Announcing Milvus 3.0: Lake-Native Vector Search and a More Powerful Retrieval Engine

What this post added

Milvus 3.0 introduces a lake-native path for indexing and serving vector data directly from object storage using open table formats (Parquet, Lance, Iceberg, Vortex) via External Collections. It enhances the retrieval engine with server-side sorting, aggregation, faceted search, StructArray, and an improved sparse index. Loon (Storage v3) is a new manifest-based columnar storage engine for S3-compatible object storage designed to reduce read amplification for point reads. Snapshots provide point-in-time, read-only views for offline jobs. A Spark DataSource V2 connector is added for batch workflow integration. Online schema evolution (adding/dropping columns) is now supported without downtime. New data model features include nullable vectors, TEXT LOB, TTL, MinHash, and Woodpecker.