Oracle early leader in pointing vectors at business data, say analysts

Big Red’s 'big announcement' strives to bring LLM technique to the business data arena

Oracle's efforts to bring natural language vector search capabilities to the relational data in business systems is being met with approval among analysts, with one placing it as an early leader in the field.

At its CloudWorld shindig in Las Vegas last week, the business application and database giant announced it was adding AI Vector Search within the 23c database, set to become the next long-term release.

Vector search is essential to exploitation of large language models within a database. It has become a feature of modern NoSQL databases such as MongoDB and Cassandra, while PostgreSQL and SingleStore have additionally introduced it. There is also a distinct category of vector databases from vendors, including Pinecone.

Oracle said it was adding to Oracle Database 23c a native Vector datatype and "optimized" vector similarity search indexing in a move designed to improve performance. New SQL functions and operators are designed to make it easier to create, manipulate, and query vectors in combination with other data models already supported in Oracle Database, including JSON, spatial, and graph data.

Big Red said the new features also support Retrieval Augmented Generation (RAG), a way of combining large language models (LLMs) and private business data to offer responses to natural language questions. It claimed RAG provides higher accuracy and avoids having to expose private data by including it in the LLM training data.

The features are included in Oracle Database 23c as an Oracle Cloud Infrastructure database service. Releases are planned for Oracle Database On-premises, Exadata, Exadata Database Service, Exadata Cloud@Customer, and Autonomous Database (ADB).

Aaron Rosenbaum, senior director analyst with Gartner, said: "Vector Datatypes and Indexes will be table stakes for all databases. The differentiation Oracle brings to the market here is how these new features interact with and leverage all the other features of their products. For example, Vector Search will be combined with relational and graph queries 'Find me books like the one I just read, that are in stock, available in English, that my friends or their friends liked.'

"Oracle has deeper features than many of the databases mentioned above so they have more ways to combine and leverage Vector search across the product," he said.

Vector search would improve on algorithms used to search across customer service, fraud, national security and many other fields. Integrating the feature within a general purpose database would also avoid extracting the data for Gen AI-tasks, he said.

"Both approaches are being used today for building Generative AI applications over business data. There will be a set of users and applications where it's best to keep the embeddings in the same database as the source data. While that won't be 100 percent of Generative AI applications, if it's convenient enough and powerful enough, it will be part of many Oracle customers data management strategy," Rosenbaum said.

Carl Olofson, research vice president at IDC, said Oracle was one of the early leaders in bringing vector search within business data systems.

"Vectors are part of a larger AI/ML solution, and while there are several such solutions for unstructured data by itself (text, image, audio) there are, to my knowledge, none that address structured data, especially relational data. We have seen early demonstrations of their GenAI capability, which will be part of a later version of 23c, in generating SQL, and responding to SQL extensions to carry out proximal searches.

"Oracle's integration of this functionality into Oracle Database makes it more efficient and more operationally efficient than other possible AI approaches aimed at the structured database," he said.

AI Vector Search was Oracle's "big announcement" in databases as it enabled support for vector embeddings within the database system, said Noel Yuhanna, veep and principal analyst with Forrester Research.

"This is still relatively new for all, including Oracle, regarding supporting vector embeddings in a database," said Yuhanna.

"Oracle's value proposition is supporting vector and non-vector data together in the same database to support semantically driven LLMs. In addition, they announced Generative AI support for Oracle databases, with SQL generation via LLM and APEX app generation.

"APEX has become a key differentiator for Oracle, a low-code tool that allows organizations to build applications quickly. APEX app generation takes this tool to the next level by offering no-code App development, which will surely help accelerate use cases," he added. ®

More about

More about

More about


Send us news

Other stories you might like