Crate vectorless 

vectorless

Hierarchical, Reasoning-Native Document Intelligence Engine.

A document indexing and retrieval library that uses tree-based navigation instead of vector embeddings for RAG applications.

Features

• Tree-Based Indexing — Documents organized as hierarchical trees
• LLM Navigation — Intelligent traversal using LLM to find relevant content
• No Vector Database — Eliminates infrastructure complexity
• Multiple Formats — Support for Markdown, PDF, HTML, and more

Quick Start

use vectorless::core::{DocumentTree, TreeNode};

// Create a document tree
let mut tree = DocumentTree::new("Root", "Root content");

// Add children
let root = tree.root();
let child = tree.add_child(root, "Section 1", "Content for section 1");

// Navigate the tree
for node_id in tree.children(root) {
    if let Some(node) = tree.get(node_id) {
        println!("Title: {}", node.title);
    }
}

Architecture

The crate is organized into the following modules:

• core — Core types: TreeNode, DocumentTree, NodeId
• llm — Unified LLM client with retry support
• concurrency — Rate limiting and concurrency control
• document — Document parsing: Markdown, PDF, HTML
• indexer — Index building: tree construction, thinning, merging
• summarizer — Summary generation
• retriever — Retrieval strategies
• ranking — Result ranking
• storage — Persistence and caching
• client — High-level API

Re-exports

pub use core::DocumentTree;

pub use core::DocumentStructure;

pub use core::NodeId;

pub use core::StructureNode;

pub use core::TreeNode;

pub use core::Error;

pub use core::Result;

pub use core::Retriever;

pub use config::Config;

pub use config::ConfigLoader;

pub use config::SummaryConfig;

pub use config::RetrievalConfig;

pub use llm::LlmClient;

pub use llm::LlmConfig;

pub use llm::LlmConfigs;

pub use llm::LlmError;

pub use llm::LlmPool;

pub use llm::RetryConfig;

pub use concurrency::ConcurrencyConfig;

pub use concurrency::ConcurrencyController;

pub use concurrency::RateLimiter;

pub use document::DocumentParser;

pub use document::DocumentFormat;

pub use document::MarkdownParser;

pub use document::RawNode;

pub use document::ParseResult;

pub use summarizer::summarize;

pub use indexer::TreeBuilder;

pub use storage::Workspace;

pub use storage::PersistedDocument;

pub use storage::DocumentMeta as StorageDocumentMeta;

pub use client::Vectorless;

pub use client::VectorlessBuilder;

pub use client::IndexedDocument;

pub use client::DocumentInfo;

pub use retriever::LlmNavigator;

pub use retriever::RetrieveOptions;

pub use retriever::RetrievalResult;

pub use retriever::ContextBuilder;

pub use registry::ParserRegistry;

pub use registry::SummarizerRegistry;

pub use registry::RetrieverRegistry;

pub use ranking::Scorer;

pub use ranking::Merger;

pub use ranking::ScoredResult;

pub use ranking::ScoringStrategy;

pub use ranking::MergeStrategy;

pub use token::estimate_tokens;

pub use token::estimate_tokens_fast;

Modules

client

High-level client API for document indexing and retrieval.

concurrency

Concurrency control for LLM API calls.

config

Configuration management for vectorless.

core

Core module containing fundamental types and traits.

document

Document parsing module.

indexer

Document indexing module.

llm

Unified LLM client module.

ranking

Result ranking and merging module.

registry

Registry module for managing pluggable components.

retriever

Document retrieval strategies.

storage

Storage module for persisting document indices.

summarizer

Document summarization module.

token

Unified token estimation module.