md

Lines changed: 6 additions & 37 deletions
Original file line numberDiff line numberDiff line change
@@ -2,8 +2,8 @@
22

33
<img src="https://vectorless.dev/img/with-title.png" alt="Vectorless" width="400">
44

5-
<h1>Reasoning-based Document Engine</h1>
6-
<h5>Reason, don't vector · Structure, not chunks · Agents, not embeddings</h5>
5+
<h1>Document Understanding Engine for AI</h1>
6+
<h3>Reason, don't vector · Structure, not chunks · Think, then answer</h3>
77

88
[![PyPI](https://img.shields.io/pypi/v/vectorless.svg)](https://pypi.org/project/vectorless/)
99
[![PyPI Downloads](https://static.pepy.tech/badge/vectorless/month)](https://pepy.tech/projects/vectorless)
@@ -14,29 +14,15 @@
1414

1515
</div>
1616

17-
**Vectorless** is a reasoning-native document engine written in Rust. It compiles documents into navigable trees, then dispatches **multiple agents** to find exactly what's relevant across your **PDFs, Markdown, reports, contracts**. No embeddings, no chunking, no approximate nearest neighbors. Every retrieval is a **reasoning** act.
17+
**Vectorless** is a document understanding engine for AI. It compiles documents into structured trees of meaning, then dispatches multiple agents to reason through headings, sections, and paragraphs — evaluating how each part relates to the whole. The problem it solves is not "where to look", but "what does this mean in context". Every answer is a reasoning act, not a retrieval result.
1818

1919
Light up a star and shine with us! ⭐
2020

2121
## Three Rules
22-
- **Reason, don't vector.** Retrieval is a reasoning act, not a similarity computation.
22+
- **Reason, don't vector.** Understanding is reasoning, not similarity.
2323
- **Model fails, we fail.** No heuristic fallbacks, no silent degradation.
2424
- **No thought, no answer.** Only reasoned output counts as an answer.
2525

26-
## Why Vectorless
27-
28-
Traditional RAG systems split documents into chunks, embed them into vectors, and retrieve by similarity. Vectorless takes a different approach: it preserves document structure as a navigable tree and lets agents reason through it.
29-
30-
| | Embedding-Based RAG | Vectorless |
31-
|---|---|---|
32-
| **Indexing** | Chunk → embed → vector store | Parse → compile → document tree |
33-
| **Retrieval** | Cosine similarity (approximate) | Multi-agent navigation (exact) |
34-
| **Structure** | Destroyed by chunking | Preserved as first-class tree |
35-
| **Query handling** | Keyword/similarity match | Intent classification + decomposition |
36-
| **Multi-hop reasoning** | Not supported | Orchestrator replans dynamically |
37-
| **Output** | Retrieved chunks | Original text passages, exact |
38-
| **Failure mode** | Silent degradation | Explicit — no reasoning, no answer |
39-
4026
## How It Works
4127

4228
### Four-Artifact Index Architecture
@@ -60,7 +46,7 @@ DocumentTree NavigationIndex ReasoningIndex Do
6046

6147
This separation means the agent makes routing decisions from lightweight metadata, not by scanning full content.
6248

63-
### Agent-Based Retrieval
49+
### Agent-Based Understanding
6450

6551
```
6652
Engine.query("What drove the revenue decline?")
@@ -74,7 +60,7 @@ Engine.query("What drove the revenue decline?")
7460
│ │
7561
│ └─ evaluate ── insufficient? → replan → dispatch new paths → loop
7662
77-
└─ Fusion ── dedup, LLM-scored relevance, return with source attribution
63+
└─ Synthesis ── dedup, evidence scoring, reasoned answer with source chain
7864
```
7965

