Milvus Vector Database Adapter

This is a community-contributed adapter for integrating Milvus with Cognee.

About Milvus

Milvus is an open-source vector database built to power AI applications. It provides high-performance similarity search and supports various index types, making it ideal for AI applications requiring fast and accurate vector searches.

Installation

1. Install the required dependencies:

   # Option 1: Install dependencies directly
   pip install pymilvus>=2.5.0
   pip install milvus-lite>=2.4.0  # Linux/Mac only
   
   # Option 2: Install as a package (if published)
   pip install cognee-milvus-adapter
   
   # Option 3: Install from source
   cd community/adapters/vector/milvus
   pip install .
   

2. Import and register the adapter in your code:

   from cognee_community_vector_adapter_milvus import register
   

Configuration

Configure Cognee to use Milvus:

# For local Milvus Lite
cognee.config.vector_db_provider("milvus")
cognee.config.vector_db_url("path/to/milvus.db")
cognee.config.vector_db_key("")  # No key needed for local

# For remote Milvus server
cognee.config.vector_db_provider("milvus")
cognee.config.vector_db_url("http://localhost:19530")  # Milvus server URL
cognee.config.vector_db_key("your_milvus_token")  # If authentication is enabled

Usage Example

import cognee

# Register the adapter (importing the module registers "milvus" with cognee)
from cognee_community_vector_adapter_milvus import register  # noqa: F401

# Configure Milvus
cognee.config.set_vector_db_config(
    {
        "vector_db_provider": "milvus",
        "vector_db_url": "./milvus.db",
        "vector_db_key": "",
    }
)

# Use Cognee normally
await cognee.add("Your data here")
await cognee.cognify()
results = await cognee.search("search query")

Features

• High-performance similarity search: Optimized for large-scale vector operations
• Multiple index types: Supports various indexing algorithms (IVF_FLAT, IVF_SQ8, etc.)
• Horizontal scaling: Can handle billions of vectors
• Hybrid search: Combines vector similarity with scalar filtering
• Enterprise-grade: Production-ready with monitoring and management tools

Testing

Run the tests to verify the adapter works correctly:

pytest tests/unit             # offline contract tests, no server needed
python tests/test_milvus.py   # integration test, needs Milvus + LLM keys

Dependencies

• pymilvus>=2.5.0,<3: Official Milvus Python client
• milvus-lite>=2.4.0: Lightweight version of Milvus (Linux/Mac only)

Deployment Options

Local Development (Milvus Lite)

• Use milvus-lite for local development and testing
• No server setup required
• File-based storage

Production (Milvus Server)

• Deploy Milvus server using Docker, Kubernetes, or cloud services
• Supports clustering and high availability
• Better performance for production workloads

Support

For issues specific to this adapter:

1. Check the Milvus documentation
2. Create an issue in the main Cognee repository with the "community-adapter" label
3. Refer to the example and test files for usage patterns

License

This adapter is licensed under the Apache 2.0 license, same as the main Cognee project.