USearch

USearch is a compact, single-file vector search engine: it delivers up to 10x faster HNSW performance than FAISS for billion-scale similarity and clustering.

USearch is a high-performance vector search engine that implements the HNSW algorithm, engineered for superior speed and portability. Benchmarks show it can be up to 10x faster than Meta's FAISS in indexing time for datasets like 100 Million 96D vectors. The core is a simple, single C++11 header library, but it offers native bindings for over 10 languages, including Python, Java, and Rust. It prioritizes efficiency: a 40-bit integer ID type allows the index to scale past 4 billion vectors, and disk-based index viewing eliminates the RAM bottleneck for massive datasets. This engine is already trusted by major databases like ClickHouse and DuckDB for its speed and low-dependency architecture.

https://unum.cloud/usearch