faiss

Perform fast nearest-neighbor search on dense vectors using FAISS indexes.

Updated Apr 10, 2026

Search Tags:#gpu #ann #hnsw #vector-search #faiss #similarity-search #ivf-pq

Ready to run

npx skills add https://github.com/KarlinskyS/hermesSkills --skill faiss-karlinskys

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: faiss
Source: https://github.com/KarlinskyS/hermesSkills/tree/main/mlops/vector-databases/faiss
Command: npx skills add https://github.com/KarlinskyS/hermesSkills --skill faiss-karlinskys

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

FAISS provides efficient nearest-neighbor similarity search and clustering for very large collections of dense vectors, enabling low-latency retrieval over millions to billions of embeddings.

Core Features & Use Cases

High-performance k-NN Search: Exact and approximate nearest-neighbor search using Flat, IVF, HNSW, and Product Quantization indexes.
Scalable & GPU-accelerated: Support for single and multi-GPU setups to accelerate index building and queries.
Production workflows: Trainable indexes, persistent save/load of indices, and integrations with LangChain and LlamaIndex for semantic search and recommendation systems.

Quick Start

Use FAISS to build an index from your embeddings, move it to GPU if available, and run k-nearest-neighbor queries to retrieve top matches for a text embedding.

Frequently Asked Questions about faiss

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I run fast similarity search for large-scale dense vector embeddings?▼

You can run fast similarity search by building a FAISS index from your dense vector embeddings. It supports exact and approximate nearest-neighbor retrieval to find top matches across millions of vectors.

What's the best way to build an approximate nearest-neighbor index for semantic search?▼

Building an approximate nearest-neighbor index for semantic search is best done using FAISS index types like IVF, HNSW, or Product Quantization. These indexes reduce retrieval latency for large-scale embedding datasets.

Can I use GPU acceleration for vector search and index building?▼

Yes, you can use GPU acceleration for vector search and index building. FAISS supports single and multi-GPU setups to accelerate both index training and k-nearest-neighbor query execution.

Does similarity search work with LangChain and LlamaIndex recommendation systems?▼

Yes, similarity search works with LangChain and LlamaIndex recommendation systems. FAISS integrates with these frameworks to provide low-latency retrieval for production semantic search and recommendation workflows.

How do I save and load trained indices for production batch processing?▼

You can save and load trained indices for production batch processing using FAISS. This allows you to persist trained index states and deploy them for large-scale k-NN batch processing without retraining.

When should I use Flat vs IVF vs HNSW indexes for nearest-neighbor retrieval?▼

Use Flat indexes for exact nearest-neighbor retrieval on smaller datasets, while IVF, HNSW, and PQ indexes provide approximate search optimized for billion-scale datasets where low-latency retrieval is prioritized over exact precision.