NEW YORK — MongoDB at present launched Atlas Agent Engine, a unified execution, reminiscence, and governance layer for manufacturing AI brokers, at its Investor Day on the Nasdaq MarketSite in New York Metropolis.
Whereas AI brokers show their worth rapidly in proof-of-concept testing, bringing them into manufacturing stays a serious bottleneck. Doing so requires correct retrieval, persistent reminiscence, and enterprise-grade safety and governance. With out a single platform, engineering groups should sew collectively disparate instruments that break each time underlying fashions or frameworks evolve.
“Investigating uncommon exercise in our fee community at present means our analysts stitching collectively information from a number of programs by hand, usually underneath time stress. We’re excited concerning the potential for an clever agent, constructed on MongoDB’s Atlas Agent Engine, to shrink the time between an issue rising and our workforce performing on it, giving our analysts extra time to deal with the judgment calls that matter most,” mentioned Amar Akshat, SVP of Structure, Paysafe.
Atlas Agent Engine offers groups a modular strategy to put brokers into manufacturing. Retrieval is powered by Voyage AI, whose embedding and reranking fashions are among the many prime performers on RTEB, a benchmark constructed to mirror actual enterprise retrieval as a substitute of educational datasets. Prospects can undertake the reminiscence and governance layers independently or with the runtime, utilizing present fashions and frameworks they know.
Atlas Agent Engine is constructed with the ecosystem enterprises already belief. Frontier Labs deliver the very best accessible fashions on to the place enterprise information already lives. System integrator companions deliver the {industry} experience and supply expertise prospects depend on for agent deployments. “Atlas Agent Engine brings the enterprise-ready capabilities, context, and constraints wanted to assist AI brokers ship real-world affect. Mixed with Accenture’s governance, structure, and deep {industry} experience, it creates a robust basis for accelerating AI transformation and delivering outcomes at scale. Our shared dedication to supply and buyer success makes this partnership significantly sturdy,” mentioned Ram Ramalingam, International Lead, SW Engineering & Head of RDE Accenture.
Constructed to get brokers into manufacturing. Enterprises constructing brokers usually run into the identical three challenges: actions no person can govern, brokers that neglect, and lock-in to a single mannequin or framework. Atlas Agent Engine solves all three, grounded in the identical operational platform that greater than 70,000 prospects already run on, with enterprises like Paysafe already constructing towards manufacturing.
“Organizations that wish to put brokers in manufacturing are being compelled right into a false tradeoff: both undertake one vendor’s runtime and settle for being locked right into a mannequin and cloud, or piece collectively a framework and handle governance and reminiscence on their very own,” mentioned Pablo Stern-Plaza, Chief Product Officer, AI and Rising Merchandise, MongoDB. “With the launch of Atlas Agent Engine, that false tradeoff ends at present. Enterprises get the real-time context their brokers want, with governance and safety inbuilt from the beginning, and the liberty to run any mannequin, any framework, and on any cloud. We didn’t wish to ask prospects to foretell the long run. We wished to construct one thing that works it doesn’t matter what they select.”
Ruled by default. Most platforms deal with id, audit, guardrails, and value controls as separate programs groups should sew collectively themselves. Atlas Agent Engine places all of it behind one management airplane: each motion is logged in opposition to an actual id, human or agent, and ruled by coverage that may’t be quietly switched off. Governance is inbuilt, not bolted on after launch. So when somebody asks what an agent did and who approved it, the reply takes seconds, not weeks. And since governance, reminiscence, and retrieval run as one system as a substitute of stitched-together companies, there’s much less to safe and fewer locations for issues to interrupt.
Reminiscence and retrieval inbuilt. With out built-in reminiscence, brokers begin each dialog from zero, and groups find yourself rebuilding reminiscence infrastructure for each new agent. Atlas Agent Engine builds reminiscence into the platform itself, utilizing Voyage AI embeddings and MongoDB’s native retrieval, so brokers get extra correct whereas spending fewer tokens. It additionally freezes an agent’s full state to disk whereas it waits on human approval or an exterior set off, then thaws it immediately when wanted, so enterprises by no means pay for idle brokers.
Standardizing on one mannequin, cloud, or framework is among the riskiest infrastructure bets a pacesetter could make in a market that strikes this quick. Atlas Agent Engine is impartial throughout AI fashions and frameworks. As a result of it’s constructed on open requirements like MCP and A2A, altering course later solely takes a configuration change quite than an costly rebuild. It can additionally run throughout any cloud, self-managed or perhaps a laptop computer, so the identical agent works all over the place with out rebuilding cloud by cloud.
Atlas Agent Engine provides ruled execution, reminiscence, and value management on prime of what groups already run, quite than asking them to interchange it. Atlas Agent Engine, MongoDB 9.0, and Atlas Infinite, additionally introduced at present, construct on one another: MongoDB 9.0 strengthens the muse each MongoDB buyer already runs on; Atlas Infinite removes the bounds on how that basis can scale and Atlas Agent Engine places AI brokers to work on prime of each, ruled and grounded in real-time information. Along with Voyage AI’s industry-leading embedding and retrieval fashions, already usually accessible, this extends MongoDB’s clever information platform for contemporary functions within the multi-cloud and AI period.
Atlas Agent Engine is out there at present in public preview. New and present Atlas prospects can get began at agentengine.mongodb.com.