8066
Worker navigation commands:
@@ -114,23 +100,6 @@ async def main():
114100
asyncio.run(main())
115101
```
116102

117-
## Key Features
118-
119-
- **Rust Core** — The entire engine (indexing, retrieval, agent, storage) is implemented in Rust for performance and reliability. Python SDK via PyO3 bindings and a CLI are also provided.
120-
- **Multi-Agent Retrieval** — Every query is handled by multiple cooperating agents: an Orchestrator plans and evaluates, Workers navigate documents. Each retrieval is a reasoning act — not a similarity score, but a sequence of LLM decisions about where to look, what to read, and when to stop.
121-
- **Zero Vectors** — No embedding model, no vector store, no similarity search. This eliminates a class of failure modes: wrong chunk boundaries, stale embeddings, and similarity-score false positives.
122-
- **Tree Navigation** — Documents are compiled into hierarchical trees that preserve the original structure — headings, sections, paragraphs, lists. Workers navigate this tree the way a human would: scan the table of contents, jump to the relevant section, read the passage.
123-
- **Document-Exact Output** — Returns original text passages from the source document. No synthesis, no rewriting, no hallucinated content. What you get is what was written.
124-
- **Multi-Document Orchestration** — Query across multiple documents with a single call. The Orchestrator dispatches Workers, evaluates evidence, and fuses results. When one document is insufficient, it replans and expands the search scope.
125-
- **Query Understanding** — Every query passes through LLM-based intent classification, concept extraction, and strategy selection. Complex queries are decomposed into sub-queries. The system adapts its navigation strategy based on whether the query is factual, analytical, comparative, or navigational.
126-
- **Checkpointable Pipeline** — The 8-stage compile pipeline writes checkpoints at each stage. If indexing is interrupted (LLM rate limit, network failure), it resumes from the last completed stage — no wasted work.
127-
- **Incremental Updates** — Content fingerprinting detects changes at the node level. Re-indexing a modified document only recompiles the changed sections and their dependents.
128-
129-
## Supported Documents
130-
131-
- **PDF** — Full text extraction with page metadata
132-
- **Markdown** — Structure-aware parsing (headings, lists, code blocks)
133-
134103
## Resources
135104

136105
- [Documentation](https://vectorless.dev) — Guides, architecture, API reference

docs/blog/2026-04-12-welcome/index.mdx

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -2,22 +2,22 @@
22
slug: welcome
33
title: Welcome to Vectorless
44
authors: [zTgx]
5-
tags: [vectorless, rag, llm, announcement]
5+
tags: [vectorless, document-understanding, llm, ai, announcement]
66
---
77

8-
Vectorless is a reasoning-native document intelligence engine written in Rust — **no vector database, no embeddings, no similarity search**.
8+
Vectorless is a document understanding engine for AI. It compiles documents into structured trees of meaning, then dispatches multiple agents to reason through headings, sections, and paragraphs — evaluating how each part relates to the whole. The problem it solves is not "where to look", but "what does this mean in context". Every answer is a reasoning act, not a retrieval result.
99

1010
{/* truncate */}
1111

1212
## Why Vectorless?
1313

14-
Traditional RAG systems rely on vector embeddings and similarity search. This approach loses document structure, requires a vector database, and often returns chunks that lack context.
14+
Understanding a document requires more than finding keywords — it requires navigating structure, cross-referencing sections, and evaluating whether the evidence is sufficient. Vectorless agents do exactly this: they reason through documents the way a human expert would.
1515

16-
Vectorless takes a different path:
16+
Key capabilities:
1717

1818
- **Hierarchical Semantic Trees** — Documents are parsed into a tree of sections, preserving structure and relationships.
1919
- **LLM Agent Navigation** — Queries are resolved by agents that navigate the tree using commands (ls, cd, cat, find, grep), making every decision through LLM reasoning.
20-
- **Zero Infrastructure**No vector DB, no embedding models, no similarity search. Just an LLM API key.
20+
- **Zero Infrastructure** — Just an LLM API key, nothing else to deploy.
2121

2222
## Quick Start
2323

docs/docs/intro.mdx

Lines changed: 3 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -4,9 +4,7 @@ sidebar_position: 1
44

55
# Introduction
66

7-
**Vectorless** is a reasoning-native document intelligence engine written in Rust — **no vector database, no embeddings, no similarity search**.
8-
9-
It transforms documents into hierarchical semantic trees and uses LLMs to navigate the structure, retrieving the most relevant content through deep contextual understanding instead of vector math.
7+
**Vectorless** is a document understanding engine for AI. It compiles documents into structured trees of meaning, then dispatches multiple agents to reason through headings, sections, and paragraphs — evaluating how each part relates to the whole. The problem it solves is not "where to look", but "what does this mean in context". Every answer is a reasoning act, not a retrieval result.
108

119
## How It Works
1210

@@ -76,12 +74,12 @@ async fn main() -> vectorless::Result<()> {
7674
## Features
7775

7876
- **Hierarchical Semantic Trees** — Preserves document structure, not flat chunks
79-
- **LLM-Powered Agent Navigation** — Worker agents navigate the tree using commands (ls, cd, cat, find, grep), making every retrieval decision through LLM reasoning
77+
- **LLM-Powered Agent Navigation** — Worker agents navigate the tree using commands (ls, cd, cat, find, grep), making every decision through LLM reasoning
8078
- **Cross-Reference Resolution** — Automatically resolves "see Section 2.1", "Appendix G" references during indexing
8179
- **Synonym Expansion** — LLM-generated synonyms for indexed keywords improve recall for differently-worded queries
8280
- **Orchestrator Supervisor Loop** — Multi-document queries are coordinated by an LLM supervisor that dispatches Workers, evaluates evidence, and replans when needed
8381
- **Cross-Document Graph** — Automatic relationship discovery between documents via shared keywords
8482
- **Incremental Indexing** — Content fingerprinting skips unchanged files
8583
- **DocCard Catalog** — Lightweight document metadata index enables fast multi-document analysis without loading full documents
8684
- **Multi-Format** — Markdown and PDF support
87-
- **Zero Infrastructure**No vector DB, no embedding models, just an LLM API key
85+
- **Zero Infrastructure**Just an LLM API key, nothing else to deploy

docs/docusaurus.config.ts

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -6,7 +6,7 @@ import type * as Preset from '@docusaurus/preset-classic';
66

77
const config: Config = {
88
title: 'Vectorless',
9-
tagline: 'Reasoning-based Document Engine',
9+
tagline: 'Document Understanding Engine for AI',
1010
favicon: 'img/favicon.ico',
1111

1212
future: {

docs/src/pages/index.tsx

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -42,7 +42,7 @@ function HomepageHeader() {
4242
{/* Left: Brand + Features */}
4343
<div className={styles.heroContent}>
4444
<h1 className={styles.mainTitle}>Vectorless</h1>
45-
<p className={styles.subTitle}>Reasoning-native Document Engine</p>
45+
<p className={styles.subTitle}>Document Understanding Engine for AI</p>
4646

4747
<div className={styles.featureList}>
4848
<div className={styles.featureItem}>
@@ -119,7 +119,7 @@ export default function Home(): ReactNode {
119119
return (
120120
<Layout
121121
title={`${siteConfig.title}`}
122-
description="Reasoning-based document engine. No embeddings, no chunking. Multiple agents navigate your documents to find exactly what's relevant.">
122+
description="Document understanding engine for AI. Agents reason through your documents — navigating structure, reading passages, cross-referencing across sections.">
123123
<HomepageHeader />
124124
<main />
125125
</Layout>

0 commit comments

Comments
 (0)