milvus

Why we included this project

Milvus is an open-source distributed vector database that has been running in production for years, and it shows. Written in Go and C++, it stores embeddings alongside scalar metadata, so you can filter by attributes while still getting fast approximate nearest-neighbor search. It supports several index types, including HNSW, IVF, SCANN, and DiskANN, with optional GPU acceleration, and you can run it as a single instance or scale it out on Kubernetes. The Python SDK is easy to pick up, and the system handles real-time streaming inserts and hybrid search over dense and sparse vectors, which makes it a natural fit for semantic search, image similarity, and RAG pipelines. If you need a dedicated storage layer for embeddings instead of bolting vector search onto an existing database, this is a strong candidate.