n8n AI Agent Tutorial (2026): The New Agents Feature vs the AI Agent Node, on Free Self-Hosted n8n

Description and use

The article demonstrates building a small chat agent on a free, self‑hosted n8n instance using both the AI Agent node and the preview Agents feature, comparing their setup, execution, and limitations.

Problem

Build a small chat agent on a free, self‑hosted n8n instance using either the AI Agent node or the newer Agents feature, and compare their requirements, execution, and limitations.

Outcome

Both approaches ran successfully on a free, self‑hosted n8n instance. The AI Agent node requires a workflow canvas, at least one tool sub‑node, and a memory node. The Agents preview feature is built in the Agent Builder, supports built‑in integrations, workflows, and custom tools, and has session memory by default. The local 1.5B model performed poorly on arithmetic tasks, revealing limitations of small models.

Workflow

• Install n8n 2.43.8 via npm on Node 24
• Create owner account and verify Community plan
• Enable Agents via Settings API
• Set up local Ollama model (Qwen2.5‑1.5B‑Instruct)
• Build workflow 1: Chat Trigger → AI Agent node → Ollama Chat Model → Simple Memory → Calculator tool
• Build workflow 2: Agent Builder → define model, instructions, tools, memory, knowledge base
• Expose chat trigger publicly via HTTP
• Test both agents by sending chat messages and observing tool usage

Evidence

Both builds used a local model through Ollama: Qwen2.5-1.5B-Instruct, a small model that runs on a CPU.

Source

The test machine had no Docker, so n8n was installed from npm. The npm package for n8n 2.43.8 declares "node": ">=24.0.0"

Source

The workflow has five nodes: Chat Trigger, AI Agent, Ollama Chat Model, Simple Memory, Calculator tool

Source