Milvus

// Overview

Milvus is an open-source vector database for storing and searching the embeddings that represent unstructured data like text, images, and audio. It keeps vectors alongside scalar fields such as integers, strings, and JSON, so you can combine similarity search with metadata filtering or hybrid search.

It is built for teams adding semantic search, recommendations, or retrieval-augmented generation to their applications. You can start small with Milvus Lite (a local file-based database you install with pip), run a single-machine Standalone deployment, or scale out to a distributed, Kubernetes-native cluster that handles billions of vectors.

Within the vector-database category, Milvus separates compute from storage so query and data nodes scale independently. It is written in Go and C++ with CPU/GPU acceleration, and is also offered as a managed service on Zilliz Cloud for those who want zero setup.

// What it does

• Stores vectors together with scalar types (integers, strings, JSON) for metadata filtering and hybrid search
• Distributed architecture that separates compute and storage, scaling query and data nodes independently
• Multiple deployment modes: Milvus Lite for local Python use, Standalone for a single machine, and Kubernetes-native clusters
• CPU/GPU hardware acceleration, with the core written in Go and C++
• Real-time streaming updates to keep indexed data fresh
• Python SDK (pymilvus) with a single MilvusClient class for collections, inserts, and search

// Getting started

Install the Python SDK, then create a local Milvus Lite database, add a collection, insert data, and run a vector search.

pip install -U pymilvus

from pymilvus import MilvusClient

client = MilvusClient("milvus_demo.db")

client.create_collection(
    collection_name="demo_collection",
    dimension=768,  # The vectors we will use in this demo have 768 dimensions
)

res = client.insert(collection_name="demo_collection", data=data)

res = client.search(
    collection_name="demo_collection",  # target collection
    data=query_vectors,  # a list of one or more query vectors, supports batch
    limit=2,  # how many results to return (topK)
    output_fields=["vector", "text", "subject"],  # what fields to return
)

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Power retrieval-augmented generation (RAG) by searching document embeddings for the most relevant context to feed an LLM
• Build semantic search over text, images, or other unstructured data using nearest-neighbor lookups
• Serve recommendations by matching user or item embeddings at low latency
• Run similarity search at very large scale (billions of vectors) on a distributed Kubernetes cluster

// How Milvus compares

Milvus alongside other open-source vector databases tools AI/TLDR tracks, ranked by GitHub stars.