PUBLIC PREVIEW
Atlas Agent Engine
Build, deploy, and govern agents in production with Atlas Agent Engine. One foundation for execution, memory, and governance.
Built for agents in production
Long-term memory, built in
Agents that pick up where you left offSemantic, episodic, taxonomic, and procedural memory are managed natively, so agents get more accurate over time while spending fewer tokens.
Retrieval powered by Voyage AI
Accuracy built for enterprise dataLeverage the top-ranked embedding and reranking models on RTEB for intelligent and performant retrieval.
Real identity on every action
Least privilege on every tool callEvery tool call runs as a real user or agent identity, authorized before it executes, inside an isolated sandbox.
Your team, your policies
Guardrails that can’t be switched offPolicies cascade from the org level and can't be overridden downstream. Route risky actions to human approval and control cost without code.
Audit what agents did and why
Every run traced, nothing to instrumentThe runtime traces every model call, tool call, and memory access, so what an agent did and why is answered in seconds instead of weeks.
No lock-in, by design
Change course with a config changeAtlas Agent Engine is neutral across models and frameworks, built on open standards like MCP and A2A. Switching later takes a configuration change, not a rebuild.
Public Preview Pricing
Pay for what you use
With consumption-based runtime pricing, you pay for agent execution you actually use, nothing more.
Adopt what you need
Adopt memory and governance independently or with the runtime. You pay for the layers you use, on the models and frameworks you already have.
Scale as you grow
Grow from one agent to thousands on the same scalable foundation, with costs that track the work your agents actually do.
Runtime
Time spent by agents executing requests$0.04 per 1000 seconds per vCPU
Memory Storage
Docs sent to long-term memory$0.25 per 1000 long-term memory docs stored
Memory Retrieval
Docs retrieved from memory$0.50 per 1000 memory docs retrieved
FAQs
MongoDB Atlas Agent Engine is an agent platform built on Atlas that enables enterprises to design, deploy, and govern AI agents in production on top of their real-time operational data, with memory, retrieval, and governance built into the platform instead of assembled from point tools.
Atlas is the data platform where your real-time data, search, and vectors live. Atlas Agent Engine is the agent layer on top of that foundation, using the same document model, search, and vector capabilities to give agents the context and state they need to act in production.
No. Atlas Agent Engine uses MongoDB Atlas as its native platform foundation, but it isn’t limited to data stored in an Atlas database. It can access data where it already lives, including sources such as Salesforce, Oracle, Snowflake, S3, SharePoint, and internal systems, without a lift-and-shift effort. Atlas provides the unified foundation for agent memory, retrieval, state, and governance, while your agents access the context they need across systems.
Many early agents run on fragile stacks—separate vector stores, caches, and custom pipelines—making them hard to scale, govern, or even audit. Atlas Agent Engine consolidates those systems into a single platform: a single control plane, unified memory and retrieval, and per-invocation isolation, so you can move from impressive demos to reliable, review-ready production systems.
Atlas Agent Engine is designed to be open and flexible: It works with any LLM or framework and uses open standards like the Model Context Protocol (MCP) for tools, A2A for agent-to-agent delegation, and OpenTelemetry for traces, so it can plug into the AI stack you already run today and adapt as it evolves.
Atlas Agent Engine enforces least-privilege policy before every tool call, runs each invocation in an isolated sandbox, and logs every action with the identity that authorized it. Audit, cost controls, and guardrails are set once at the org level and enforced automatically, not stitched together across systems.
Instead of agents starting every conversation from zero, Atlas Agent Engine provides platform-level memory—semantic, taxonomic, episodic, and procedural—backed by Atlas search and vectors so agents retrieve precise, current context, answer faster, and consume far fewer tokens in the process.
Bring your next generation of agents to production
- Govern every agent centrally
- Inherit memory and retrieval
- Enforce least privilege by design
- Prove isolation and auditability
- Stay flexible across frameworks and LLMs