Skip to main content
Three building blocks let an LLM do more than answer a single prompt: Agents, Sub-agents, and Tools. They sound similar but play different roles. This page explains each and shows how they connect.

​
At a glance

​
Agent

An Agent is the everyday LLM step. You give it a prompt and a model; it runs as part of the workflow and produces an output the next node consumes. Pick this when the node should always run as part of the flow. Create one with an LLM node set to Agent mode (the default).

​
Sub-agent

A Sub-agent is a specialized helper that other agents can call on demand — like a junior expert sitting on the bench. It does not run on its own. A main agent reads its tool description and decides whether the current task needs it. Use sub-agents to split a big problem into focused roles — a researcher, a writer, a validator — that the main agent orchestrates. Each sub-agent is an LLM node set to Sub-agent mode, which adds two settings:
  • Tool name — a short identifier the parent agent uses to call it (e.g. research_assistant).
  • Tool description — tells the parent agent when to call it. Be specific about the task it handles.

​
Tool

A Tool is a ready-made capability — like web search or consulting another model — that an agent can invoke while it works. You add tools to an agent with a Tool node in Agent tool mode: pick the tools you want, and the node exposes a single tool-list output you wire into the agent. A Tool node can also run in Standalone mode, where a single tool runs directly as its own step instead of being handed to an agent.

​
Wiring it together

Both Tool nodes and Sub-agents feed an Agent through the same place: the Agent’s Tools input handle (type tool-list, which accepts multiple connections).
1

Build the main Agent

Add an LLM node in Agent mode. It has a Tools input handle on its left.
2

Add tools

Add a Tool node in Agent tool mode, select your tools, and connect its tool-list output to the Agent’s Tools input.
3

Add sub-agents

For each specialist, add an LLM node in Sub-agent mode and connect its Tool out handle into the same Tools input. The main agent now sees each sub-agent as one of its tools.
Because the Tools input accepts multiple connections, one Agent can combine several Tool nodes and several sub-agents at once.

​
Limits

Sub-agents can call other sub-agents, but the chain is capped at a depth of 5. Tool names must be unique within a single parent agent, and sub-agent call chains can’t form a cycle.

​
Next steps

LLM node

Agent and Sub-agent modes in detail.

Tool node

Agent-tool bundles and standalone tools.