N8n agent team

Questions
1

IMG_0593
IMG_05931920Ă—886 62 KB

n8n currently have sub agent, is there any way to get an agent team that talks to each other ?

2

Hi @Mohammed_Rifad
You can simulate an agent team by configuring other AI Agent workflows as Tools for a main “Supervisor” Agent, effectively allowing them to “talk” by passing data back and forth.

check these helpfull links :
Multi-agent systems: Frameworks & step-by-step tutorial
n8n AI agent builder

3

Yes @Mohammed_Rifad You can create multi-agent systems in n8n where agents communicate with each other. Here is an approache:

Create a “manager” agent that delegates to specialist agents:

  1. Create Specialist Agents (separate workflows):

    • Research Agent (workflow with webhook trigger)

    • Writing Agent (workflow with webhook trigger)

    • Analysis Agent (workflow with webhook trigger)

  2. Create Manager Agent with tools that call specialists:

    • Add HTTP Request Tool nodes to your manager agent

    • Each tool calls a specialist agent’s webhook

    • The manager decides which specialist to consult

Example Structure:

User Input → Manager Agent → [Tools]:
                              - Call Research Agent (HTTP Request)
                              - Call Writing Agent (HTTP Request)
                              - Call Analysis Agent (HTTP Request)
                           → Synthesize responses → Output

Hope this helps!

1 Like
4

Hi @Mohammed_Rifad, welcome to the n8n community! As far as I know, n8n does not support a native agent team where agents freely talk to each other. The recommended approach is a workflow orchestrated multi agent setup. In practice, I use multiple agents with clear and isolated responsibilities, and the workflow decides when and how each one is called, often via sub workflows. Letting agents decide this on their own or share memory implicitly usually leads to non deterministic behavior, loops, higher costs, and harder debugging. Treating agents as specialized functions and letting n8n handle orchestration has been the most reliable approach for me.

5

Hi @Mohammed_Rifad that is a nice concept and it depends on the project outcome that how your workflow design would be, for an instance if you are creating a LLM council where multiple LLMs generate an output which gets valuated by one admin agent, you can refine this concept and make this in a loop until the Main Agent approves to a certain parameter, again the concept of multiple agent depends on the outcome you are looking forward to achieve, an example of something like that would be:

This is just a sample structure of how the agent should be passing down the information after computing their respective results, please clarify your agenda towards a specific tasks you want to incorporate this multi agentic communication kind of system in, that way we can suggest clearer pattern. Hope this helps.

2 Likes