Milvus

What is Milvus?

Milvus is an open-source vector database originally developed by Zilliz and now a Linux Foundation AI & Data project. It is built specifically for storing, indexing, and querying dense vector embeddings at scale, which makes it a common backend for retrieval-augmented generation (RAG), recommendation systems, and image/audio similarity search.

Unlike general-purpose databases retrofitted with a vector extension, Milvus was architected around approximate nearest neighbor (ANN) search from the start, separating storage, indexing, and query execution into independently scalable components.

Key Features

• Multiple index algorithms: Supports HNSW, IVF_FLAT, IVF_PQ, and DiskANN, letting teams trade off recall, latency, and memory footprint per collection rather than being locked into one algorithm.
• Distributed architecture: Milvus’s cluster mode separates proxy, query, data, and index nodes, allowing storage and compute to scale independently — useful for collections in the billions of vectors.
• Hybrid filtering: Combines vector similarity search with structured scalar filters (e.g., “similar products under $50”) in a single query.
• Deployment flexibility: Milvus Lite runs embedded in a Python process for local development; Milvus Standalone runs in Docker for small production deployments; Milvus Distributed (via Kubernetes) or the managed Zilliz Cloud handle large-scale workloads.

Who is it For?

Milvus is aimed at engineering teams building large-scale RAG pipelines, semantic search, or recommendation engines who need an index that scales past what a single-node vector store can handle. It’s a common choice when vector counts run into the hundreds of millions or billions.

Pricing & Plans

The core Milvus database is open source under the Apache 2.0 license and free to self-host. Zilliz Cloud, the managed offering, is priced by compute and storage usage with a free tier for small projects.

Strengths & Limitations

Strengths: Proven at very large scale, flexible index selection, strong Kubernetes-native operational tooling, active open-source community.

Limitations: Operating a distributed Milvus cluster yourself carries real infrastructure overhead compared to simpler managed vector databases like Pinecone; for small collections (under a few million vectors), the operational complexity may outweigh the scaling benefits.

Milvus’s Governance Under the Linux Foundation

Milvus’s status as a Linux Foundation AI & Data project, rather than being solely controlled by its originating company Zilliz, provides a governance structure some enterprises view favorably when evaluating open-source infrastructure dependencies — vendor-neutral governance reduces the risk that a single company’s business decisions could unilaterally change the project’s licensing or direction in ways that would disrupt existing production deployments built on it.