Milvus is a high-performance vector database written in Go and C++ that enables scalable vector similarity search for AI applications. The system is designed to efficiently organize and search vast amounts of unstructured data such as text, images, and multi-modal information by storing and indexing vector embeddings alongside scalar data types like integers, strings, and JSON objects.
The architecture separates compute and storage layers, allowing horizontal scaling across distributed Kubernetes clusters. This design enables Milvus to handle tens of thousands of concurrent search queries on billions of vectors while maintaining real-time data freshness through streaming updates. The system offers both a fully distributed deployment mode for production environments and a Standalone mode for single-machine deployments, with Milvus Lite providing a lightweight Python-based option for quick prototyping via pip installation.
Milvus implements hardware acceleration for both CPU and GPU processing to achieve high vector search performance. The system supports multiple vector index types optimized for different scenarios, including HNSW, IVF, FLAT brute-force search, SCANN, and DiskANN, with quantization-based variations and memory-mapped file support. Beyond dense vector search, Milvus natively supports sparse vectors for full-text search using BM25 and learned sparse embeddings like SPLADE and BGE-M3, enabling hybrid search capabilities that combine semantic and full-text search within the same collection.
The repository demonstrates significant community engagement with 6521 tracked issues and pull requests. The median response latency for issues and PRs is 0.0 hours, while the mean latency is 2697.1 hours, indicating rapid initial triage followed by longer resolution timelines for complex issues. The most active issue labels are kind/bug with 4465 items, triage/accepted with 3417 items, and stale with 2375 items. Top contributors include yanliang567 with 15592 events, xiaofan-luan with 4274 events, and zhuwenxing with 3362 events, reflecting sustained development activity.
Milvus provides flexible multi-tenancy strategies with isolation at the database, collection, partition, or partition key level, allowing a single cluster to serve hundreds to millions of tenants. The system implements hot and cold storage mechanisms to optimize costs by keeping frequently accessed data in memory or on SSDs while storing less-accessed data on slower, more cost-effective storage. Enterprise security features include mandatory user authentication, TLS encryption for network communications, and Role-Based Access Control for fine-grained permission management.
The project is distributed under the Apache 2.0 license and operates under the LF AI & Data Foundation with Zilliz as its major contributor. A fully managed cloud service called Zilliz Cloud offers Serverless, Dedicated, and Bring-Your-Own-Cloud deployment options. The repository overlaps with contributors from microsoft/vscode, microsoft/typescript, and rust-lang/rust, indicating cross-project collaboration. Milvus powers production applications including text and image search, Retrieval-Augmented Generation systems, and recommendation engines for both startups and enterprises.
