llmdb

A lightweight, embeddable vector database built with Rust and SQLite. llmdb provides both a library and REST API for storing documents with vector embeddings and performing semantic similarity searches, making it ideal for local development, testing, embedded applications, or distributed environments.

Overview

llmdb combines SQLite's reliability with HNSW vector indexing to create a minimal yet powerful document store. The project features:

• Vector Search: Store and query documents using high-dimensional embeddings (configurable dimensions)
• HNSW Indexing: Fast approximate nearest neighbor search with configurable parameters
• Dual Usage Modes:
  • Embedded: Use as a Rust library for direct integration
  • REST API: Run as a server and connect via HTTP client
• Flexible Embedding: Support for external embedding services with caching
• Background Workers: Asynchronous embedding generation and processing
• Graph Operations: Document relationship tracking and graph queries
• Feature Flags: Configurable features for different deployment scenarios

Architecture

llmdb is structured as a Rust workspace with three crates:

• llmdb-core - Core library with storage, indexing, search, and embedding logic
• llmdb-client - HTTP client for connecting to llmdb REST API
• llmdb - Main binary and server runtime (can also be used as library)

Quick Start

As Library (Embedded)

use llmdb_core::*;

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    let mut store = store::DocumentStore::new("./data".to_string()).await?;
    
    // Add document
    let doc_id = store.add_document(/* ... */).await?;
    
    // Search
    let results = store.search(/* ... */).await?;
    
    Ok(())
}

As Server + Client

# Start server
cargo run --release

# In another terminal/application

use llmdb_client::Client;

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    let client = Client::new("http://localhost:8080");
    
    // Add document
    let doc_id = client.add_document(
        "Hello world".to_string(),
        None
    ).await?;
    
    // Search
    let results = client.search("greeting".to_string(), 10).await?;
    
    Ok(())
}

Building

# Build entire workspace
cargo build --release

# Run server
cargo run --release

For detailed build instructions, cross-compilation, CI/CD, and development commands, see BUILD.md.

---

Disclaimer

This project was generated with the assistance of Claude Sonnet 4.5 via GitHub Copilot in VS Code.

I include these tools for transparency and provenance tracking in case this is ever useful to others in the future.

Tool	Version	Date
VsCode	1.107.1	2026.01.07
github.copilot-chat	0.35.3	2026.01.07
Claude Sonnet	4.5	2026.01.07