How to build a MultiAgent?

flowchart TD 
  C["Agents"] --> n2["Agent"] & D["FlowAgent"] & n1["MultiAgent"]
n4["DevSpace"] --> C & n3["Workflows"]
n3 --> n5["Workflow"]
n2@{ shape: rect}
n1@{ shape: rect}
n4@{ shape: rect}
n3@{ shape: rect}
style n2 color:#000000
style D color:#000000
style n1 color:#D50000
style n3 color:#000000
style n5 color:#000000

MultiAgent is a robust solution designed for orchestrating multiple specialized agents to collaboratively solve complex, multi-step tasks. Unlike a single-agent setup, a MultiAgent system leverages the collective strengths of diverse agents, coordinated through a centralized Planner and enhanced by a modular configuration.

This guide provides a deep dive into creating, configuring, debugging, and publishing a MultiAgent system, highlighting best practices for high-impact B2B deployment.

Understanding the MultiAgent Architecture

The Pivotal Role of the Planner

At the core of any MultiAgent system is the Planner. It acts as the "brain," responsible for interpreting user requirements, planning the execution path, and conducting rigorous quality reviews.

Key capabilities of the Planner include:

• Dynamic Task Adjustment: Real-time modification of tasks as conditions evolve during execution.
• User Takeover: Enables human intervention to provide guidance or resolve bottlenecks when automated progress stalls.
• Retry Limits: Configure maximum retry counts for individual tasks to optimize performance and control token consumption.

The Planner’s intelligence can be further expanded by integrating long-term memory, advanced tools, knowledge bases, and external databases.

Assigning Specialized Sub-Agents

By clicking the “Add Node” button, you can access a diverse library of agent types to assist the Planner:

• Preset Agents: System-configured and ready for rapid deployment (e.g., Insight Writers, Online Search tools).
• Blank AI Agents: Fully customizable agents requiring manual prompting to fit specific, proprietary business logic.

Once selected, these agents are placed on the visual canvas and connected to the Planner for seamless task allocation.

Exploring Specialized Agent Types and Their Functions

A successful MultiAgent system relies on specialized nodes. Here are the key agent types available:

• Computer Use Agent: Operates within a sandboxed Linux environment. It can use applications, access file systems, and execute system commands.
• Browser Use Agent: Designed for web-based automation within a sandboxed browser, ideal for tasks not requiring LLM-based image recognition.
• Coder IDE Agent: A specialized environment for code generation, API execution, and web development within a secure CLI.
• Online Search Agent: Focuses on real-time web content retrieval, making it the perfect choice for market monitoring and information gathering.

Pro Tip: Providing clear, accurate descriptions of each agent's abilities is crucial. The Planner uses these descriptions to decide which agent is best suited for a specific sub-task.

Executing and Managing MultiAgent Workflows

To illustrate the power of orchestration, consider this task: “Write a feature article about the hottest pizza spots in New York in 2025.”

1. Task Initiation: Submit the request. The Planner assesses whether it has enough information to proceed or if it needs to ask the user for clarification.
2. Plan Generation: The Planner generates a structured task list. Once you approve the plan, execution begins.
3. Collaborative Execution:
   • The Online Search Agent crawls the web for the latest 2025 food trends.
   • The Planner monitors the search results and adjusts the next steps based on the data found.
4. Final Synthesis: The Insight Writer Agent summarizes the findings into a cohesive article. Users can then interact further to refine or expand the content.

Key Takeaways for Multi-Agent Success

• Centralized Intelligence: Treat the Planner as the primary coordinator for task monitoring and adaptive control.
• Modular Flexibility: Use Preset Agents for speed and Blank Agents for highly specific business scenarios.
• Supervised Execution: Utilize the task approval and human takeover features to ensure high-quality outputs in enterprise environments.

Conclusion

Building a MultiAgent system enables you to orchestrate complex workflows with unparalleled efficiency. By combining specialized roles—like searchers, coders, and writers—under a centralized Planner, you create a scalable, maintainable AI solution for any business challenge.

Learn more about MultiAgent deployment: https://www.gptbots.ai/docs/tutorial/bot/multi-agent/overview