en.Wedoany.com Reported - On June 29, 2026, Zilliz announced the launch of Loon, a new storage engine powering Zilliz Vector Lakebase and built into Milvus 3.0, in Redwood City, California. Loon adopts a lake-native foundation, enabling a single copy of vector data to serve real-time search, large-scale discovery, and batch analysis simultaneously, forming the storage layer foundation for Zilliz Cloud's evolution from a vector database to a unified AI data platform.
Vector Lakebase is designed on the premise that a single logical copy of vector data should serve all AI workloads, including production search, discovery, and batch analysis, without the need to replicate or move data between systems. Its storage layer faces challenges: the same dataset must simultaneously support fast record-level lookups for serving and wide scans for analysis, both based on cost-effective object storage. The system must also handle continuously changing data, as teams re-embed, re-label, and re-index the same records as models improve.
James Luan, co-founder and CTO of Zilliz, stated that vector retrieval is no longer the entire problem, and Vector Lakebase is the answer to what comes after the success of vector databases. The winning system will make continuous serving and continuous discovery feel like part of the same machine, which is only achievable when the storage layer serves a single copy of data for each workload. Loon is that storage layer.
To achieve this goal, Loon treats vector datasets as physically heterogeneous in their actual form and is built on three design principles. A hybrid file format stores each column in its most suitable format: scalar and filter fields use Parquet for efficient scanning; dense and sparse vectors use the open Vortex format for fast, byte-accurate row-level reads on object storage; raw videos, PDFs, and images remain in object storage, referenced rather than copied into the database. Row ID alignment enables columns split across different formats to still function as a logical table, allowing new embedding models to be added as their own columns without rewriting stored data. Versioned manifests define the current version of the dataset, including files, indexes, deletion logs, and statistics, allowing serving clusters, on-demand compute, and external engines like Spark and Ray to read and update the same dataset without maintaining separate copies.
These designs enable a single copy of data on object storage to feed multiple engines simultaneously. In internal tests on Zilliz object storage, Loon's Vortex-based layout reduced data pulled per record read by approximately 135 times compared to Parquet. Adding a new embedding model becomes a lightweight version update rather than a massive rewrite. The Vector Lakebase architecture includes real-time serving clusters, elastic on-demand compute, and External Collections indexing, all operating on the same semantic foundation without duplicate pipelines or ETL. Over 10,000 enterprises and AI-native teams have built on Milvus and Zilliz Cloud, including MiniMax, OpenEvidence, Filevine, Exa, and Salesforce.
Loon now powers Milvus 3.0 and serves as the storage layer for Zilliz Vector Lakebase on Zilliz Cloud, available in over 30 regions across AWS, Google Cloud, and Microsoft Azure, with Serverless, Dedicated, and BYOC deployment options. Teams that have scattered online serving, offline analysis, backfilling, and external data lake workflows across multiple systems can create a free account, receive $100 in free credits by registering with a new work email, or contact the Zilliz team to discuss specific use cases.
Zilliz is a leading AI data infrastructure company and the creator of Milvus, the world's most widely adopted open-source vector database, with over 44,000 GitHub stars and more than 100 million Docker pulls. Zilliz helps enterprises and AI startups make their unstructured data searchable, analyzable, and governable, transforming text, images, audio, video, and more into strategic assets for production AI. Zilliz's technology is centered on Milvus and Zilliz Cloud, with Zilliz Cloud extending this foundation into a fully managed Vector Lakebase platform, combining the high-throughput, low-latency serving capabilities of a vector database with the openness, scalability, and cost-effectiveness of a multimodal data lake.
