Workflow Nodes

Agent Call

Delegate a step to a specific agent within a workflow. The workflow pauses, the agent processes the input, and the result is injected back into the execution context.

How It Works

The Agent Call node references an existing agent by ID. When the workflow reaches this node, the executor fetches the agent configuration, runs the LLM with the provided input, and returns the structured output. If the agent is not found, the step falls back to the workflow's default agent.

Specialized Agents

Assign different agents to different steps. Each agent can have its own provider, model, and system instructions.

Context Isolation

The called agent receives only the input payload you specify. Its output is returned to the workflow context.

Structured Output

Agent responses are validated and stored as step results, just like any other step type.

Agent Call Configuration

agent-call-node.json

{
  "type": "agent_call",
  "agentId": "agent_001",
  "input": "Summarize the following research: {{steps.s1.output}}",
  "waitForResponse": true
}

Notes

• The agent must be owned by the same user as the workflow.
• If waitForResponse is true, the workflow waits for the agent to finish before continuing.
• Agent call steps support retry policies and timeouts like any other step.