Infinity is an AI-native database for LLM applications, with its core positioning being to provide extremely fast hybrid search across dense vectors, sparse vectors, tensors (multi-vectors), and full-text retrieval. It is aimed at LLM applications such as search, recommendation, question answering, conversational AI, copilot, and content generation, as well as various RAG (retrieval-augmented generation) scenarios.
Main capabilities:
- High performance: achieves 0.1 millisecond query latency and 15K+ QPS on million-scale vector datasets; achieves 1 millisecond latency and 12K+ QPS in full-text retrieval over 33 million documents.
- Powerful search: supports hybrid search across dense vectors, sparse vectors, tensors, and full text, with filtering support; supports multiple rerankers, including RRF, weighted sum, and ColBERT.
- Rich data types: supports multiple data types such as strings, numbers, and vectors.
- Deployment and usage: single-binary architecture, dependency-free, easy to deploy; can be used embedded as a Python module and provides an intuitive Python API.
Usage: Supports deploying the Infinity database via Docker, with the client and server running as independent processes. Installation requirements: CPU is x86_64 with AVX2 support; operating system is Linux (glibc 2.17+), Windows 10+ (WSL/WSL2), or macOS; Python 3.11+. On Windows 10+, WSL or WSL2 must be enabled to use Docker deployment. When installing the server, first create the data directory and grant permissions, then pull the image and start the container. The commands are as follows: sudo mkdir -p /var/infinity && sudo chown -R $USER /var/infinity docker pull infiniflow/infinity:nightly docker run -d --name infinity -v /var/infinity/:/var/infinity --ulimit nofile=500000:500000 --network=host infiniflow/infinity:nightly If using Windows Docker Desktop 4.29+, you need to select "Enable host networking" in "Settings > Features in development". In addition,The material also mentions installing the Infinity client and running vector search, as well as deploying via binary (with server and client separated) and building from source. The specific steps are not provided in detail for now.
Applicable scenarios: Search, recommendation systems, question answering, conversational AI, copilot, content generation, and various RAG applications.
Limitations: The material does not provide information such as charging, pricing, platform support scope, and performance guarantees; some deployment and build steps (binary deployment, source build, and the specific commands for client installation) are not provided for now.
