Milvus

Milvus is a cloud-native, open-source vector database built for massive-scale vector search — handling billions of vectors with GPU-accelerated indexing and a fully distributed architecture. It is the most feature-rich vector database for enterprise deployments that demand high throughput, horizontal scaling, and multi-tenancy.

• Billion-scale search — distributed architecture with sharding and replication supports datasets far beyond what single-node vector databases can handle
• GPU-accelerated indexing — build indexes on GPU for dramatically faster index construction on large datasets
• Multiple index types — choose from IVF_FLAT, IVF_SQ8, HNSW, DiskANN, and GPU-based indexes depending on your latency, memory, and accuracy requirements
• Hybrid search — combine dense vector search with sparse (keyword) search and scalar filtering in a single query
• Schema-defined collections — define typed schemas with multiple vector fields, scalar fields, and dynamic JSON fields
• Time travel — query data as it existed at a specific point in time for reproducibility and debugging
• Streaming inserts — ingest data in real-time without pausing searches, with consistent read-your-write semantics

Enterprise ML teams, search infrastructure engineers, and organizations operating at massive scale use Milvus when they have outgrown single-node vector databases. It is the vector database of choice when your dataset is in the hundreds of millions to billions of vectors range, or when you need features like multi-vector search and GPU indexing.

Milvus is maintained by Zilliz, which also offers Zilliz Cloud — a fully managed version of Milvus that eliminates operational overhead. For lighter workloads, Milvus Lite provides an embedded mode similar to Chroma for development and testing.

1. Run Milvus Lite (embedded, for development):

   pip install pymilvus
   
   from pymilvus import MilvusClient
   client = MilvusClient('milvus_demo.db')  # Local file-based storage

2. Or run with Docker Compose (standalone mode):

   wget https://github.com/milvus-io/milvus/releases/download/v2.4.0/milvus-standalone-docker-compose.yml -O docker-compose.yml
   docker compose up -d

3. Or use Zilliz Cloud — create a free managed cluster at cloud.zilliz.com.
4. Install the Python SDK:

   pip install pymilvus

5. Create a collection and insert data:

   from pymilvus import MilvusClient
   
   client = MilvusClient('http://localhost:19530')
   
   client.create_collection(
       collection_name='documents',
       dimension=1536
   )
   
   client.insert(
       collection_name='documents',
       data=[
           {'id': 1, 'vector': [0.1, 0.2, ...], 'text': 'ML basics', 'source': 'docs'},
           {'id': 2, 'vector': [0.3, 0.4, ...], 'text': 'Deep learning', 'source': 'wiki'}
       ]
   )

6. Search:

   results = client.search(
       collection_name='documents',
       data=[[0.1, 0.2, ...]],  # Query vector
       limit=5,
       filter='source == "docs"',
       output_fields=['text', 'source']
   )

Pricing: Open-source: free, self-hosted (standalone or distributed). Milvus Lite: free, embedded. Zilliz Cloud Free: 1 cluster, 1M vectors. Zilliz Cloud Standard: from ~$65/month. Enterprise: custom pricing with dedicated infrastructure and SLAs. Full pricing details.