Milvus

Vector databases Advanced Open source Open core

A vector database built for scale, with distributed indexing, several index types and GPU acceleration, aimed at retrieval over very large collections. It is more infrastructure than library: you run it as a service, usually in containers, which brings operational weight that smaller embedded stores avoid.

Strengths

Scales horizontally, with separated compute and storage
Multiple index types (HNSW, IVF, DiskANN) plus optional GPU indexing
Supports dense and sparse vectors, and hybrid search

Weaknesses

The distributed architecture demands operational expertise at scale
Heavier than an embedded store for small local projects
A steeper learning curve than the lightweight alternatives

At a glance

Licence: Apache-2.0
Pricing: Open core
Platforms: Linux, macOS, Docker
GPU required: No

Works with

BGE-M3

Related guides

Building a local RAG system
Improving RAG retrieval quality

Glossary

Vector database
Embedding
RAG

Entry last verified 18 August 2026.