Milvus
Milvus is an open-source, cloud-native vector database for approximate nearest-neighbour search over unstructured data (text, images, multimodal). The README says it is written in Go and C++ with CPU and GPU acceleration. Zilliz, the company behind it, is named in the milvus.io footer.
Repository snapshot (GitHub API, 2026-11-19)
| Item | Detail |
|---|---|
| Repo | milvus-io/milvus, Apache-2.0 |
| Created | 2019-09-16 |
| Activity | last push 2026-11-19; latest release v3.0.6 (2026-11-11) |
| Size | 46,696 stars, 4,337 forks |
Capabilities (README and milvus.io)
- Fast search at scale; the site claims tens of billions of vectors.
- Fully distributed, Kubernetes-native horizontal scaling and real-time streaming updates.
- Metadata filtering, hybrid search and multi-vector support.
- A wide index family (milvus.io index docs): FLAT, IVF variants (including IVF_RABITQ), HNSW variants, DISKANN, SCANN, GPU indexes (CAGRA, IVF), binary indexes (BIN_FLAT, BIN_IVF_FLAT, MINHASH_LSH) and one sparse index (SPARSE_INVERTED_INDEX). See ann-index-algorithms.
Deployment and billing
- Milvus Lite: embedded,
pip install, for notebooks and prototypes. - Standalone (one machine) and Distributed (cluster).
- Zilliz Cloud: managed, with Serverless, Dedicated and bring-your-own-cloud options, a free tier and on-demand search billed per query. Rates are on Zilliz’s pricing page and are not recorded here.
Multi-tenancy and quantisation were not stated on the pages read.
Where it fits
Milvus is the choice for very large, distributed deployments and the widest index selection; compare qdrant, weaviate and the library-level faiss. See vector-databases-compared.
Sources
- milvus.io, its index docs, zilliz.com, the Milvus README and GitHub API (2026-11-19)