Documentation

quantvec is a data-oblivious, zero-training vector quantization and nearest-neighbor search library for TypeScript — a clean-room implementation of Google Research’s TurboQuant  with the RaBitQ  unbiased-estimator correction. It runs anywhere JavaScript runs: Node, browsers, Bun, Cloudflare Workers, and React Native.

These pages are rendered from the repo’s canonical docs in docs/.

• Getting Started — install and run your first index. Start here.
• Usage Guide — end-to-end walkthrough: building, searching, filtering, and persisting.
• Architecture — the encode pipeline and flat-scan search path, end to end.
• Serialization Format — the versioned QVEC binary layout, field by field.
• API Reference — TurboQuantIndex, IdMapIndex, options, and result types.
• Benchmarks — recall@k, throughput, and compression vs float32.
• Roadmap — what’s shipped and what’s next (WASM-SIMD kernel, IVF, filter DSL).

Everything here lives in the GitHub repository . quantvec is an independent clean-room implementation and is not affiliated with or endorsed by Google.