FAISS

FAISS

FAISS (Facebook AI Similarity Search) is the industry-standard C++ library (with Python bindings) for nearest-neighbor search over high-dimensional embedding vectors. It powers the vector retrieval layer in countless RAG systems, recommendation engines, and semantic search applications.

Key Features

• Multiple index types — Flat (exact), IVF (inverted file), HNSW, PQ (product quantization), and hybrids
• GPU support — single and multi-GPU acceleration via CUDA for billion-scale search
• Compression — Product Quantization and Scalar Quantization reduce memory by 4–64×
• Billion-scale — benchmarked on datasets with 1B+ vectors
• Python & C++ APIs — use from Python for prototyping, C++ for production integration
• LangChain / LlamaIndex integration — FAISS vector store is a first-class integration in both frameworks

Quick Start

pip install faiss-cpu  # or faiss-gpu for CUDA

import faiss
import numpy as np

d = 128  # vector dimension
nb = 10000  # database size

xb = np.random.random((nb, d)).astype("float32")
index = faiss.IndexFlatL2(d)
index.add(xb)

xq = np.random.random((5, d)).astype("float32")
D, I = index.search(xq, k=5)  # 5 nearest neighbors
print(I)  # indices of nearest neighbors

Install via ai-supply

npx ai-supply add faiss-vector-search

Curated mirror of the open-source FAISS project (MIT). Install upstream from the repository.