The Pulse
MongoDB Launches Atlas Agent Engine for Production AI Agents
MongoDB introduced Atlas Agent Engine at its September 29 Investor Day, offering runtime, memory, retrieval and governance features for AI agents. The service is in public preview, with consumption-based pricing for its runtime and memory c
AI.info Team · Written with AI ·
MongoDB announced Atlas Agent Engine on September 29 at its Investor Day in New York, pitching the service as a way to run AI agents without assembling separate systems for execution, memory, retrieval and governance. The product is in public preview, and MongoDB says new and existing Atlas customers can start using it now.
One control plane for agent actions
Atlas Agent Engine combines an agent runtime with memory and retrieval capabilities, while adding controls for identity, audit, policy and cost. MongoDB says each action is logged against an identity—human or agent—and governed by policy. The company’s pitch is that these functions can reduce the work of connecting and securing separate services as teams move agents beyond prototypes.
“Organizations that want to put agents in production are being forced into a false tradeoff: either adopt one vendor's runtime and accept being locked into a model and cloud, or piece together a framework and manage governance and memory on their own,” said Pablo Stern-Plaza, Chief Product Officer, AI and Emerging Products, MongoDB.
He said the engine is intended to let customers use different models, frameworks and clouds rather than commit to one provider.
Voyage AI supplies retrieval and memory
MongoDB says Atlas Agent Engine uses Voyage AI embedding models and MongoDB’s native retrieval for memory and search. Its announcement says Voyage AI’s embedding and reranking models rank among the top performers on RTEB, a benchmark the company describes as designed to reflect enterprise retrieval. The release does not provide a specific score or ranking for the models.
Customers can adopt the memory and governance layers separately or use them alongside the runtime, MongoDB says. The company also says the product is designed to work with existing models and frameworks, and to run across cloud, self-managed and laptop environments. Support for open standards including MCP and A2A is intended to make it easier to change frameworks or deployment choices without rebuilding an agent application.
Public preview, with usage-based charges
The Atlas Agent Runtime and Atlas Agent Memory carry consumption-based pricing during the public preview. MongoDB says usage draws on customers’ existing Atlas commitments; the announcement does not give per-unit rates. The release describes the memory and governance layers as optional components, rather than requiring teams to adopt every part of the engine.
Paysafe is among the companies building toward production use. Amar Akshat, Paysafe’s senior vice president of architecture, said its analysts currently piece together data from multiple systems while investigating unusual activity in the payment network. “We're excited about the potential for an intelligent agent, built on MongoDB's Atlas Agent Engine, to shrink the time between a problem emerging and our team acting on it,” he said.
MongoDB’s bet is flexibility with guardrails
The launch extends MongoDB’s database and search offerings into software that can take actions, not just retrieve information. The company says its platform already serves more than 70,000 customers, and frames Atlas Agent Engine as an added layer for organizations that want to build agents on infrastructure they already use.
That breadth is also the product’s main promise: MongoDB is offering a managed route to agent runtime, persistent memory and governance while leaving model and framework choices open. The service remains in public preview, so the announcement establishes availability and the company’s design goals, not evidence of production results across customers.