infiniflow/infinity

What it solves

Infinity is an AI-native database designed to handle the high-performance retrieval needs of LLM applications. It addresses the challenge of searching across different data types—such as dense and sparse vectors, tensors, and full-text—simultaneously and quickly, which is essential for advanced Retrieval-augmented Generation (RAG) systems.

How it works

Infinity operates as a single-binary architecture that can be deployed via Docker or as a standalone binary. It provides an intuitive Python API and HTTP API for developers to manage data and perform hybrid searches. It supports various reranking methods, including ColBERT, RRF, and weighted sum, to refine search results.

Who it’s for

It is built for AI developers creating search engines, recommenders, question-answering systems, conversational AI, and other RAG-based applications that require low-latency, high-throughput retrieval of rich data types.

Highlights

• Hybrid Search: Combines dense embeddings, sparse embeddings, tensors, and full-text search with filtering.
• High Performance: Achieves sub-millisecond query latency on million-scale vector datasets and high QPS for full-text search.
• Ease of Deployment: Single-binary architecture with no dependencies, available as a Docker image or a Python module.
• Rich Data Support: Handles strings, numerics, vectors, and structured data.