This project provides a lightweight SQLite extension for vector search, enabling efficient storage and querying of vector embeddings directly within SQLite databases. It targets developers and researchers needing a portable, dependency-free solution for integrating vector search into applications, particularly those leveraging SQLite.

How It Works

The extension implements vector search using the vec0 virtual table in pure C. It supports various vector types (float, int8, binary) and allows storing non-vector data in auxiliary columns. This approach offers broad compatibility, running on diverse platforms including browsers via WASM, and avoids external dependencies, simplifying deployment.

Quick Start & Requirements

• Install: pip install sqlite-vec (Python), npm install sqlite-vec (Node.js), gem install sqlite-vec (Ruby), go get -u github.com/asg017/sqlite-vec/bindings/go (Go), cargo add sqlite-vec (Rust), datasette install datasette-sqlite-vec (Datasette), rqlited -extensions-path=sqlite-vec.tar.gz (rqlite), sqlite-utils install sqlite-utils-sqlite-vec (sqlite-utils).
• Prerequisites: SQLite.
• More Info: sqlite-vec with Python, sqlite-vec with Node.js, sqlite-vec with Ruby, sqlite-vec with Go, sqlite-vec with Rust, sqlite-vec with Datasette, sqlite-vec with rqlite, sqlite-vec with sqlite-utils.

Highlighted Details

• Pure C implementation, no external dependencies.
• Supports float, int8, and binary vector types.
• Runs anywhere SQLite runs, including Linux, macOS, Windows, and WASM.
• Integrates with popular tools like Datasette, rqlite, and sqlite-utils.

Maintenance & Community

• Developed by Alex Garcia (asg017), a Mozilla Builders project.
• Sponsored by Mozilla, Fly.io, Turso, SQLite Cloud, and Shinkai.
• GitHub Sponsors for individual support.

Licensing & Compatibility

• MIT License. Permissive for commercial use and closed-source linking.

Limitations & Caveats

The project is pre-v1, indicating potential for breaking changes. Performance is described as "fast enough," suggesting it may not be suitable for extremely high-throughput or latency-sensitive applications compared to specialized vector databases.