How to Use n8n's AI Agent Node

n8n’s AI Agent node turns any workflow into a tool-calling agent powered by LangChain under the hood.

n8n is a fair-code workflow automation platform, and its AI Agent node is the bridge between traditional integration flows and LLM reasoning. The node wraps LangChain’s agent executor, letting an LLM decide which of your connected nodes to call as tools. You get agents that can send Slack messages, query Postgres, or hit any of n8n’s 400+ integrations.

What it is

A node in the LangChain category of n8n. It accepts a chat model, optional memory, and a list of tool nodes as sub-nodes. Each tool is described by a name and a natural-language description that the LLM uses to plan. Output is the agent’s final answer after any tool loop.

Install / set up

# self-host with docker
docker volume create n8n_data
docker run -d -p 5678:5678 \
  -v n8n_data:/home/node/.n8n \
  --name n8n n8nio/n8n

First run

Open http://localhost:5678, create your owner account, and start a new workflow. Add a Chat Trigger, then an AI Agent node, then an OpenAI Chat Model as a sub-node. Drag a tool like HTTP Request or Google Sheets into the Tool slot and describe what it does.

$ curl -X POST http://localhost:5678/webhook/chat \
  -d '{"chatInput":"summarize last 5 rows in sheet X"}'
{"output":"Here are the 5 rows..."}

Everyday workflows

• Build an internal “ops bot” that takes plain-English requests and calls your CRM, billing, or support APIs.
• Attach a Window Buffer Memory node so the agent remembers context across turns in a single chat session.
• Stack multiple agents with the Agent-as-Tool pattern to let a planner agent delegate to specialist agents.

Gotchas and tips

Tool descriptions matter more than you think. The LLM picks tools by reading the description field, not the node name, so “HTTP Request” with a blank description will be ignored. Write one clear sentence per tool describing when the agent should call it.

The agent can loop. If your tool returns an error the LLM doesn’t understand, it may retry forever until it hits the max iterations cap. Set maxIterations explicitly, return structured errors from tools, and watch the execution log the first few times you run a new agent.

Who it’s for

Teams already running n8n for automations who want to layer LLM reasoning on top. If you’ve got 20 existing workflows and want an agent that can invoke them, this is the easiest on-ramp — no new platform to learn.