ParlayANN

Category: SDKs & Libraries
Website: https://github.com/cmuparlay/ParlayANN
License: See LICENSE file in repository

Description

ParlayANN is a C++ library of scalable, deterministic, parallel graph-based approximate nearest neighbor (ANN) search algorithms for high-dimensional vector similarity search. It is designed for use as a core search component in large-scale vector databases and retrieval systems, and builds on parallel primitives from ParlayLib.

Features

• Multiple ANN algorithms implemented
  • DiskANN
  • HNSW
  • HCNNG
  • pyNNDescent
• Graph-based indices suitable for high-dimensional ANN search
• Parallel execution using ParlayLib primitives for scalability
• Range search support via algorithms from the paper Range Retrieval with Graph-Based Indices
• Tools for algorithm design: utilities and data tools to help design and experiment with ANN algorithms
• C++ implementation for performance and low-level control
• Python directory present in the repo, indicating Python-related bindings or tooling
• Build system support
  • CMake configuration (CMakeLists.txt)
  • Bazel configuration (WORKSPACE, .bazelrc)

Technology Stack

• Language: C++ (with associated Python directory)
• Parallel library: ParlayLib
• Focus: Graph-based approximate nearest neighbor and range search in high dimensions

Use Cases

• Core similarity search component in vector databases
• Large-scale vector retrieval and recommendation systems
• Research and experimentation on graph-based ANN algorithms

Pricing

ParlayANN is an open-source library. No pricing plans are specified; usage terms are governed by the license provided in the repository (LICENSE).