Startup Zilliz Inc. today debuted a new release of its flagship offering, a managed vector database called  Zilliz Cloud that artificial intelligence models can use to hold information.

Redwood Shores, California-based Zilliz is backed by more than $110 million in funding. It offers an open-source vector database called Milvus. Zilliz Cloud, the offering the company updated today, is a paid cloud-based version of Milvus that includes additional features and removes the need for customers to manage the underlying infrastructure.

AI models don’t process files such as documents in their original form, but first turn them into embeddings. Embeddings are a type of vector, a mathematical structure that lends itself well to storing information. Vector databases such as Zilliz Cloud are specifically optimized to hold AI models’ embeddings.

Zilliz Cloud ships with a search engine that enables an AI application to enter a piece of data and find all the similar records on file. A neural network optimized to generate shopping suggestions, for example, could enter a smartphone description provided by a shopper and have Zilliz Cloud return a list of matching devices. The new release of the platform that debuted today promises to process such queries up to 10 times faster than earlier versions.

The upgraded search engine can process multiple types of queries. According to Zilliz, it’s capable of running so-called single-vector searches in which an AI model enters one record and the database retrieves all the similar files it stores. The platform also supports more advanced use cases such as hybrid searches. Those are requests that specify the files Zilliz Cloud retrieves should not only be relevant to the user’s query, but must also contain the same keywords. 

To speed up searches, developers often equip their databases with a so-called index. This is a collection of shortcuts that can be used by AI models to more quickly sift through contents of a database and find the record requested by the user. According to Zilliz, the new version of Zilliz Cloud includes a feature called AutoIndex that automates the index generation process to save time for customers. 

“As companies scale up similarity search for recommendation systems, RAG and image retrieval, they face a choice between overprovisioning hardware or spending months tuning vector indexes,” said Zilliz founder and Chief Executive Officer Charles Xie.

Zilliz Cloud also promises to ease other aspects of AI projects. The new release supports a specialized type of index that is a unified IVF-based and graph-based index that not only includes shortcuts to the files an AI model uses, but also narrows down the amount of information the model must sift through to find specific items. Additionally, Zilliz has equipped the platform with performance optimizations for Intel Corp. and Arm Holdings plc processors.

Image: Zilliz