Milvus is an open-source vector database built for GenAI applications. Developers install it with pip, run similarity searches quickly and scale to tens of billions of vectors with minimal performance loss. It comes in several forms: Milvus Lite, a library for notebooks and laptops; Milvus Standalone for single-machine production or testing; Milvus Distributed for horizontal scaling to billions of vectors; and Zilliz Cloud, a fully managed service offered in serverless and dedicated clusters. Features include metadata filtering, hybrid search and multi-vector search. The open-source software is free to run.
Machine learning engineers, data scientists, backend developers and solution architects building retrieval applications are the core audience. Beginners can prototype with Milvus Lite in a notebook, small teams can run Standalone for datasets up to millions of vectors, and enterprises needing reliability can adopt the distributed version or choose managed hosting with SaaS and bring-your-own-cloud options. The project has an active community and more than 46 thousand GitHub stars.
Common use cases include retrieval-augmented generation, image search, multimodal search, hybrid search that mixes keywords and vectors, Graph RAG, recommendation systems and similarity search across large unstructured collections. A typical workflow creates a collection with the Python client, inserts embeddings, and queries for the nearest neighbors, as the Quickstart shows. Teams can start small on a laptop and later move the same code to production clusters, while those preferring not to operate infrastructure can use the managed Zilliz Cloud service instead. Data engineers can also pair it with embedding models of their choice, store metadata beside each vector for filtered queries, and keep experiments reproducible by saving collections between runs.