MongoDB launches Atlas Agent Engine for AI production

MongoDB has launched Atlas Agent Engine, a new layer for running AI agents in production. The product is its latest effort to make the Atlas platform more central to enterprise AI deployments.

The launch comes alongside MongoDB 9.0, a new version of its database, and Atlas Infinite, a deployment option for Atlas designed to handle sudden jumps in demand. MongoDB has also partnered with Cognition on a service intended to help companies move legacy applications onto MongoDB Atlas more quickly.

Atlas Agent Engine targets a problem many companies have encountered while experimenting with generative AI and software agents. Early demonstrations can be built quickly, but moving those systems into production often requires teams to connect separate tools for data retrieval, memory, identity management, audit trails and policy controls.

Built on Atlas, Atlas Agent Engine combines execution, memory and governance in a single layer. Customers can use the memory and governance components on their own or with MongoDB's runtime, while continuing to use existing AI models and frameworks.

The product is available in public preview. Pricing for Atlas Agent Runtime and Atlas Agent Memory will be consumption-based and will count against existing Atlas commitments.

Governance focus

Governance is a central part of the launch. Atlas Agent Engine places identity, auditability and policy controls under one control plane so each action taken by an agent can be traced to a human or machine identity.

MongoDB is also positioning the product as an alternative to adopting a single supplier's AI stack. The system is designed to work across models, frameworks and cloud environments, using open standards including MCP and A2A.

Memory and retrieval are another part of the offering. Atlas Agent Engine uses Voyage AI embeddings and MongoDB's retrieval technology so agents can maintain context across tasks and interactions.

"Organisations 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.

"With the launch of Atlas Agent Engine, that false tradeoff ends today. Enterprises get the real-time context their agents need, with governance and security built in from the start, and the freedom to run any model, any framework, and on any cloud. We didn't want to ask customers to predict the future. We wanted to build something that works no matter what they choose."

Platform changes

MongoDB 9.0 and Atlas Infinite expand the broader platform on which Atlas Agent Engine sits. MongoDB 9.0 is generally available, while Atlas Infinite is in public preview.

According to MongoDB, version 9.0 delivers up to 35% faster find-one queries, up to 30% faster update-one queries and as much as 2x throughput on large instances compared with MongoDB 8.0. The release also adds support for prefix, suffix and substring searches on encrypted data through its Queryable Encryption feature.

Atlas Infinite separates compute from storage so each can scale independently. This is intended to help customers handle unpredictable traffic and AI-related workloads without rewriting applications.

MongoDB is also renaming its current Atlas offering Atlas Core. Atlas Core and Atlas Infinite will be the two deployment options within the Atlas platform.

"The growing use of AI agents is causing applications to use and store more data," said Ben Cefalo, Chief Product Officer, Core Products, MongoDB. "MongoDB 9.0 hardens the foundation that more than 70,000 customers already trust with extreme performance, making it significantly faster at scale for mission-critical transactional workloads and AI-powered applications. Meanwhile, Atlas Infinite delivers the elasticity, responsiveness, and accuracy that agentic workloads demand, so enterprises never have to stretch legacy architecture to keep up."

Customer and partner use

MongoDB used customer and partner examples to illustrate the intended uses for the new products. Paysafe said it saw potential for Atlas Agent Engine in payment-network monitoring, while Okta pointed to Atlas Infinite as a way to manage unpredictable authentication traffic.

Coinbase said MongoDB 9.0's Intelligent Workload Management feature could help clusters remain responsive during periods of intense trading activity. PicPay, a Brazilian fintech company, said it had tested Atlas Infinite at four times its normal peak traffic for two hours without failures.

MongoDB also announced a partnership with Cognition, the company behind Devin, an AI coding product. The joint service, called Devin for MongoDB Modernizations, links Cognition's code transformation tools with MongoDB's Application Modernization Platform.

The aim is to help enterprises rewrite business logic, data access layers and related code while moving application data into Atlas. Engineering teams would still make decisions on target data models and cutover sequencing, while the combined tooling would handle more of the migration process.

"Most enterprises know they need to modernize their legacy applications, but the cost, complexity, and sheer amount of code involved have forced them to do it one application at a time," said Dev Ittycheria, Chief Executive Officer, MongoDB. "AMP was built to change the economics of modernization by giving customers an automated path from legacy applications to MongoDB. By integrating Devin's code transformation and agentic capabilities into AMP, we can automate even more of that journey and dramatically expand the amount of an enterprise's application estate that can be modernized at once. The result is bigger than a faster migration. Customers can move decades of legacy applications onto MongoDB in months instead of years, giving them a modern data foundation built to run AI applications and agents in production on live operational data."