Milvus

What it does

Milvus takes in vectors, which are numerical representations of text, images, or other unstructured data, along with scalar fields such as integers, strings, and JSON. It stores these alongside metadata and returns nearest-neighbor search results, including filtered, hybrid, and full-text queries, for applications that need to find similar items quickly.

Developers typically install the pymilvus Python SDK, create a collection with a vector dimension matching their embedding model, insert embeddings, and run search calls. Milvus Lite can run a local database from a single file for prototyping. The project emphasizes a distributed architecture that separates storage and compute, multiple index types such as HNSW and DiskANN, and hybrid dense-plus-sparse search for retrieval-augmented generation pipelines.

Milvus suits engineers building AI search, recommendation, or RAG systems who can manage their own deployment or prefer a managed option. Self-hosting requires infrastructure work. Zilliz Cloud offers managed Serverless, Dedicated, and BYOC deployments, with pricing that depends on usage, cluster type, and region.

What it can do

• Vector similarity search with top-K and range ANN queries
• Metadata filtering and grouping search
• Hybrid search combining dense, sparse, and full-text retrieval
• Multiple index types including HNSW, IVF, FLAT, SCANN, and DiskANN with GPU acceleration
• Multi-tenancy at database, collection, partition, or partition-key level
• Role-based access control, TLS encryption, and user authentication

Works with

• LangChain
• LlamaIndex
• OpenAI
• HuggingFace
• Spark
• Kafka
• Fivetran
• Airbyte
• Prometheus/Grafana

Example workflows

Semantic document search

Input: text chunks embedded by an external model. Steps: create a collection with the embedding dimension, insert records with metadata, run a search with a query embedding and output_fields. Output: top-K matching chunks with requested fields.

Filtered hybrid retrieval

Input: dense query vector plus keyword text and metadata constraints. Steps: store dense and sparse vectors in one collection, define a rerank function, and search with filters. Output: ranked results combining semantic and keyword relevance.

Before you choose

• Self-hosted Milvus is open source under Apache 2.0; running compute and storage is your cost, and open-source code does not include hosted infrastructure.
• Zilliz Cloud pricing is usage-based for Serverless (vCU and storage) and per-CU or per-million-vector for Dedicated; use the pricing calculator for estimates.
• Capacity figures on the pricing page assume 768-dimensional vectors and primary keys only; scalar fields may reduce capacity, so test your own workload.

Listed under

• Data
• AI infrastructure
• Developers
• Open source
• RAG
• Self-hosted
• Semantic search

Get started

Run pip install -U pymilvus to get started locally, or sign up for a managed cluster.

1. Install the Python SDK with pip install -U pymilvus.
2. Create a MilvusClient using a local .db file for Milvus Lite, or a URI and token for a deployed server or Zilliz Cloud.
3. Create a collection with the vector dimension matching your embedding model, then insert data.

The software can be installed under its repository license. Infrastructure and model-provider charges remain separate. Optional commercial hosting and support have their own terms.

Milvus is free to use. See their pricing page ↗

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Milvus: Open-Source Vector Database Built by Zilliz ↗zilliz.com
• GitHub - milvus-io/milvus: Milvus is a high-performance, cloud-native vector database built for scalable vector ANN search · GitHub ↗github.com
• Zilliz Cloud Pricing - Fully Managed Vector Database for AI & Machine Learning ↗zilliz.com
• Home | Cloud | Zilliz Cloud Developer Hub ↗docs.zilliz.com

Company website

zilliz.com: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.