Milvus

Milvus is an open-source vector database for developers building RAG applications, image search and recommendation systems. It stores embeddings alongside metadata so applications can retrieve related text, images or multimodal data. You can run it on your own hardware, from a laptop prototype to a distributed production cluster.

Milvus Lite runs as a Python library in notebooks and on laptops, with data persisted in a local file. Standalone provides a complete database on one machine. Distributed deployments use Kubernetes and separate compute from storage, so teams can expand read and write capacity independently as workloads grow.

Search combines semantic matching with native BM25 full-text retrieval. Milvus also supports sparse embeddings such as SPLADE and BGE-M3, metadata filters and multi-vector search. These capabilities let developers combine keyword relevance with embedding similarity and restrict results using fields stored alongside the vectors.

For larger workloads, Milvus supports CPU and GPU acceleration, including NVIDIA CAGRA indexing. Its index choices include HNSW and DiskANN. Streaming updates keep searchable data fresh, while replicas support fault tolerance. Teams sharing a cluster can isolate tenants, and hot/cold storage places frequently accessed data in memory or on SSDs.

Milvus uses the Apache 2.0 license. Zilliz Cloud offers a separate managed service, with hosted deployments and an option to run in your own cloud account.