Skip to main content

Sub Agent Node

The Sub Agent node runs an AI agent inside your workflow as a background processing step. Unlike the Forward to Agent node (which transfers a live call), the Sub Agent runs behind the scenes and returns structured results for the workflow to use.

How It Works

When the workflow reaches this node, it spins up a separate AI agent that processes data, performs research, or completes tasks independently. The sub agent can use tools, access external services, and produce output that feeds into subsequent workflow nodes.

The caller does not interact with the sub agent directly. It operates in the background while the workflow continues or waits for results.

Capabilities

The sub agent has access to powerful tools:

  • MCP tools --- connect to external services and APIs through the Model Context Protocol.
  • Web search --- look up real-time information from the internet.
  • Structured output --- return data in a defined format that other nodes can use.

Sub Agent vs. Forward to Agent

Sub AgentForward to Agent
InteractionBackground processing, no caller interactionTransfers the live call to another agent
PurposeData processing, research, complex reasoningHanding off a conversation to a specialized agent
Caller awarenessCaller does not know it is runningCaller speaks directly with the new agent
OutputReturns structured data to the workflowTakes over the conversation

Common Use Cases

  • Data research --- Look up a company or contact using web search and return a summary for the sales rep.
  • Complex calculations --- Process order details, apply discounts, and calculate totals.
  • Multi-step reasoning --- Analyze conversation data, cross-reference with policies, and determine eligibility.
  • External API calls --- Use MCP tools to query CRMs, ERPs, or other business systems and bring data back into the workflow.

Tips

  • Define clear, specific instructions for the sub agent so it knows exactly what to accomplish.
  • Keep sub agent tasks focused. If you need multiple types of processing, use multiple sub agent nodes.
  • Test sub agent outputs to make sure the returned data format works with your downstream nodes.

Next Steps