Milvus

Milvus is an open-source vector database designed to store, index, and query large collections of high-dimensional vectors produced by machine learning models. It enables similarity search over embeddings for applications such as natural language processing, computer vision, and recommender systems at scale. Milvus is developed by Zilliz and released under the Apache 2.0 license.

Milvus stores vector data alongside associated metadata in collections. It supports both float and binary vectors, as well as scalar fields for attributes. The platform provides multiple approximate nearest neighbor (ANN) search algorithms, including index types based on inverted files and graph-based methods, to enable efficient retrieval on large datasets. It also supports filtering and scalar predicates to refine results.

Milvus is designed for distributed deployment and can run on CPU or GPU hardware. It supports horizontal scalability through multi-machine clusters and can be deployed on premises or in the cloud. Containers and orchestration with Kubernetes are commonly used to manage Milvus deployments, enabling scalable and resilient operation.

Ecosystem and usage: Milvus offers client libraries and SDKs for languages such as Python, Java, Go, Node.js, and C++, facilitating integration with machine learning pipelines and data systems. It can ingest data from a variety of sources and integrates with tools for data processing, analytics, and visualization. Milvus Insight provides a web-based UI for management and monitoring, though usage of the UI is optional.

Milvus serves AI applications that require fast similarity search across large vector datasets and continues to evolve through an active open-source community and commercial offerings.