Agent Call node
The Agent Call node invokes another agent from within your workflow. This lets you compose complex behaviors by delegating tasks to specialized agents.
In the palette it appears under Actions with the description: “Call another agent inside this workflow.”
When to use
- Break complex workflows into smaller, reusable agents
- Call a specialized agent (e.g., a classification agent, a summarization agent)
- Compose multi-step pipelines where each step is its own agent
Configuration
Click the Agent Call node to open its configuration panel.
| Setting | Type | Description |
|---|---|---|
| Select Agent | Search text field | Search and select which agent in the project to call |
| Input Mapping | Add button → variable mapper | Map variables from this workflow to the target agent’s input schema |
| Output Mapping | Add button → variable mapper | Map the called agent’s output to workflow variables |
| Result Variable | Text field (default: agent_result) | Name of the variable that holds the full response |
| Timeout | Number input (default: 60 seconds) | Maximum time to wait for the called agent to respond |
Retry Configuration
| Setting | Type | Description |
|---|---|---|
| Max Attempts | Number input (default: 3) | Number of retry attempts on failure |
| Backoff | Number input (default: 1000 ms) | Delay between retries in milliseconds |
ℹ️
The called agent must be in the same project. Both Task Agents and AI Agents can be called.
Output
The output variable contains the called agent’s response. The structure depends on the target agent’s output schema.
Common patterns
- Classifier → Specialist — Use one agent to classify input, then route (via Switch) to specialized agents via Agent Call.
- Pipeline composition — Chain multiple agents sequentially, passing output of one as input to the next.