Multi-Agent System
Multi-Agent System
Multi-Agent System
The Multi-Agent System allows you to create multiple conversation flows within a single agent. Instead of building one massive flow or separate agents for different scenarios, you can organize specialized sub-agents inside one unified agent.
Use the Trigger Subflow node to hand off from your main conversation to a specialized sub-agent. When the sub-agent finishes, the conversation automatically continues from the Trigger Subflow node—right where it left off.
Key benefit: Keep your main flow clean and organized while handling complex scenarios in dedicated sub-agents.
Benefits:
Example: Create a “Pricing Questions” sub-agent. Whenever a user asks about pricing (from anywhere in your main flow), trigger that sub-agent. It answers their questions, then returns control to your main flow exactly where it left off.
The Trigger Subflow node allows your main flow to hand off to a specialized sub-agent, then automatically return.
In your agent settings, create separate sub-agents for different scenarios:
Common sub-agents:
Each sub-agent is a complete conversation flow with its own nodes (Conversation, Branch, Message, etc.).
Add Trigger Subflow nodes in your main conversation:
Example:
Variables automatically flow between your main flow and sub-agents.
Main Flow → Sub-Agent:
Sub-Agent → Main Flow:
You can view all available variables (from both your main flow and sub-agents) in the variable drawer. This makes it easy to see which variables are accessible at any point in your flow and helps you reference them correctly in nodes.
Both Trigger Subflow and Jump To nodes allow you to redirect your flow, but they serve different purposes. Understanding the difference helps you choose the right tool for your use case.
Key difference: Trigger Subflow hands off to a separate sub-agent and automatically returns when done. Jump To redirects to a node within the same flow and doesn’t return—it’s a permanent redirect.
For more details on when and how to use Jump To nodes, see the Jump To Node documentation.
There’s no hard limit, but we recommend keeping it manageable. Beyond that, consider whether you need multiple separate agents.
Yes, but avoid going more than 2 levels deep. Deep nesting makes flows hard to debug and maintain.
It automatically returns to the Trigger Subflow node in your main flow, and the conversation continues with the next node. The user experiences a seamless transition.
Trigger Subflow hands off to a separate sub-agent and automatically returns. Jump To redirects to a node within the same flow and doesn’t return.
Use Trigger Subflow for reusable logic. Use Jump To for loops within the same flow.
Yes! Variables collected in your main flow are accessible in sub-agents, and sub-agents can update variables that the main flow will see when it returns.