Teradata adds Enterprise Vector Store to augment RAG

Teradata is adding vector support to its platform in the form of a new Enterprise Vector Store, an in-database offering, to enable and augment use cases that require retrieval augmented generation (RAG), such as agentic AI.

Other features of Enterprise Vector Store include vector data management starting from embedding generation and indexing to metadata management and intelligent search.

It also supports frameworks such as LangChain and planned temporal vector embedding capabilities, which is designed to aid explainability by tracking changes to data over time, thereby improving accuracy and decision making, the company said.

“Teradata’s new in-database vector store will allow enterprises to add AI capabilities by handling high-dimensional data in the form of processing large volumes of vectors, combining structured and unstructured data in one database,” said Devin Pratt, research director of data management at IDC.

Analysts believe that Teradata is not the only data platform provider, which is adding vector store capabilities to augment RAG use cases in order to alleviate the need for enterprise customers to adopt specialist vector databases alongside data platform products.

“Almost all data platform providers are adding support for the storage and processing of vectors as they update their products to support enterprise requirements for applications based on generative AI,” said Matt Aslett, director of research, analytics & data at ISG.

Typically, vector databases are used to store vectors representing features or attributes of data and information, and these vectors can be used to augment generative AI models by providing semantically similar information from an enterprise’s own data via RAG.  

Rival data platform providers such as Snowflake, Databricks, and Google BigQuery have all added vector storage capabilities, said Doug Henschen, principal analyst at Constellation Research.Another example, according to IDC’s Pratt, is Oracle that unveiled HeatWave GenAI with in-database LLMs and vector store capabilities.

Through 2027, almost all enterprises developing applications based on generative AI will invest in data platforms with vector search and retrieval-augmented generation to complement foundation models with proprietary data and content, said ISG’s Aslett. 

Nvidia partnership to accelerate RAG use cases

Teradata has also partnered with Nvidia for the new vector store offering with an aim to accelerate RAG use cases via NIM — a containerized inference microservice that is specifically designed to accelerate the deployment and operational performance of applications, agents and assistants based on generative AI models.

“Developers can fine-tune NeMo Retriever microservices, a part of NIM, in combination with community or custom models to build scalable document ingestion and RAG applications which can be connected to proprietary data wherever it resides,” the company said in a statement.

According to Aslett, NIM is proving attractive to data management software providers as it enables them to accelerate the delivery of support for high-performance operationalization of AI models in a containerized architecture that can be deployed on-premises or in the cloud.

Enterprise Vector Store is currently available in private preview with general availability expected in July.

However, it is unclear whether it will be added to Teradata’s Vantage Cloud. However, several analysts, such as Henschen, believe that although Enterprise Vector Store is debuting on Teradata’s lake environment, it will be integrated into its VantageCloud and IntelliFlex offerings by the second half of this year.

If the new vector store is added to IntelliFlex, Henschen believes that Teradata will become the first and only one to offer a vector store on-premises. “The on-premises vector store capability becomes very important for existing Teradata customers in hybrid and on-premises scenarios looking to tap generative AI capabilities,” Henschen explained.