Zilliz Launches Vector Lakebase, Extending Vector Database with Unified AI Data Platform

Zilliz, the company renowned for the open-source vector database Milvus, has unveiled the public preview of its new product, Vector Lakebase. This significant release for Zilliz Cloud is designed to transform how organizations manage and utilize data for artificial intelligence applications by combining a production vector database with a shared, lake-native data foundation. The core objective of Vector Lakebase is to provide a unified platform capable of handling real-time data serving, interactive data discovery, and batch analytics, all operating on a single logical copy of data. This approach eliminates the need for data duplication and complex custom pipelines, thereby simplifying data management from gigabytes to petabytes. A crucial innovation within Vector Lakebase is its tiered storage model. This model intelligently distributes data across different storage mediums—memory, NVMe, and object storage—to optimize both performance and cost. For instance, frequently accessed data might reside in memory for ultra-low latency, while less active data is moved to object storage, offering substantial cost savings. The platform explicitly highlights its use of object storage for "significantly lower cost" in its Tiered-Storage option, which supports various query per second (QPS) ranges and latency requirements. Furthermore, Vector Lakebase introduces "object-storage-aware indexes." These specialized indexes are built on Vortex, an open columnar format, and are designed to facilitate faster and cheaper random reads compared to traditional formats like Lance and Parquet. By integrating these indexes with object storage, Zilliz claims to cut read amplification by over 90%. This technical advancement is vital for AI agents and models that require rapid and efficient access to vast datasets stored in object storage without incurring high operational costs or performance bottlenecks. The platform also emphasizes its "Full-Spectrum AI Search" capabilities, allowing searches across vectors, text, JSON, and geospatial data with advanced retrieval methods. This is complemented by "On-Demand Search," a pay-as-you-go compute model that bills directly for object storage and compute, rather than relying on serverless markups, offering greater cost control. The overall vision is to create an "active knowledge base" where self-describing objects enable AI agents to reason over data instantly, minimizing manual data preparation and maximizing the value of unstructured data.