Concepts

Managed Agents

Claude Managed Agentsis the Anthropic API that takes care of hosting, running, and the lifecycle of your agents. It's the technical foundation of Forgent: we generate the blueprint, Anthropic runs it.

The problem it solves

Before Managed Agents, running an agent in production required standing up an infrastructure:

  • A server (VM, container, lambda)
  • An orchestrator (cron, queue, trigger)
  • A logging and monitoring system
  • Error handling and retry logic
  • A security layer for the tools (who has access to what)

For a POC, that's several days of setup before you even write the first line of business logic.

What Managed Agents handles

You send a blueprint to Anthropic, Anthropic returns an agent_id. From there, Anthropic takes care of everything:

  • Hosting — the agent lives on their infra, you have no server to manage.
  • Scheduling — if your agent runs on a trigger (cron, webhook, event), Anthropic calls the agent at the right moment.
  • Tool sandboxing— each tool (bash, files, web fetch) runs in an isolated environment. Your data doesn't leak to other Anthropic customers.
  • Sessions and resumes — if an agent crashes in the middle of a long task, its session is saved and can be resumed.
  • Logs and monitoring — accessible via the Anthropic console.

What stays on your side

  • Nothing to pay or plug in — every agent runs for free, funded by the platform. There is no API key and no usage bill.
  • The MCP integration— connecting your Notion, Gmail, GitHub, etc. to your Anthropic account. Forgent doesn't do it for you (Anthropic secured it for good reasons).
  • The business validation— Anthropic can run the agent, but it can't verify that its outputs are correct for your business.

The relationship between Forgent and Managed Agents

This matters: Forgent is not Anthropic. We're a layer on top.

  • Anthropic provides the model (Claude), the API (Managed Agents), and the hosting.
  • Forgent provides the no-code interface that generates a correct blueprint and pushes it to Managed Agents.

Concretely: if Forgent disappears tomorrow, your agents keep running at Anthropic. You retain access via their console. We don't hold you hostage.

Beta statusManaged Agents is officially in beta at the time we're writing this doc (May 2026). Anthropic regularly adds new tools, models, and MCPs. Forgent tracks these changes — check the changelog for updates.

Going further

The official Anthropic documentation covers the Managed Agents API in detail (endpoints, request format, rate limits). You don't need it to use Forgent, but it's useful if you want to hit the API directly or understand what happens under the hood.