Couchbase Recasts Itself as an AI Data Plane and Bets on Self-Managed and the Edge

Most of us, me included, still file Couchbase under NoSQL databases, with a strong edge story. With the recent AI Data Plane announcement, they want to be the operational data layer that production AI agents run on. Couchbase has been saying a version of this for a year, and this launch is the version with real product behind it, the AI layer running self-managed and out to the edge, beyond their own cloud.

Self-Managed and Edge Get the Same AI Layer as Capella

Couchbase pulled their agent capabilities under a single name, the AI Data Plane, and made it run across Capella, self-managed deployments, and the edge as a single architecture. A lot of these capabilities already shipped on managed Capella as AI Services back in December, so the managed side is mostly repackaging. The same agent memory, context retrieval, and governed access now run self-managed too, on the clusters a company operates itself, for the enterprises that can’t or won’t put their agents in someone else’s cloud.

Agent Memory Is Only as Good as the Guardrails Around It

The centerpiece is Agent Memory, a single persistence layer for the context an agent carries across sessions. Couchbase argues that a company that grew up on caching and high-transaction workloads handles agent memory better than one that came at it from search or analytics, and there’s something to that. The single store is the easy part. The controls wrapped around it are what make it production-grade. Couchbase can expire a memory once it goes cold and cap how much any one session or agent can spend retrieving it. Let memory pile up unchecked, and the token bill climbs with it. And an agent that grabs stale context will act on it without blinking. A clock on how long memory lives and a ceiling on what it costs are the difference between a memory feature and memory you’d run on a system of record. Agent Memory works across the usual agent frameworks, too, so a team can swap LangGraph for CrewAI or LlamaIndex without rebuilding its memory layer. The one thing I’d want to see more of is conflict resolution. When two memories conflict, which one does the system trust?

MCP Server and Agent Catalog Cover Connections and Governance

Two more pieces sit alongside memory. First, the MCP server provides agents with a standard way to access Couchbase data, with little to no integration code to write. Couchbase has taken it from a community project to an enterprise-grade offering supported by their own engineering team and made it self-managed. Second, Couchbase placed the Agent Catalog under governance, and it shows what your agents are up to. Underneath, it’s really a registry of the functions an agent can call, surfaced inside the platform and wired to the MCP server. While an MCP server is close to table stakes now, the catalog is the more useful piece, serving as a ready-made set of tools an agent can reach for without anyone wiring them up first.

Enterprise Analytics 2.2 Opens Couchbase Data to the Iceberg Lakehouse

Couchbase also shipped Enterprise Analytics 2.2, which lets their analytics engine query Apache Iceberg tables alongside Couchbase’s operational data, with change data capture from Oracle and SQL Server. Plenty of platforms now pitch one governed copy across operations, analytics, and AI, so the idea isn’t Couchbase’s alone. Couchbase federates to the lakehouse from the operational and edge write paths, rather than treating the warehouse as the center and operational data as something to reach back to. Starting where the writes happen is the ground they own.

At the Edge, Couchbase Turns Shared Context into Token Savings

Couchbase continues to harden at the edge, with new offline-first peer-to-peer sync and on-device search that can run without a network connection. While I’m a fan of their increased reach, the cost angle is getting overlooked. Picture a dozen agents on the same retail floor, or handling a rush of hotel check-ins. If each agent calls the model and pulls the same context, the token bill will run up quickly. By caching shared context on the device and syncing it across nodes, the agents can draw from a single copy of data instead of each paying for the same lookup. That’s the token math worth paying attention to.

Where the Repositioning Leaves Couchbase and What Comes Next

With virtually every operational database now racing to be the place agents run, Couchbase had little choice but to recast themselves as an AI Data Plane. It was a necessary move up the stack. The good news is that they now compete in areas where they are already strong: operational and edge write paths, self-managed, and on-prem environments. They leave the model and cloud-scale fights to the platforms built for them. That’s a focused bet. Couchbase can back it up.

The bet Couchbase is making fits the company. The team has spent years learning and building in the operational and edge areas, and the agent era is pulling demand straight toward them. Running agents on your own infrastructure and out at the edge is becoming more common, and that’s exactly the moment Couchbase was built for. The two areas I’m excited to learn more about are memory and the analytics federation. Adding even more details in those two areas can make their strong repositioning even stronger. Couchbase is focused and sure of their lane, and they walk into the agent conversation with something they can call their own.

Mike Leone
VP & Principal Analyst | + posts

Mike Leone is a principal analyst at Moor Insights & Strategy covering data platforms and analytics, data infrastructure and storage, and data governance and enterprise data strategy. He brings 15 years of analyst experience from his work at Enterprise Strategy Group, where he rose to practice director for data management, analytics, and AI. Mike's work is grounded in a strong technical and strategic foundation, including early roles in software and hardware engineering.

Recent Posts