Once a database specialist, Couchbase is growing beyond its roots toward becoming a data platform for agents with the launch of the AI Data Plane.
Made generally available on Tuesday, the AI Data Plane is designed to provide a governed data infrastructure for agents by unifying frequently disparate capabilities such as agent memory, real-time data retrieval and consistent data access that, when separate, can stall AI development projects.
Specific features include Agent Memory, which enables agents to store and retrieve information from previous interactions, Agent Catalog to help agents discover the tooling that enables them to execute tasks, and a self-managed Model Context Protocol server to standardize integrations between agents and models.
Beyond the AI Data Plane, Couchbase unveiled a new version of its Enterprise Analytics platform that includes Apache Iceberg federation capabilities, so users can query real-time data from Couchbase and lakehouse tables without having to move or duplicate data. In addition, the vendor updated its Capella iQ natural language assistant and introduced tools that extend the AI Data Plane to edge devices.
"Collectively, the AI Data Plane and accompanying features represent a significant update because they collapse previously fragmented data services into a single, governed architecture," William McKnight, president of McKnight Consulting, told TechTarget. "With this update, users can run AI agents more efficiently with lower token costs and higher accuracy."
Devin Pratt, an analyst at IDC, likewise called the update significant, noting that what makes it valuable is what it removes -- the need to piece together separate systems to inform agents -- rather than what it adds.
"This lands right where the market already is," he said. "Enterprises are cutting down the number of data systems they run, and most already have AI agents in or near production."
Based in San Jose, Calif., Couchbase originated as a NoSQL document database platform but, like fellow former database specialists such as MongoDB and Redis, it is expanding to include broader data management and AI development capabilities. Couchbase, which was founded in 2011, was acquired by Haveli Investments in September 2025 for $1.5 billion and is now led by CEO BJ Schaknowski, who replaced Matt Cain following the acquisition.
A foundation for AI
Innovation in data management is largely aimed at helping enterprises build agents and other AI applications.
Despite investing heavily in AI development, many organizationscan't move projects past the pilot stage and into production because they struggle to discover and operationalize the relevant data required for AI tools to deliver accurate outputs.
Collectively, the AI Data Plane and accompanying features represent a significant update because they collapse previously fragmented data services into a single, governed architecture. With this update, users can run AI agents more efficiently with lower token costs and higher accuracy.
William McKnightPresident, McKnight Consulting
Providers from tech giants such as AWS and Microsoft, through full-featured data platform vendors Databricks and Snowflake, to niche specialists including Domo and Tableau, have all introduced new features this year aimed at enabling customers to better connect agents with the right data.
With the AI Data Plane, Couchbase is similarly trying to tackle the problem of providing agents with an accessible data foundation.
Customer feedback combined with Couchbase's own observations drive the vendor to create the new capabilities, according to Barry Morris, the vendor's chief product and strategy officer. He noted that customers repeatedly requested integrated, out-of-the-box memory so they didn't have to integrate disparate systems, and the requests came from enterprises ranging from a toy manufacturer to a payments platform.
"When companies with almost nothing in common independently ask for the same capability in the same language, that's not a feature request — it's a category forming," Morris said. "So we built the layer that removes those walls with Agent Memory, an MCP Server and Agent Catalog shipping as enterprise-supported components in this release."
Agent Memory is designed to provide a unified memory layer within a broader data platform so that teams don't have to piece together tools such as vector databases, document stores and caching capabilities. In addition, it is built to be framework-agnostic so that developers don't have to build new memory layers each time they use a different development framework, such as LangGraph or LlamaIndex.
Meanwhile, Couchbase's inclusion of Agent Catalog and the MCP server within the AI Data Plane consolidate previously available capabilities into a single architecture.
Given that insufficient memory is one of the problems that often stalls AI initiatives, Agent Memory is perhaps the most valuable feature within the AI Data Plane, according to Pratt.
"What stops companies from scaling AI agents usually isn't the model, it's the data underneath," he said. "Get memory right and you've solved the real bottleneck."
In addition, he noted that although Couchbase is on the right strategic path by adding data platform capabilities and consolidating previously fragmented capabilities in a unified architectural layer, the vendor is in step with competitors rather than pacing the market.
Where Couchbase does stand out, however, is with its extension of its capabilities to the edge through new features such as Couchbase Lite that syncs devices over Bluetooth and Wi-Fi, Pratt continued.
"The idea [of a unified data layer] isn't unique anymore," he said. "Couchbase's real advantage is reach, one platform that runs the same way everywhere, which is how enterprises actually operate."
McKnight likewise noted that while Couchbase distinguishes itself from its competitors in some ways, it is largely providing similar features rather than market-moving ones.
"Couchbase differentiates itself from competitors by offering high memory performance, a unified infrastructure that eliminates fragmented tools, and advanced edge capabilities like offline, peer-to-peer syncing," he said. "However, its updates also align with broader industry trends by integrating with standard data governance catalogs and supporting infrastructure consolidation."
Regarding the strategy to evolve beyond its database roots, McKnight added that that Couchbase is making smart moves as it expands.
"Couchbase is clearly on the right track by evolving into a unified operational data platform," he said. "Consolidating previously fragmented vectors, documents and caches into a single architectural layer is what enterprises need to overcome the primary bottleneck of scaling AI agents from simple pilots to production."
Future features
Just as the AI Data Plane is designed to aid AI development, Couchbase's upcoming product development plans focus on enabling customers to move AI projects into production at scale, according to Morris.
Toward that end are the evolution from a database to a data foundation for AI, more consolidation of capabilities, and extension beyond a single environment to include the many systems and devices where users do their work.
"We believe a new category is forming, and that while our heritage … positions us to define it, it's much more than a growth beyond our roots," Morris said. "The database is part of what we do as the operational data platform, but only part of it."
As Couchbase expands, it would be wise to add graph capabilities that discover and connect data in different ways than others, according to McKnight.
"To truly succeed in the next wave of agentic AI, combining this new AI Data Plane with Graph retrieval-augmented generation capabilities would be a competitive differentiator," he said.
Pratt, meanwhile, suggested that Couchbase make governance a focus as it concentrates on enabling AI development.
"The next step isn't storing AI memory, it's making it something companies can trust and afford," he said. "That's what our research keeps pointing to as the real holdup."
Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than three decades of experience. He covers analytics and data management.
