# Call other agents from one agent

Build an answer agent that calls three published agents inside its own turn and writes one incident note from what they return.

A manager agent receives the work, decides during the turn which other agent handles each part, and puts the result together. On Katonic it is an answer agent with other agents attached in **Tools**. This tutorial builds a network operations manager that takes an alarm, calls a diagnosis agent, a customer impact agent and a ticketing agent, and writes one incident note. It is the eighth form in [Which builder for which agent](/agents/choose-a-builder).

**Who can use it:** the **Builder** role or higher, in a tenant licensed for agents.

## Prerequisites

- The three agents it calls, already published: a fault diagnosis agent, a customer impact agent and a ticketing agent. Only published agents are listed. If they do not exist yet, build each one first as an answering or acting agent; see [Build an agent that answers from your documents](/agents/forms/answer-from-your-documents) and [Build an agent that acts on a request](/agents/forms/act-on-a-request). To try this tutorial quickly, any two published agents will do.
- A team you belong to, for the **Owner team**, and a guardrail profile. [Before you build a form of agent](/agents/forms/before-you-build) says who provides each.

## Build the agent

  
**Step 1. Start a blank draft**

In Studio, open **My Agents** and select **New agent**. Select **Answer & assist**, then **Answer questions**. Pick the **Blank** card and select **Continue to build**.

  
**Step 2. Name it**

On **Identity**, type the **Display name** `Network Operations Manager`, a one-sentence **Description** and pick the **Owner team**.

  
**Step 3. Say which agent does what**

Open **Behavior** and paste the instructions from the next section into the system prompt, the standing instruction the agent follows on every turn. The model decides when to call each agent, so the instructions name each one, say what to hand it and what to do when it returns nothing. Use the names your own agents carry.
    ![The agent builder on Behavior for the Network Operations Manager draft: the System prompt card with instructions that name three agents and say what to hand each one, and the Model tier card below.](https://docs.katonic.ai/img/agents/forms/call-other-agents-from-one-agent/forms-manager-behavior.png)
*The instructions name each agent this one may call, what to hand it and what to do when it returns nothing.*

  
**Step 4. Attach the agents**

Open **Tools** and scroll to the **Agents** card. It lists the published agents with their kind and description. Switch on the three this agent should call. You should now see the card count three attached. An attached agent answers inside this agent's turn.
    ![The Tools section scrolled to the Agents card for a draft named Network Operations Manager: the count of agents attached, the line saying these are agents this one can call as tools, and one row per published agent with its kind, its description and a switch, three of them on.](https://docs.katonic.ai/img/agents/forms/call-other-agents-from-one-agent/forms-manager-agents.png)
*The Agents card lists the published agents this agent can call inside its own turn. Switch on the ones it should use.*

  
**Step 5. Keep the execution strategy**

Open **Advanced**. Leave the **Execution strategy** on **Tool-calling**, which lets the model decide when to call a tool and when to answer.

  
**Step 6. Pick a guardrail profile**

Open **Guardrails** and pick a profile, a named set of safety checks that run on every turn.

## The agent's instructions

```text
You receive a network alarm and write one incident note.

Call the Fault Diagnosis Agent with the alarm to get the likely cause. Call the Customer Impact Agent with the affected site to get who is affected. Call the Ticketing Agent to open a ticket once you have both.

Then write the incident note: the alarm, the likely cause, who is affected, the ticket reference and what happens next. Say which agent each part came from.

Call each agent once. If an agent fails or returns nothing, say so in the note. Do not invent a cause, a customer count or a ticket reference.
```

## Test and publish

  
**Step 1. Save the alarm as a test prompt**

Open **Test** and add a prompt with the plus button:

    ```text
    ALARM 14:02 site NTH-04: loss of signal on the fibre uplink, 3 sectors down.
    ```
    ![The Test tab for the Network Operations Manager draft: one saved test prompt holding a site alarm, the sandbox message box with Run, and the empty Run trace on the right.](https://docs.katonic.ai/img/agents/forms/call-other-agents-from-one-agent/forms-manager-test.png)
*Save the alarm as a test prompt. A run lists its steps in the Run trace, the calls to other agents among them.*

  
**Step 2. Run it and read the trace**

Select **Run**. The **Run trace** on the right lists the steps of the real run, tool calls among them. A call to an attached agent appears there as a tool call.

  
**Step 3. Publish**

Select **Publish**, fix any blocker, choose who can use the agent and tick **Work Room** under **Where it runs**. The screen that follows says whether the version is live, in a lane or not live yet; see [Before you build a form of agent](/agents/forms/before-you-build).

## Verify

Run the alarm above. The trace shows three calls, one for each attached agent, and no agent called twice. The incident note has five parts: the alarm, the likely cause, who is affected, a ticket reference and what happens next. Each part names the agent it came from, and the cause, the customers and the reference match what those agents return when you ask them directly.

Then detach one agent and run again. The note says that part is missing; it does not make one up.

## Troubleshooting

- **The agent answers from its own knowledge and calls nobody.** Read the trace. If it shows no calls, name each agent and its job more plainly in the instructions, and check that the three switches on the **Agents** card are on.
- **An agent is called and returns nothing.** Open that agent and run the same input on its own **Test** tab. A tool that an attached agent calls is checked against tool policies in that agent's own run, so a write it attempts can be parked for approval there; see [Before you build a form of agent](/agents/forms/before-you-build).
- **Couldn't load your published agents.** The list could not be read. Reload the builder and open **Tools** again.
- **An agent you expected is not listed.** Only published agents appear, and the agent you are building is left out of its own list.
- **The turn is slow or costly.** Every call adds time and cost. Attach only the agents the job needs.
- **You wanted one agent to take over the conversation.** That is a coordinator, which hands the whole turn to one specialist; see the next form.

## Related

- [Which builder for which agent](/agents/choose-a-builder)
- [Build an answer agent](/agents/build-an-answer-agent)
- [Coordinate a team](/agents/coordinate-a-team)
- Next form: [Route requests to specialists](/agents/forms/route-requests-to-specialists)

Source: https://docs.katonic.ai/agents/forms/call-other-agents-from-one-agent
