Multi-Agent Architecture in n8n

A single AI agent quickly reaches its limits — complex tasks require specialization. Multi-agent systems solve this problem by orchestrating multiple focused agents that collaborate. n8n, with its sub-workflows, webhooks, and flexible data processing, provides the ideal platform for this.

The Orchestrator Pattern

The most common multi-agent pattern is the Orchestrator Pattern: A central agent (the orchestrator) coordinates specialized sub-agents and aggregates their results.

                    ┌─────────────────┐
                    │   Orchestrator   │
                    │   (Main Workflow) │
                    └────────┬────────┘
                             │
              ┌──────────────┼──────────────┐
              ▼              ▼              ▼
     ┌──────────────┐ ┌──────────────┐ ┌──────────────┐
     │  Researcher   │ │   Writer     │ │  Reviewer    │
     │  Agent        │ │   Agent      │ │  Agent       │
     │ (Sub-Workflow)│ │ (Sub-Workflow)│ │ (Sub-Workflow)│
     └──────────────┘ └──────────────┘ └──────────────┘

Why Sub-Workflows?

n8n sub-workflows are perfect for multi-agent systems:

Advantage	Description
Isolation	Each agent runs in its own context — errors stay local
Reusability	An agent can be used in multiple orchestrators
Scaling	Sub-workflows can be scaled independently
Testability	Each agent is individually testable
Versioning	Agents can be updated independently

Agent Communication via Webhooks

For asynchronous agent communication, use webhook triggers:

Synchronous: The orchestrator calls sub-workflows directly and waits for the result
Asynchronous: Agents communicate via webhook endpoints — ideal for long-running tasks
Event-based: Agents publish events, other agents react to them

Synchronous Call (Default)

{
  "nodes": [
    { "type": "n8n-nodes-base.executeWorkflow", "parameters": { "workflowId": "researcher-agent", "mode": "each" } },
    { "type": "n8n-nodes-base.executeWorkflow", "parameters": { "workflowId": "writer-agent", "mode": "each" } }
  ]
}

Asynchronous Call (Webhooks)

{
  "nodes": [
    { "type": "n8n-nodes-base.httpRequest", "parameters": { "url": "https://n8n.example.com/webhook/researcher", "method": "POST" } }
  ]
}

Shared State Management

Multi-agent systems need shared state. In n8n, there are three strategies:

Strategy	Use Case	Tool
Workflow variables	Short-lived state within an execution	n8n Static Data
Database	Persistent state across executions	PostgreSQL / Supabase
Redis	Fast, ephemeral state for real-time coordination	Redis Node

Practical tip: Start with the synchronous orchestrator pattern and simple PostgreSQL state. Only switch to webhooks and Redis when latency requirements demand it. Simplicity beats elegance in production.