Skip to content

Call Agent Tool

The Call Agent tool allows Agents to call other Agents for task assistance, enabling division of labor and collaboration.

When to Use

  • Tasks involve multiple specialized domains
  • Want to reuse capabilities of already-created Agents
  • Need layered handling of complex processes

Use Case Examples

Main Agent Called Agent Scenario
Customer Service Hub Technical Support Agent, Billing Agent Route based on question type
Sales Assistant Product Specs Agent, Quotation Agent Complex inquiries
Project Management Scheduling Agent, Resource Query Agent Multi-faceted coordination

Prerequisites

Before adding a Call Agent tool, you need to:

  1. Create Target Agent: Create at least one Agent in the workspace
  2. Publish Target Agent: Ensure the target Agent has published at least one version

Must Have Published Version

  • If the target Agent has never published any version, even if you check "Always choose latest published version", the tool will be automatically skipped during execution.
  • Please ensure the target Agent has published at least one version.

Setup Steps

Step 1: Add Call Agent Tool

In the Tools section, click "Add Tool" to open the tool selector.

Step 2: Select Target Agent

In the "Agent Tools" group of the tool selector, select the Agent you want to call. The system will automatically filter out already selected Agents to avoid duplicate selection.

Cannot Add When No Agents Available

  • If there are no created and published Agents in the workspace, the "Agent Tools" option in the tool selector will appear grayed out with a note "All agent tools selected", and you cannot add a Call Agent tool at this time.
  • Please create and publish at least one Agent first.

Step 3: Choose Version

When adding a Call Agent tool, the system defaults to selecting "Latest Published Version".

You can choose:

Option Description Suitable Scenario
Latest Published Version (Default) Automatically use the target Agent's latest published version Rapid iteration, testing phase
Fixed Version Specify a particular version, won't auto-update Stable operation, production environment

Defaults to Latest Published Version

  • The system defaults to selecting "Latest Published Version".
  • If the target Agent has never published a version, please publish at least one version first, or switch to "Fixed Version" and select an existing version.

Severity of Unpublished Versions

  • If you choose "Latest Published Version" but the target Agent has never published any version, the Call Agent tool will be automatically skipped during execution, and the Agent cannot call that tool. The system will display a yellow warning icon in the tool settings to indicate this issue.
  • Strongly recommended: Before configuring a Call Agent tool, ensure the target Agent has published at least one version, or switch to "Fixed Version".

Step 4: Set "When to Use"

Tell the Agent when to call this Agent.

Examples:

When users ask technical questions, troubleshooting, or API usage,
call the Technical Support Agent for assistance.
Pass the user's original question to it.

When needing to calculate quotes or query prices, call the Quotation Agent.
Pass necessary specification and quantity information.

Step 5: Confirm Addition

Click confirm, and the tool will appear in the Tools list.

Version Selection Recommendations

Use "Latest Published Version"

  • During development and testing phases, but you want improvements to take effect automatically
  • Want to automatically use the target Agent's latest published version

Default Option

When adding a Call Agent tool, the system defaults to selecting "Latest Published Version". If the target Agent hasn't published a version yet, please publish first or switch to "Fixed Version".

Use "Fixed Version"

  • Production services
  • Need stable and predictable behavior
  • Target Agent has risk of major changes

Version Number Display

  • When selecting "Fixed Version", the system will display all available version numbers of the target Agent (v1, v2, v3...), making it easy to choose the appropriate version.

Execution Method

When an Agent needs to call multiple Agents:

  • Sequential execution: One completes before calling the next
  • Result integration: Main Agent integrates all call results to reply to the user

Tool List Display

Added Call Agent tools are displayed in the "Agent Tools" group under the "Tools" section, each tool showing:

  • Target Agent's name (e.g., "Collect Name and Phone")
  • Version label:
  • If "Latest Published Version" is selected: Shows "Use Published Version" label
  • If "Fixed Version" is selected: Shows version number (e.g., "v3")
  • If target Agent hasn't published a version: Shows yellow warning icon
  • Can be expanded to view detailed settings (including Agent description, version selection, "When to Use")
  • Can be removed by clicking the delete button

Quantity Limit

  • Maximum 5 Call Agent tools can be added
  • This limit ensures the Agent can effectively select and use tools

Call Depth Limitation

The system limits to a maximum of one level of call depth. Even if you configure multiple Call Agent tools, called Agents cannot call other Agents again.

Best Practices

Clear Division of Labor

Each called Agent should have a clear specialty:

Technical Support Agent: Handle API, SDK, troubleshooting
Billing Agent: Handle billing, subscriptions, invoice issues
Product Agent: Handle feature introductions, usage tutorials

Use in Conjunction with Agent's "Core Instruction"

Explain in the main Agent's "Core Instruction":

## Collaboration Rules
- Refer technical questions to Technical Support Agent
- Refer billing questions to Billing Agent
- After integrating each Agent's responses, answer user with unified tone
- If question spans domains, determine primary domain first before calling

Pass Sufficient Context

Explain in When to Use what information to pass:

When calling, pass:
- User's original question
- Relevant conversation history
- Known user information (if any)

Testing Calls

After adding, test if calls work properly:

  1. Ask a question that should trigger a call
  2. Observe if the target Agent was called
  3. Check if the answer integrates results from the called Agent

Common Questions

Can a called Agent call other Agents?

No. The system limits to a maximum of one level of call depth:

  • ✅ Allowed: Agent A → Agent B (one level of calling)
  • ❌ Not allowed: Agent A → Agent B → Agent C (nested multi-level calls)

When Agent B is called, the system automatically disables its Call Agent tools, preventing it from calling other Agents again. This limitation is to:

  • Prevent infinite loops and excessive complexity
  • Reduce debugging difficulty
  • Control resource consumption
What happens if a call fails?

The main Agent will receive an error message and typically inform the user that the request cannot be completed. It's recommended to add error handling instructions in the Instruction.

Can users choose which Agent to answer?

Yes, you can design this flow in the main Agent's "Core Instruction", for example: "First ask the user about the question type, then decide which specialized Agent to call".

What happens if the called Agent is deleted?

If the target Agent in the workspace is deleted, the system will automatically remove the corresponding Call Agent tool from the Agent's tool configuration and display a notification message (Toast).

It's recommended to regularly check the Agent's tool configuration to ensure all Call Agent tools are functioning properly. If you find a tool has been automatically removed, recreate and publish the target Agent, then add the tool again.

Why can't I select a certain Agent?

Possible reasons:

  1. The Agent doesn't belong to the workspace where the current Agent is located
  2. The Agent has already been selected
  3. The Agent has been deleted from the workspace

Please check the Agent's status and type, or contact an administrator to confirm workspace permissions.

Next Steps