Repolode
Submit
nmslib

nmslib/hnswlib

Header-only C++ library with Python bindings for fast approximate nearest neighbor search using the HNSW graph algorithm.

View on GitHubWebsite
Stars
5,364
Forks
862
Open issues
311
Last commit
2 months ago

What is it?

Hnswlib is a header-only C++ implementation of Hierarchical Navigable Small World (HNSW) graphs for approximate nearest neighbor search. It indexes high-dimensional vectors so that finding the closest matches to a query vector is much faster than brute-force comparison, at the cost of returning results that are approximately, not exactly, the nearest.

The library ships Python bindings that work with NumPy arrays and support L2, inner product, and cosine distances. Indexes can be built incrementally, updated by re-inserting labels, marked as deleted, and saved to or loaded from disk. Python indexes can also be pickled. Custom distance functions can be supplied from C++.

The C++ core has no dependencies beyond C++11. Recent versions add stream-based save and load and an optional no-exceptions API for builds compiled with -fno-exceptions.

Key features

  • 01Header-only C++ implementation with no dependencies beyond C++11.
  • 02Python bindings for NumPy arrays with L2, inner product, and cosine distances.
  • 03Incremental insertion, updates of existing labels, and deletion marking.
  • 04Index save and load to disk, plus pickle support in Python.
  • 05Filtered search with a boolean function, and custom distances in C++.

Use cases

  • Building a semantic search or recommendation backend over embedding vectors.
  • Finding similar images or products by their feature vectors.
  • Adding a fast nearest-neighbor index to a C++ application without pulling in a large dependency.
  • Maintaining a vector index where items are frequently added, updated, or removed.

Things to know

  • The README describes version 0.10.0rc2, a release candidate that should not be treated as a stable release.
  • Filtered search in Python is limited by the GIL and runs slowly in multithreaded mode.

In collections

Vector databases

Topics

#vector-search#approximate-nearest-neighbor#hnsw#similarity-search#cpp#python#embeddings

FAQ

What is hnswlib?
Header-only C++ library with Python bindings for fast approximate nearest neighbor search using the HNSW graph algorithm. Hnswlib is a header-only C++ implementation of Hierarchical Navigable Small World (HNSW) graphs for approximate nearest neighbor search. It indexes high-dimensional vectors so that finding the closest matches to a query vector is much faster than brute-force comparison, at the cost of returning results that are approximately, not exactly, the nearest.
Is hnswlib free to use? What license does it have?
hnswlib is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is hnswlib still maintained?
Updated within 6 months. The last commit to hnswlib was on Sep 15, 2026.
What can I use hnswlib for?
Building a semantic search or recommendation backend over embedding vectors. Finding similar images or products by their feature vectors. Adding a fast nearest-neighbor index to a C++ application without pulling in a large dependency. Maintaining a vector index where items are frequently added, updated, or removed.
Where can I get hnswlib?
The source code and installation instructions are on GitHub: https://github.com/nmslib/hnswlib

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects