A C++ library and command-line tool for fast approximate nearest neighbor search over high-dimensional vectors, using graph and tree indexes.
NGT provides commands and a library for approximate nearest neighbor search over large collections of high-dimensional vectors, typically from tens to thousands of dimensions. It builds neighborhood graph and tree indexes so that similar objects can be found without comparing a query against every stored item.
The project offers several methods: a graph and tree-based index (NGT), a quantized graph method (QG), and a quantized blob graph method (QBG) intended for very large datasets. Supported distance functions include L1, L2, cosine, angular, Hamming, Jaccard, inner product, and hyperbolic (Poincare and Lorentz) distances. Vectors can be stored as 4-byte or 2-byte floats or as 1-byte unsigned integers, and indexes can be placed in memory-mapped shared memory.
NGT includes a CLI, a C++ API, and bindings for Python, Rust, Go, Ruby, PHP, and JavaScript. Objects can be added and removed after an index is built. Builds are supported on Linux, macOS, and Windows through MinGW-w64/MSYS2, with QG and QBG requiring BLAS and LAPACK.
Install on macOS with Homebrew
brew install ngtDownload the source archive from the Releases page
Configure the build on Linux without QG and QBG
cmake -DNGT_QBG_DISABLED=ON ..Summary generated by AI from the project's README. GitHub stats refresh daily.
A C++ library for large-scale approximate nearest neighbor vector search, with index build, search, and distributed serving toolkits.
Open-source vector similarity search extension for Postgres, storing and querying embeddings alongside your relational data.
An embedded retrieval library and database for multimodal AI, offering vector, full-text and SQL search over data stored in the Lance format.
An AI-native database for LLM applications, offering hybrid search over dense vectors, sparse vectors, tensors and full-text.
A pure-Python approximate nearest neighbor search library built on Nearest Neighbor Descent, with broad distance metric support.
CUDA-accelerated C++ header-only library and Python wrappers of reusable primitives for machine learning, linear algebra, and information retrieval on NVIDIA GPUs.