Milvus SDK Evolution

4/16/2025

Milvus SDK v2 introduces native async support (e.g., AsyncMilvusClient in Python), a Schema Cache to reduce repeated schema validation and improve performance, and a unified API across languages. The Python SDK's MilvusClient simplifies collection creation, schema definition, index building, and data loading into a single, streamlined operation. The RESTful API now has near feature parity with gRPC. The MCP Server is introduced for AI integration.

7/28/2023

Milvus 2.2.12 introduces RESTful API support, allowing users to access Milvus without installing a client, simplifying client-server operations. It also enables vector retrieval during Approximate Nearest Neighbor (ANN) searches by allowing the vector field to be an output field in HNSW-, DiskANN-, or IVF-FLAT-indexed collections. The `json_contains` function is added for JSON array operations. Performance enhancements include reduced overhead for extensive topk searches, improved write performance in partition-key-enabled and multi-partition scenarios, and optimized CPU usage for large machines. Various issues related to disk usage, compaction, data deletion, and bulk insertion failures are also addressed.

3/21/2022

This post details the Milvus Protocol API, including milvus.proto, common.proto, and schema.proto, which define the RPC interfaces and data structures for interacting with Milvus. It also introduces the ORM-style API of PyMilvus 2.0, explaining its object-oriented approach to managing Milvus connections, collections, partitions, and indexes, and how it abstracts away underlying RPC calls.

12/31/2021

Introduces Milvus_CLI, a command-line tool built on the Milvus SDK for Python. It enables users to connect to Milvus, create collections, list/describe collections, import data, query data, and delete collections using interactive commands. It supports all platforms and offers both online and offline installation methods.