Skip to content

What is Milvus?

Open-source vector database for RAG, semantic and hybrid search, from a pip-installed Lite mode up to a distributed cluster holding billions of vectors.

Pricing
Free
Free plan
Yes
Made by
Zilliz
Updated

Compare Milvus alternatives

Milvus screenshots

Milvus features & details

Milvus is an open-source vector database for storing embeddings and searching them by similarity. It sits behind retrieval-augmented generation (RAG), semantic and image search, and recommendation systems, and it is one of the longest-running projects in the space: started in 2019, written in Go and C++, licensed under Apache 2.0 and hosted by the LF AI & Data Foundation, with Zilliz as the main contributor.

One API, three sizes

The same Python client (pymilvus) talks to every deployment, so code written against a laptop prototype carries over to production:

  • Milvus Lite runs inside your process after a pip install, persisting to a local file — enough for notebooks and prototypes.
  • Milvus Standalone is a single-machine server (Docker) for testing or production datasets up to millions of vectors.
  • Milvus Distributed is Kubernetes-native, separates compute from storage and scales query nodes horizontally to billions of vectors.

If you would rather not run it, Zilliz Cloud offers it fully managed (serverless or dedicated, with a free tier and BYOC options) — a separate, paid service from the same company.

What it does well

  • Index choice: HNSW, IVF, FLAT, SCANN and DiskANN, with quantised variants and memory-mapped storage, plus GPU acceleration.
  • Hybrid search: dense vectors, learned sparse embeddings (SPLADE, BGE-M3) and built-in BM25 full-text search in the same collection, with reranking across the result sets.
  • Metadata filtering over scalar fields (integers, strings, JSON) alongside vector similarity.
  • Multi-tenancy at database, collection, partition or partition-key level, and hot/cold storage to cut cost.
  • Security: user authentication, TLS and role-based access control.
  • A built-in web UI for inspecting collections, data, queries, slow requests and server configuration.

It integrates with LangChain, LlamaIndex, Haystack, DSPy and other AI frameworks, and the docs carry worked tutorials for RAG, multimodal search and graph RAG.

Who it is for

Teams that need a self-hostable vector store that can grow from a prototype to very large collections without changing client code. For a small project a single Lite file is enough; the distributed mode is real infrastructure and expects Kubernetes experience.

No reviews yet — sign in to write the first one

Apps similar to Milvus

The closest matches are Weaviate and Qdrant. See every Milvus alternative compared by price and free plan.

An open-source AI-native database combining vector search, structured filtering, and built-in model integrations.

Data & AnalyticsFreemium

An open-source vector search engine written in Rust — fast, memory-efficient, and self-hostable.

Data & AnalyticsFreemium

A fully managed vector database for semantic search and retrieval at billion-scale, with no infrastructure to run.

Data & AnalyticsFreemium

Turn a Postgres database into working memory for AI agents — relational rows, graph traversal and vector search fused into one ranked, token-budgeted context block.

Data & AnalyticsFreemium

The embedding database that starts as a Python import and grows into a distributed cloud service.

Data & AnalyticsFreemium