Granné is a Rust library for approximate nearest neighbor (ANN) search, designed for efficient indexing of billions of vectors by minimizing memory usage. It targets developers and researchers working with large-scale similarity search, offering a memory-mapped, extensible HNSW graph implementation.

How It Works

Granné implements Hierarchical Navigable Small World (HNSW) graphs, a popular algorithm for ANN search. Its key advantage lies in its focus on memory efficiency, enabling the indexing of massive datasets. The library supports memory-mapped indexes for reduced RAM footprint and offers extensibility, allowing elements to be added to an existing index without rebuilding.

Quick Start & Requirements

• Rust: Install via rustup.rs. Build with cargo build --release.
• Python Bindings: Available via pip install granne.
• Optional BLAS: For performance, enable with cargo build --release --features "blas". Requires libblas-dev and libopenblas-dev on Debian/Ubuntu.

Highlighted Details

• Graph-based Approximate Nearest Neighbor Search (HNSW).
• Focus on reducing memory usage for billions of vectors.
• Memory-mapped indexes.
• Extensible indexes (add elements to existing index).
• Python bindings available.

Maintenance & Community

The project is associated with Cliqz Search. Further community or maintenance details are not provided in the README.

Licensing & Compatibility

The README does not explicitly state a license.

Limitations & Caveats

The README mentions potential issues with BLAS integration on macOS. The project's status (e.g., alpha, beta, stable) and comprehensive documentation are not detailed.