Skip to the operations

AI

Claude Managed Agents in a decision flow.

4 Claude Managed Agents operations a decision flow can call directly, with the credentials your own contract issued. The response is data the rest of the flow reads, branches on, and keeps in the trace.

Category
AI
Type
Integration
Authentication
API key
Test environment
Production host only

Who they are

Hosted Claude agents that work in their own sandbox.

Claude Managed Agents runs a saved Claude agent for you: Anthropic runs the agent loop and gives each session its own sandbox, where the agent reads and writes files, runs commands and works on mounted GitHub repositories. A session can have a hard spend cap in US dollars, and it reports its cost at list price as it runs. The API is in beta.

What a flow can call

4 operations, each one a step you can place on the canvas.

  1. POST/v1/sessions

    Start a session

    Starts a session of a saved agent in a saved environment and sends the first user message in the same call.

  2. POST/v1/sessions/$data.managed_session_id/events

    Send a message to a session

    Sends a user.message event to a session. Use it to continue an idle session with a new instruction.

  3. GET/v1/sessions/$data.managed_session_id

    Get a session

    Reads the status and the cost of a session. Set data.managed_session_id to the session ID.

  4. GET/v1/sessions/$data.managed_session_id/events?order=desc

    List the events of a session

    Reads the events of a session, newest first. Set data.managed_session_id to the session ID.

Where it sits in the decision

AI calls have a natural place in a flow.

A model call is for the judgment a table cannot express — reading a document, summarising a file, weighing a narrative. Because the output is not deterministic, a flow constrains it with a schema and keeps the prompt and the response in the trace, so the reasoning behind a referral stays as inspectable as a rule.

Whatever Claude Managed Agents returns is part of the run, so it is part of the record. When someone asks months later why an applicant was declined, the answer cites what came back at the time rather than re-fetching from a service whose answer has since changed.

  • 01Add Claude Managed Agents as a connection authenticating with an API key.
  • 02Claude Managed Agents has one host for both environments, so guard test runs with your own credentials and limits.
  • 03Place a Connection node and pick an operation — “Start a session” is usually the first one a flow needs.
  • 04Map the response into the fields your rules read, then test the whole path before it carries live traffic.

Common questions

Using Claude Managed Agents in a flow.

How do I connect Claude Managed Agents to a decision flow?

Add Claude Managed Agents as a connection in your workspace authenticating with an API key, with the credentials your own contract issued — ArboRule calls the provider as you, and never holds a contract on your behalf. Once the connection exists, any flow in the workspace can place a Connection node and choose one of its operations. The credentials live on the connection, not in the flow, so a policy owner can use Claude Managed Agents in a decision without ever seeing the secret.

Can I test Claude Managed Agents without touching production?

Claude Managed Agents exposes one host for both environments, so there is no separate sandbox to point at. Test runs still execute in Sandbox and are recorded separately in decision history, but the call goes to the same place as production — so guard it with your own test credentials, rate limits, or data.

What can a flow call on Claude Managed Agents?

4 operations, including “Start a session”, “Send a message to a session”, “Get a session”. Each one is a step you place on the canvas and map into the fields your rules read, and most flows start with “Start a session”. The list comes from the same manifest the engine uses to make the call, so this page cannot describe an operation the product does not have.

Where in a decision should Claude Managed Agents be called?

A model call is for the judgment a table cannot express — reading a document, summarising a file, weighing a narrative. Because the output is not deterministic, a flow constrains it with a schema and keeps the prompt and the response in the trace, so the reasoning behind a referral stays as inspectable as a rule.

Also in AI

Ready when you are

Wire Claude Managed Agents into a real decision.

Build the flow in Sandbox, connect your account, and watch the decision pull what it needs before it answers.

Read the docs