Skip to main content

Atlas Agent Engine: Frequently Asked Questions (FAQ)

Written by Venkat Cherukuri

Product Overview & Architecture

1. What is an agentic platform, in general?

An agentic platform is purpose-built infrastructure designed to run and scale AI agents in enterprise environments. Rather than requiring users/builders to assemble fragmented services themselves, an agentic platform provides:

  • Lifecycle Management: Developing, local testing, packaging, deploying, and versioning agents.

  • Enterprise Governance: Role-based access control (RBAC), API key management, audit logging, and runtime policy enforcement.

  • Observability: Real-time distributed tracing, execution logging, and debugging across model and tool calls.

  • Secure Sandboxed Tool Execution: Allowing agents to run tools, call APIs, and connect to internal systems within strict network and credential boundaries.

  • State & Memory: Providing short-term checkpointing and long-term semantic memory so agents maintain context over time.

2. What is Atlas Agent Engine?

Atlas Agent Engine is MongoDB’s managed service for building, deploying, and operating secure, observable AI agents in production.

It bridges your existing application data and LLMs with a managed execution layer. By combining MongoDB’s document and vector search capabilities with managed execution sandboxes, Atlas Agent Engine helps teams move AI agents from experimental prototypes to enterprise-grade production applications with built-in governance, memory, and audit controls.

3. How does Atlas Agent Engine work?

Atlas Agent Engine integrates directly with your MongoDB Atlas environment:

  • Develop Familiar Workflows: Build your agent locally in Python or TypeScript using native LangGraph abstractions and the agentengine CLI.

  • Deploy to Managed Workspaces: Deploy code into isolated workspace namespaces running on managed infrastructure.

  • Automated Orchestration & Audit: Every LLM invocation, memory lookup, and tool execution is routed through the Orchestration Engine (OE), which verifies permissions and records immutable audit traces.

  • Persistent State & Memory: Conversation states and long-term semantic memories are durably stored in MongoDB collections, enabling multi-turn context and human-in-the-loop approvals.

Please refer to our official documentation for additional details - Atlas Agent Engine.

4. Does the Atlas Agent Engine favour specific models or cloud vendors?

No. Vendor neutrality is a foundational design principle of Atlas Agent Engine.

Enterprises are not locked into a single model family or cloud provider:

  • LLM Freedom: Connect to OpenAI, Anthropic, Google Gemini, AWS Bedrock, Azure OpenAI, or self-hosted models using your own API credentials.

  • Framework Support: Public Preview provides native support for LangGraph (Python and TypeScript), with integrations for frameworks like CrewAI and AutoGen on the roadmap.

  • Portability: Swap models or tools through configuration updates without re-architecting your entire agent.

Please refer to our official documentation for additional details - Atlas Agent Engine.

5. Can I use Atlas Memory without using Atlas Agent Engine?

Yes. You can use the Atlas Agent Engine Memory SDK directly from your own application or service, without deploying an agent through Atlas Agent Engine. The SDK connects to the hosted Memory API and uses a project service account for authentication.

If you are running an agent on Atlas Agent Engine, you can also enable integrated memory in your agent configuration. This lets the agent use memory automatically while it handles platform requests.

Please refer to our official documentation for additional details: Use the Standalone Memory Service

6. Do I need a MongoDB Atlas account? What type of cluster do I need?

Yes. A MongoDB Atlas account and a running Atlas cluster are required to deploy an agent to Atlas Agent Engine. Atlas stores agent data such as workflow checkpoints and, when memory is enabled, memory data and search indexes.

Use one of the following cluster types:

  • Flex

  • Dedicated cluster, M10 or higher

Free clusters not supported for Atlas Agent Engine deployments.

If you are new to Atlas, the setup workflow can help you select or create the required Atlas resources.

Please refer to our official documentation for additional details: Set up Atlas resources.

Getting Started & Development

1. Where should I start with Atlas Agent Engine?

Start with the official Quick Start Guide. It walks through prerequisites, CLI setup, running your first agent locally, and deploying it to the cloud.

2. Where can I find the full documentation?

The documentation portal contains complete guides, conceptual deep-dives, and platform specifications:

3. How do I download, install, or update the CLI?

Install the agentengine CLI and authenticate with your MongoDB Atlas account following the installation guide:

4. I already have an existing agent. Can I migrate it to Atlas Agent Engine?

Yes. Follow our migration guide to adapt existing LangGraph workflows, configure your agent.yaml manifest, and containerize your dependencies for deployment:

5. How do I create a new agent project from scratch?

Use the CLI project scaffolding workflow to generate a structured project with sample graphs and configuration:

6. Where can I find configuration syntax and agent.yaml specifications?

The Agent Contract Reference is the authoritative specification for all agent.yaml fields, schemas, runtime expectations, and scaling parameters:

Administration, Security & Tools

1. How do I manage access for my team?

Access is structured hierarchically across Organizations, Projects, and Workspaces, leveraging Atlas Role-Based Access Control (RBAC):

2. How do API keys and service accounts work?

To automate deployments from CI/CD pipelines or invoke agents programmatically from your applications, generate scoped service accounts and API keys:

3. Where can I find the Platform REST API reference?

Consult the API reference for endpoints covering agent invocation, execution status, streaming responses (SSE), and workspace management:

4. How are network egress, security, and governance handled?

Atlas Agent Engine runs workloads in isolated sandboxes . Outbound network traffic (to third-party APIs or external tools) is restricted by default and managed through explicit egress policies:

5. Where can I find guidance on Memory, Multi-Agent communication, and Remote MCP?

Billing & Pricing

1. What is the cost of Atlas Agent Engine during Public Preview?

During Public Preview, Atlas Agent Engine uses a pay-as-you-go, consumption-based billing model, that draws down directly from your existing MongoDB Atlas credits - no separate contract or upfront license is required. Please refer to this resource for additional details.

Charges are determined by two primary platform meters:

  • Agent Runtime: The compute time your agents spend actively executing requests (metered per vCPU-second).

  • Agent Memory Operations: The volume of data documents stored in and retrieved from long-term memory.

Note: Underlying Atlas database usage (such as clusters, vector search storage, Voyage AI services etc.,), third-party LLM provider API token fees, and network data transfer are billed at their respective standard rates.

Support & Troubleshooting

1. What can Chat Support assist with?

In-app chat support is available for all Atlas users to help with:

  • Navigating public documentation, product concepts, and Quick Start guides

  • Verifying known Atlas Agent Engine service issues reported on status.mongodb.com.

  • Collecting non-sensitive triage details

Chat support does not provide live custom application debugging, proprietary code reviews, prompt optimization, or log forensics.

2. What information should I provide when reporting an issue?

To help our team resolve your inquiry efficiently, please share:

  • Stage of failure: Did the issue occur during CLI login, local development (agentengine dev up), image build, cloud deployment, or invocation?

  • Error message: The exact visible, non-sensitive error text.

  • Guide followed: The specific documentation page, sample template, or command executed.

⚠️ Security Reminder: Never paste API keys, passwords, database connection strings, or proprietary customer data into support chat.

3. How do I open a formal Support Case for hands-on investigation?

Did this answer your question?