Milvus is a cloud-native vector database designed to scale to billions of vectors across distributed infrastructure. Built by the Zilliz team (who also maintain the managed Zilliz Cloud), Milvus is battle-tested at companies like Alibaba, PayPal, and Roblox for massive-scale similarity search.
Milvus uses a decoupled storage-compute architecture where query nodes, data nodes, and index nodes scale independently. It supports multiple index types (HNSW, IVF-FLAT, IVF-SQ8, DiskANN) so you can tune for speed vs memory vs disk tradeoffs. Deployment is Kubernetes-native — helm charts, horizontal pod autoscaling, and persistent volume claims are first-class. For smaller workloads, Milvus Lite runs in-process like Chroma.
Open source: free. Zilliz Cloud (managed Milvus): Starter $0, Standard $99/month, Enterprise custom. Self-hosted costs depend on your Kubernetes cluster.
Alibaba, PayPal, Ebay, Roblox, Nvidia — essentially any company doing billion-scale vector search. Strong in the e-commerce and recommendation space.