Milvus
Start Milvus with a local-file retrieval prototype, then choose production deployment and operations separately.
- Starting price
- Apache-2.0; no licence fee
- Free access
- Open-source software; no included hosted capacity
- Billing unit
- Infrastructure and external model usage
- Platform
- Local Lite, Standalone or distributed service
- Verified
- 30 Oct 2026
Overview
What you provide
- Supply vectors or records with identifiers, define the collection dimensions and submit similarity queries with optional filters.
What you get
- Milvus returns ranked matching records; it does not independently verify or write a final natural-language answer.
Choosing this tool
Milvus is suitable for developers who need to control a vector database rather than buy an assistant. Lite makes a local retrieval experiment straightforward, while Standalone and distributed services bring different operational work. Weaviate Cloud is a useful alternative when managed capacity matters more than owning deployment; Qdrant offers another self-hosted retrieval option. Compare the workload’s filters, updates and recall under realistic data before making scale claims.
A decision worked through
A local Milvus Lite experiment has USD 0 in Milvus licence fees. If you already spend USD 12 on storage or compute and USD 3 on embedding requests, the project subtotal is USD 15, not a Milvus subscription price; those external amounts are illustrative assumptions.
Milvus alternatives for open-source vector database deployment
Compare Milvus with Weaviate, Qdrant for this workflow, including the billing unit and the work left to your team.
| Product | Price / unit | Free allowance | Input → output | Key limit |
|---|---|---|---|---|
| MilvusRead profileOfficial source ↗ | Apache-2.0; no licence fee | Self-hosted software without cloud credits | Vectors and filters return ranked neighbours. | Deployment operations are separate from the Lite demo. |
| WeaviateRead profileOfficial source ↗ | Flex from USD 45/month | 100,000 objects on managed Free | Objects and queries become retrieval matches. | AI-service usage is separately metered. |
| QdrantRead profileOfficial source ↗ | Apache-2.0 software; infrastructure extra | No software licence charge for self-hosting | Vectors and filters yield matching records. | You operate the self-hosted database and embedding pipeline. |
Setup & workflow
- Install pymilvus[milvus-lite] and create a client pointing to a local database file.
- Create a collection, insert sample vectors and check known-neighbour search results.
- Evaluate Standalone or distributed deployment with backup and access requirements before production migration.
Requirements & installation
- Install the documented Python SDK and Lite extra for a local-file test; reserve persistent storage and a supported environment for the chosen release.
Limits & review
- A local Lite example does not establish production high availability or distributed capacity. No universal hardware minimum is assigned here; size the selected deployment using its official requirements and your dataset.
Your part
- Check retrieval relevance, permissions and deletion behaviour before using matches in an AI application.
Sources & verification6
This profile is based on official sources, not a hands-on product test.
Vendor descriptions and demos document advertised features. Editorial guidance is based on these sources.
- Official Milvus pricingmilvus.io
- Official Milvus productmilvus.io
- Official Milvus installation and licencegithub.com
- Milvus official referencemilvus.io
- Weaviate official referenceweaviate.io
- Qdrant official referencegithub.com
