Cloud database-as-a-service provider Couchbase Inc. today added some powerful new capabilities to its platform that should enhance its ability to support more advanced generative artificial intelligence workloads.

The company said its database is getting support for vector search and retrieval-augmented generation, as well as integration with LlamaIndex and LangChain. The new features will help to streamline the development of generative AI apps, the company promised.

Over the years, Couchbase has become popular for its operational data store, which offers a novel mix of transactional and analytical functionality. Last year, it beefed up its offering with support for SQL++ and the addition of a columnar data type, illustrating its desire to respond to its user’s latest needs.

The company is continuing that trend with the release of Couchbase 7.6, and today’s updates will become generally available in Couchbase Server, the on-premises version of its database, and Capella, the cloud-hosted version, before the end of the quarter. They’ll also be added to Couchbase’s mobile and edge applications in beta within the same timeframe.

The platform is gaining support for vector similarity search, hybrid search and support for open-source tools that can help customers connect their data to large language models. Vector search is a key addition that will enable Couchbase to leverage vector embeddings stored directly in JSON documents to dig up more relevant search results. Vector embeddings are mathematical representations of data, and they’re especially useful for unstructured information that cannot be stored in traditional rows and columns.

By enabling enterprises to search unstructured data more efficiently, Couchbase is supporting the development of more powerful generative AI models. After all, the capabilities of generative AI are limited by the information they can access. Given that something like 90% of all enterprise data is said to be unstructured, vector search is becoming an increasingly essential requirement for every organization.

Scott Anderson, Couchbase’s senior vice president of product management and business operations, said vector search will enhance the results of each query its database handles. He added that the company’s ability to support vector search from the cloud to the edge will be a game-changer for many customers.

“Couchbase is seizing this moment, bringing together vector search and real-time data analysis on the same platform,” he explained. “Our approach provides customers a safe, fast and simplified database architecture that’s multipurpose, real-time and ready for AI.”

It’s a timely update because businesses today are racing to build more adaptive applications powered by generative AI, with common use cases including chatbots, recommendation systems and semantic search. For instance, if a customer is looking to buy some shoes from an online retailer that complement a particular outfit, Couchbase’s new capabilities will enable them to narrow down their search by uploading a photo of that outfit to a mobile app. Couchbase will then perform a hybrid search that encompasses vectors, text, numerical ranges, geospatial data and operational inventory queries to help the customer find a better match.

Those search capabilities are further boosted by retrieval-augmented generation, or RAG, which enables LLMs to tap into customer’s private databases to enhance their knowledge beyond the training data they were initially taught with.

The integrations with LangChain and LlamaIndex are also important, the company said. They enable the database to connect to a broad library of open-source LLMs that are customized for various applications. According to the company, the integrations will accelerate query prompt assembly, improve response validation and facilitate accelerated RAG workloads.

LangChain co-founder and Chief Executive Harrison Chase said retrieval has become the predominant way to combine data with LLMs. “Many LLM-driven applications demand user-specific data beyond the model’s training dataset, relying on robust databases to feed in supplementary data and context from different sources,” he said. “Our integration with Couchbase provides customers another powerful database option for vector stores so they can more easily build AI applications.”

Constellation Research Inc. analyst Doug Henschen said that although many of today’s generative AI applications are already very impressive, the next generation will be far more advanced. However, before these applications can be built, companies must put the tools and infrastructure in place to support them. And if possible, they want to keep things as simple as possible.

“Organizations are increasingly looking at ways to consolidate and simplify technology stacks and manage cost,” Henschen said. “With the addition of vector search capabilities, Couchbase is reducing complexity and delivering a multipurpose database platform that addresses needs from cloud to edge to on-premises. This will let organizations do more on one, unified platform to accelerate the development of adaptive applications.”

Image: GaryKillian/Freepik