Lesson 2easy10 min
n8n AI Agent Live Demo
Walk through an operational AI agent workflow running inside n8n canvas.
Learning Objectives
- Inspect the anatomy of an active n8n AI Agent workflow
- Observe the sub-node connections: Model, Memory, and Tool nodes
- Track an execution run through the step-by-step debugger
n8n AI Agent Live Demo
Before assembling your own agent, examining a fully operational agent canvas clarifies how n8n represents agentic architectures.
Canvas Topology
An n8n AI agent is not a single node. It is an orchestration cluster centered around the AI Agent node:
[ Trigger Node (Schedule / Webhook) ]
│
▼
[ AI Agent Node ]
├── Chat Model (e.g. OpenAI GPT-4o / Claude 3.5 Sonnet)
├── Window Buffer Memory (Session state)
├── Tool: Google Calendar
├── Tool: OpenWeatherMap API
└── Tool: SerpAPI Search
│
▼
[ Delivery Node (Gmail / Telegram) ]
Advertisement
Sub-Node Connections
Notice the colored connection inputs at the bottom of the AI Agent node:
- Model Port (Purple): Supplies the intelligence. This accepts nodes like
@n8n/n8n-nodes-langchain.lmChatOpenAior@n8n/n8n-nodes-langchain.lmChatAnthropic. - Memory Port (Blue): Maintains conversation history across multi-turn interactions using Window Buffer or Vector Store backends.
- Tools Port (Orange): Grants the agent external capabilities. Any standard n8n node can be converted into an agent tool.
Observing Agent Thought Logs
When you click Test step on an agent node, n8n displays the internal reasoning trail:
json
{
"thought": "I need to check the user calendar for today before checking the weather.",
"action": "google_calendar_get_events",
"action_input": { "timeMin": "2026-09-15T00:00:00Z", "timeMax": "2026-09-15T23:59:59Z" },
"observation": "[{ title: 'Product Review', start: '10:00 AM' }, { title: '1-on-1 Sync', start: '2:00 PM' }]"
}
The agent determines next steps dynamically based on what the tools return.