AI agents need context everywhere they run, even where the cloud can't follow

couchbase context smk1
The competitive edge in enterprise AI is changing in context: which platform can give the agent the right memory, the right retrieval, and the right data at decision time.

Couchbase on Tuesday announced its AI Data Plane, which combines persistent agent memory, real-time context retrieval, and an enterprise-managed MCP server into a single operational platform.

Couchbase has its roots in caching and high-transaction databases – an architecture the company argues makes it better suited to agent memory than vendors who ran into the problem with search or analytics. The AI ​​data plane runs in cloud, on-premises, and disconnected edge environments alike, extending agent memory and local vector search to devices without a network connection.

"How do you ensure that the intelligence you get from these models is the same as the database experts?" Gopi Duddy, CTO of Couchbase, told VentureBeat. "How do you get that value out of the storage system, which is still going to be the database?"

What does AI data plane provide

The AI ​​Data Plane packages three components that are designed to replace the fragmented stacks most enterprises currently run.

Agent Memory: A unified persistence layer for conversational context, structured operational data, and vector embeddings. Couchbase says the guardrails are what differentiate it from standalone memory services: per-session token constraints, time-out limits on stored memories, and metering controls that calculate consumption per agent session.

Enterprise MCP Server: An enterprise-supported self-managed server for standardized model-reference protocol integration, shipping as part of the platform rather than requiring a separate service.

Agent List: A function-level catalog of searchable agent tooling built by Couchbase. Duddy distinguished it from metadata catalogs like Databricks Unity or AWS Glue – in his own words, describing it as closer to a glorified MCP that acts as a callable tool within the platform.

Memory-first architecture moves agent context to disconnected edge

Duddy says Couchbase’s pedigree and its core architectural foundation give it an edge when it comes to context.

"Before there was a database we were a cache," Duddy said.

Writing to memory is 10 times faster than writing to disk, Duddy said — a speed advantage he argues differentiates Couchbase from NoSQL databases that layer in-memory workloads on top of disk-based storage.

Couchbase isn’t the only data technology that has its roots in the caching layer. Redis is similarly rooted in caches and also recently announced an agentic AI context layer. Duddy argued that Couchbase is different in that it maintains an ACID (atomicity, consistency, isolation, and durability) compliant database that makes sense for transactional workloads. Couchbase also has a long history in various deployment modalities.

That architecture extends to the edge through the platform’s on-device runtime, Couchbase Lite. It runs SQL, full-text searches, and vector searches locally without a network connection, using a proprietary sync mechanism to replicate bilaterally to the cloud or between edge nodes when connectivity returns. Target environments are retail floor operations, field service, industrial deployments, and regulated settings where agent data cannot leave the device.

Doody cited hotel reservations as an early example: multiple agents serving customers simultaneously, each pulling local context and running vector searches on the device, with shared session memory synchronized centrally. The practical benefit is nominal efficiency. Instead of each agent retrieving and processing the same data independently, the platform caches the shared context so that concurrent sessions can draw on it without burning tokens again and again.

View of Agora from production

Agora, a platform that helps developers embed real-time voice, video, and conversational AI in enterprise applications, has run Couchbase in production since February 2024.

The initial use case was its signaling product, managing channel setup and state synchronization for live calls. The expansion into conversational AI agents revealed strict requirements: memory-first architecture, full JSON support for storage and query, cross-datacenter replication for high availability, and enterprise-grade vendor support.

"Based on these criteria Couchbase was the best fit," Patrick Ferriter, SVP of product at Agora, told VentureBeat.

Agora is now expanding that relationship to support context retrieval for conversational AI agents.

"This will simplify the architecture and provide enterprise grade RAGs with the predictable low latency needed for conversational AI use cases." Ferriter said.

For data professionals trying to figure out the best approach to context, there is no single answer. On platform selection, Ferriter was direct.

"It depends on the priority and goals of the organization including time." Ferriter said. "If they want something enterprise grade and optimal for immediate production and scale versus customizing and maintaining an open-source solution with community support. We wanted that before and that’s why we considered an expanded partnership with Couchbase."

Competitive Context: Following the Right Trend

The reference layer has become a crowded space in 2025.

Oracle inserted a memory core into its databases in March, providing a context layer. Redis added a context layer in May, as did vector-native database vendor Pinecone.

"Couchbase is following the trend, not setting it, but it’s the right one to follow," Devin Pratt, research director of AI, automation, data and analytics at IDC, told VentureBeat. "The real benefit of this is the reach, running a single platform from cloud to mobile through which enterprises really operate. Now the test is to compete with big names."

For teams navigating the vendor landscape, Pratt’s framing is straightforward. "Match the equipment to the workload. Consolidate where it makes sense, use a specialized engine like a graph database where relation-heavy logic earns it, and let governance calls run instead of treating memory as plumbing," Pratt said.



<a href

Leave a Comment