Single governed platform reduces token consumption by up to 45% and infrastructure complexity across cloud, data center, and edge deployments
SAN JOSE, Calif. – October 20, 2026 – Couchbase, the Operational Data Platform for AI®, today announced a major expansion of its AI Data Plane with new capabilities including native Amazon Bedrock integration with semantic caching, managed Agent Memory and a managed MCP server on AWS, in addition to vector search within Couchbase Operational Insights. Together, these innovations give enterprise teams a single governed platform to replace fragmented data services that slow agent deployments, reducing both infrastructure burden and the cost of running production AI at scale.
Enterprise teams deploying production AI agents are finding that general-purpose databases and stitched-together data services cannot keep pace with agentic workloads. They need a converged data layer that serves both agents and applications with live, trusted data at scale in the cloud, in a private data center, and at the edge, even offline. Couchbase is the only operational data platform built to do exactly that. The new capabilities announced today extend that foundation with managed cloud services and token cost reductions that help enterprises accelerate production AI.
Agents have become the primary consumers of enterprise data infrastructure, and most of that infrastructure was not built for them. Machine traffic surpassed human traffic in June 2026, with 57.5% of web requests now automated rather than human-initiated, eighteen months ahead of industry forecasts1. Agents generate 450% more data-layer traffic per task than humans performing the same work2, and 83% of organizations report needing infrastructure upgrades before they can run production-grade agentic AI3. Most production databases were designed for human-scale query patterns, predictable throughput, and data that lives in one place. Agents are highly concurrent, fan out into parallel tasks, need both operational and historical data in a single reasoning loop, and increasingly operate where the data is born, whether at the edge, on devices, or across regions.
Couchbase’s elastically scalable, distributed architecture serves millions of concurrent low-latency operations without bottlenecks. Its memory-first design and shared-nothing topology mean parallel agent requests resolve at sub-millisecond latency rather than queuing behind each other. Couchbase unifies key-value, SQL, full-text search, vector search, e real-time columnar analytics on a single JSON data set, so agents get operational and historical context in one call instead of a round trip through separate systems. Couchbase Mobile lets enterprises run AI-capable applications on devices and at the edge with full functionality even when connectivity drops. It runs the database and native vector search directly on-device, and active-active cross-data-center replication keeps every region writable and in sync, so agents reason and act where data is born while sharing one source of truth.
Mavenir Networks carries calls, messages, and data for hundreds of millions of people. The data layer underneath cannot blink. The company has run its subscriber profiles and live session state with Couchbase since 2017, and is now expanding that deployment to support its AI initiatives.
“Couchbase has been a key component in achieving high availability in Mavenir’s voice solutions for years,” said Brandon Larson, SVP & General Manager, Cloud, AI & IMS Business Strategy. “Now we’re creating new AI voice services that require getting reliable context fast and Mavenir is building that context layer with Couchbase because of their proven performance in production.”
Native Amazon Bedrock Integration with Semantic Caching
Couchbase has expanded its Model Service with native Amazon Bedrock integration, giving enterprise teams serverless, pay-per-token access to hundreds of foundation models along with vectorization over private networking. Inference and embedding generation now run alongside the data, without separate GPU instance management or overhead. This opens a direct path to semantic caching for applications and agents built on Bedrock, where semantically similar queries are served from cache instead of re-invoking the model.
At production scale, matching the right model to each task is one of the most effective levers for controlling cost and latency. Native Amazon Bedrock integration gives Couchbase customers governed access to hundreds of available models without adding infrastructure or leaving the platform.
Managed Agent Memory and Managed MCP Server on AWS
In June, Couchbase launched Agent Memory and an enterprise-supported MCP server as self-managed capabilities. One quarter later, both are available as fully managed cloud services on AWS.
Managed Agent Memory gives agents a durable, low-latency memory layer that persists profiles, conversation history, and vectorized facts across sessions, so agents pick up where they left off instead of starting cold on every interaction. Managed MCP server gives agents and tools a governed, open-standard path to Couchbase data, with nothing to stand up or operate. Both are framework-agnostic, validated with LangGraph, CrewAI, and LlamaIndex, so engineering teams can switch or combine orchestration frameworks without rebuilding their memory layer.
Vector Search in Couchbase Operational Insights
Infrastructure-level requirements change when agents drive query volume. Data freshness is measured in seconds, pattern recognition across structured and unstructured data must be completed in a single pass, and durable context must be stored alongside the data it describes. Couchbase Operational Insights now supports vector search alongside structured analytical queries against fast-moving operational data. Teams can run semantic and structured analysis together, segmenting by meaning with vector search and then aggregating the results in a single SQL query, with no ETL pipeline and no secondary system to synchronize.
Traditional analytics answers questions built from counts, sums, and group-bys. Vector search answers questions built from similarity, intent, and themes drawn from unstructured, nested JSON. Combined, they allow queries that were previously impossible without multiple systems. A query like – revenue by region, filtered by clusters of frustrated customers – resolves in one pass because JSON is analyzed in place and never flattened. Couchbase’s write-back function then sends query results straight into the operational data that applications and agents already read, so an insight lands while it can still change the outcome.
Convergence, Not Just Consolidation
For production AI deployments, the data layer increasingly determines both the speed to production and the cost of staying there. Consolidation reduces the number of systems an enterprise pays for. Convergence changes the architecture itself. Couchbase customers who have converged fragmented data services onto a single platform report TCO reductions exceeding 50% across more than half of deployments. The AI Data Plane extends that cost advantage into AI economics. One platform now serves caching, documents, queries, search, analytics, and AI context against a single copy of the data. It does so everywhere the enterprise operates, using the standards its teams already chose, from SQL to open table formats to the agent frameworks its developers picked.
Semantic caching through native Bedrock integration has been demonstrated to reduce LLM calls by 40 to 60%. Agent Memory, benchmarked against leading alternatives on the LoCoMo standard, cuts token usage by 33 to 45%. Vector-driven context retrieval produces prompts 70% smaller than full-context alternatives, shrinking input token volume proportionally. Every agent built on the AI Data Plane starts with its memory, context retrieval, caching, and operational data already governed and connected, so teams spend engineering cycles on agent logic rather than integration plumbing. Early adopters report compressing production timelines from quarters to weeks.
“Customers building production AI agents at scale need a converged data layer, but most are still running three or four separate systems for memory, vectors, operational data, and analytics, each carrying its own sync logic, security model, and failure mode,” said Barry Morris, Chief Product and Strategy Officer at Couchbase. “What we’re shipping now collapses that stack. The new capabilities all sit on one governed platform that runs identically in the public cloud, a private data center, or at the edge. This is helping Couchbase customers reduce costs, accelerate productivity, and realize faster ROI on AI initiatives.”
Disponibilità
All capabilities listed above are generally available today. Pricing and packaging details are available at www.couchbase.com/pricing.