Questions

Docs · Questions

LOPsUsage

How do I use the LOPs `Agent Orchestrator` with multiple agents?

Problem

The Agent Orchestrator is a powerful LOP for creating complex, multi-agent workflows, but its structure can be confusing. Users are often unsure how to define tasks and prompts for the orchestrator and its sub-agents.

The Core Concept: A Manager for Your Agents

Think of the Agent Orchestrator as a “manager” agent. Its job is not to do the work itself, but to delegate tasks to a team of “worker” agents.

  • Agent Orchestrator: The manager. It holds the main goal or task.
  • Sub-Agents: The workers. Each has a specific skill or purpose (e.g., one agent for writing code, another for analyzing text).
  • Validation Agent: The “brain” of the orchestrator. It examines the main task and decides which sub-agent is the best tool for the job.

How to Structure Your Prompts

  1. Define the Main Goal: The high-level goal or task should be placed in the Task DAT parameter of the Agent Orchestrator.
  2. Define the Worker’s Skills: The specific capabilities, instructions, and personality for each sub-agent should be placed in their respective System Message parameters.

How to Debug the Orchestrator

The key to understanding the orchestrator is to see how the “brain” (the validation agent) perceives its “workers” (the sub-agents). You can do this with the Tool Debugger LOP.

  1. Place a Tool Debugger LOP in your network.
  2. Connect the Agent Orchestrator to the input of the Tool Debugger.
  3. The Tool Debugger will now show you exactly how the sub-agents are being presented as “tools” to the validation agent. This will help you understand if your agent descriptions are clear and if the validation agent is able to correctly choose the right agent for the task.