infiniflow/infinity

The AI-native database built for LLM applications, providing incredibly fast hybrid search of dense vector, sparse vector, tensor (multi-vector), and full-text.

What it solves

Infinity is an AI-native database designed to handle the complex data retrieval needs of LLM applications. It solves the problem of needing multiple separate systems for different types of search—such as vector search for semantic meaning and full-text search for keyword matching—by combining them into a single, high-performance engine.

How it works

Infinity provides a unified system for hybrid search, allowing users to combine dense embeddings, sparse embeddings, tensors, and full-text search with filtering. It utilizes a single-binary architecture with no external dependencies, which simplifies deployment. For AI developers, it can be embedded directly into Python as a module or run as a separate server via Docker.

Who it’s for

This project is built for AI developers creating RAG (Retrieval-augmented Generation) applications, such as conversational AI, copilots, recommenders, and question-answering systems.

Highlights

  • Hybrid Search Capabilities: Supports dense and sparse embeddings, tensors, and full-text search in one place.
  • High Performance: Achieves sub-millisecond query latency on million-scale vector datasets and high QPS for full-text search.
  • Advanced Reranking: Includes support for various rerankers, including ColBERT, RRF, and weighted sum.
  • Easy Deployment: Offered as a single-binary architecture or a Docker image for rapid setup.