GitHub - 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.

The AI-native database built for LLM applications, providing incredibly fast hybrid search of dense embedding, sparse embedding, tensor and full-text

Document | Benchmark | Twitter | Discord

Infinity is a cutting-edge AI-native database that provides a wide range of search capabilities for rich data types such as dense vector, sparse vector, tensor, full-text, and structured data. It provides robust support for various LLM applications, including search, recommenders, question-answering, conversational AI, copilot, content generation, and many more RAG (Retrieval-augmented Generation) applications.

⚡️ Performance

🌟 Key Features

Infinity comes with high performance, flexibility, ease-of-use, and many features designed to address the challenges facing the next-generation AI applications:

🚀 Incredibly fast

Achieves 0.1 milliseconds query latency and 15K+ QPS on million-scale vector datasets.
Achieves 1 millisecond latency and 12K+ QPS in full-text search on 33M documents.

See the Benchmark report for more information.

🔮 Powerful search

Supports a hybrid search of dense embedding, sparse embedding, tensor, and full text, in addition to filtering.
Supports several types of rerankers including RRF, weighted sum and ColBERT.

🍔 Rich data types

Supports a wide range of data types including strings, numerics, vectors, and more.

🎁 Ease-of-use

Intuitive Python API. See the Python API
A single-binary architecture with no dependencies, making deployment a breeze.
Embedded in Python as a module and friendly to AI developers.

🎮 Get Started

This section provides guidance on deploying the Infinity database using Docker, with the client and server as separate processes.

Prerequisites

CPU: x86_64 with AVX2 support.
OS:
- Linux with glibc 2.17+.
- Windows 10+ with WSL/WSL2.
- MacOS
Python: Python 3.11+.

Install Infinity server

Linux x86_64 & MacOS x86_64

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

Windows

If you are on Windows 10+, you must enable WSL or WSL2 to deploy Infinity using Docker. Suppose you've installed Ubuntu in WSL2:

Follow this to enable systemd inside WSL2.
Install docker-ce according to the instructions here.
If you have installed Docker Desktop version 4.29+ for Windows: Settings > Features in development, then select Enable host networking.

Pull the Docker image and start Infinity:

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

Install Infinity client

pip install infinity-sdk==0.7.0.dev4

Run a vector search

import infinity

infinity_obj = infinity.connect(infinity.NetworkAddress("<SERVER_IP_ADDRESS>", 23817)) 
db_object = infinity_object.get_database("default_db")
table_object = db_object.create_table("my_table", {"num": {"type": "integer"}, "body": {"type": "varchar"}, "vec": {"type": "vector, 4, float"}})
table_object.insert([{"num": 1, "body": "unnecessary and harmful", "vec": [1.0, 1.2, 0.8, 0.9]}])
table_object.insert([{"num": 2, "body": "Office for Harmful Blooms", "vec": [4.0, 4.2, 4.3, 4.5]}])
res = table_object.output(["*"])
                  .match_dense("vec", [3.0, 2.8, 2.7, 3.1], "float", "ip", 2)
                  .to_pl()
print(res)

🔧 Deploy Infinity using binary

If you wish to deploy Infinity using binary with the server and client as separate processes, see the Deploy infinity using binary guide.