Creating a Multi-Agent System for Request Processing Using AI Agent

Hello, community!

I am working on creating a system similar to the one implemented in the n8n service, and I would like to get advice on developing a module that will include several multi-agents.

System Description:

My goal is to create a manager agent that will accept requests from users and determine which functions and tools need to be used for processing them. For example, these could include:

  • Performing calculations using a calculator

  • Making API requests to retrieve data

  • Sending emails or creating documents using third-party APIs

What I Want to Implement:

  1. Manager Agent: The main agent that will receive requests from users and route them to the appropriate sub-agents for specific tasks.

  2. Sub-Agents: These agents will perform specific functions, such as:

  • Processing text requests

  • Performing mathematical calculations

  • Interacting with external APIs to retrieve or send data

  1. Interaction Structure: I want to understand how to organize the interaction between the manager agent and the sub-agents. How can I properly pass requests and data between them to ensure efficient task execution?

Example Use Case:

I envision that a user might ask a question like, “Create a document with the sales report.” The manager agent should:

  • Determine that a sub-agent for document creation is needed to fulfill this request.

  • Gather the necessary data using other sub-agents (e.g., to retrieve sales information via an API).

  • Return the final response to the user, such as sending the document via email.

Questions:

  • What is the best way to organize the architecture of such a system?

  • What tools and technologies would you recommend for implementing a multi-agent system like this?

  • Are there any examples or resources that could help me in the development?

I would appreciate any advice and ideas!

Thank you!

Wowza @Artem_Larin; you’re not asking for much.

Despite my skills and wanting something similar for an internal chatbot of ours, the initial query processing is very open-ended and likening to opening up pandoras box. As such, I’m awaiting CrewAI’s cloud servicing via an API instead of building out the system.

This way, I stick with what I know well, my business operations and team member needs, and not attempting to open myself up to a time suck that is better served by off the shelf options.

CrewAI has some decent alternatives, including open source that might interest you to let Make be the controller, and other the processers. :four_leaf_clover: