Milvus
Vector database software
Database software
- Features
- Ease of use
- Ease of management
- Quality of support
- Affordability
- Market presence
Take the quiz to check if Milvus and its alternatives fit your requirements.
Completely free
Free TrialFree version
Small
Medium
Large
- Agriculture, fishing, and forestry
- Retail and wholesale
- Information technology and software
What is Milvus
Milvus is an open-source vector database designed to store, index, and search high-dimensional embeddings for similarity search and retrieval-augmented generation (RAG) workloads. It targets engineering teams building semantic search, recommendation, and AI applications that require approximate nearest neighbor (ANN) retrieval at scale. Milvus separates compute and storage components and supports multiple index types and distance metrics to tune performance and recall. It is commonly deployed in containers and Kubernetes environments and is also available through a managed service from the same vendor.
Purpose-built vector indexing
Milvus focuses on vector similarity search with configurable ANN index types and distance metrics (for example, L2, inner product, and cosine). This specialization helps teams optimize for latency, recall, and memory footprint based on workload characteristics. It also supports scalar fields and filtering alongside vector search, which is important for production retrieval use cases.
Scalable distributed architecture
Milvus supports distributed deployment with components that can scale horizontally, which fits large collections and high query throughput. It is commonly run on Kubernetes and integrates with object storage for persistence in typical production setups. This architecture aligns with teams that need to scale vector search independently from application services.
Open-source ecosystem and APIs
Milvus provides open-source code, client SDKs, and integrations that make it easier to embed vector search into applications. It supports common ingestion and query patterns used in AI pipelines, including batch inserts and similarity queries with filtering. The open-source model can reduce vendor lock-in compared with proprietary-only offerings.
Index tuning and trade-offs
Achieving good results often requires choosing and tuning index types and parameters based on dataset size, dimensionality, and latency/recall targets. Misconfiguration can lead to higher memory usage, slower queries, or reduced recall. Teams may need benchmarking and ongoing tuning as data and query patterns evolve.
Operational complexity at scale
Running Milvus in distributed mode typically requires managing multiple services and dependencies (for example, coordination and storage backends), which increases operational overhead. Teams often need Kubernetes expertise and monitoring to maintain performance and reliability. This can be heavier than using a general-purpose database with built-in vector features for smaller deployments.
Not a general-purpose database
Milvus is optimized for vector retrieval rather than full relational or document database functionality. It does not aim to replace transactional databases for complex joins, multi-row transactions, or broad SQL workloads. Many production systems still pair it with another database for system-of-record data.
Plan & Pricing
| Plan | Price | Key features & notes |
|---|---|---|
| Self-hosted (Milvus open-source) | $0 — Free (Apache 2.0) | Milvus is a 100% free open-source vector database distributed under the Apache 2.0 license; self-host and operate at your own infrastructure (official docs state Milvus is free). |
Zilliz Cloud (managed Milvus) — usage-based (Serverless & Dedicated)
Pricing model: Pay-as-you-go (Serverless: vCU-based per-operation billing); Dedicated: CU-based resource billing (per CU-hour) and storage charged separately.
Free tier / free cluster: Available — Zilliz Cloud free cluster: 5 GB storage, 2.5M vCUs/month, up to 5 collections.
Free trial: Available — Zilliz Cloud free trial / credit-based trials and 30-day free trials for clusters (no credit card required to start free cluster).
Serverless (example unit pricing & examples):
- vCU unit price: $4 per million vCUs (applies to both read and write vCUs in Serverless).
- Write examples: writing 1M 768-dim vectors ≈ 0.75 million vCUs → $3 (doc example).
- Read examples: 1M 768-dim vectors read ≈ 15 million vCUs → $60 (doc example).
- Notes: Read operations cost a minimum of 6 vCUs per operation; import/bulk insert may not incur write costs (see docs for exact rules).
Dedicated clusters:
- Billing by Compute Units (CUs) × runtime (per-CU-hour) + storage + optional add-ons.
- Storage: official docs/release notes indicate storage pricing examples (e.g., standardized storage pricing $0.04 per GB per month announced in product updates). Exact CU/hour rates vary by region/provider and are shown as examples in docs; contact sales or cloud marketplace listing for region-specific CU/hour rates.
Example cost references (official docs):
- Serverless vCU pricing and read/write examples (Zilliz Cloud docs).
- Free cluster details and free trial information (Zilliz Cloud docs).
- Storage pricing example & release note standardization ($0.04/GB/month) (Zilliz Cloud release notes).
Discounts & billing options: Advance pay, marketplace subscriptions, and enterprise/commitment discounts available; contact sales for custom/enterprise pricing.
Seller details
Zilliz, Inc.
Redwood City, CA, USA
2017
Private
https://zilliz.com/
https://x.com/zilliz_universe
https://www.linkedin.com/company/zilliz/
