Authority AI Editorial
n8n: workflow automation with AI agent nodes
n8n is a workflow automation platform built for people who want to connect apps, APIs, and AI models without writing a full application around each connection. It runs under a fair-code license: the source is visible and you can self-host it for free, with a paid Enterprise license and a hosted Cloud option for teams that want that handled. The project's own AI Agent node is what makes it relevant here. It lets you drop a model into a workflow as one step, give it tools and memory, and keep the surrounding logic in plain, inspectable workflow steps instead of hidden in a prompt. For a reader who already runs coding agents, n8n is the tool you reach for when the job is less about writing code and more about connecting systems that already exist, with an agent doing one part of the routing or decision making.
What it does
At its core n8n is a visual workflow builder: a canvas of nodes, each one a step, wired together by connections that pass data forward. Long before AI agents existed, this was already used for things like syncing a CRM to a spreadsheet or posting a Slack message when a form is submitted. The AI Agent node adds a step where, instead of fixed logic, an LLM decides what to do next: which of the available tools to call, in what order, and when it has enough information to answer. According to the project's README, n8n supports OpenAI, Anthropic, and Google as model providers for this node without locking you into one of them, and the wider platform connects to more than 1,500 integrations you can hand the agent as tools.
How you use it
You build a workflow on the canvas, add a trigger (a webhook, a schedule, an incoming email), and drop in the AI Agent node where you want a model to make a decision instead of a fixed rule. You attach tools to that node, which are just other n8n nodes exposed to the agent, so anything you could already call from a normal workflow (a database query, an API request, a file operation) becomes something the agent can choose to invoke. n8n's own AI product page describes pairing "AI agents, human approvals, and code" in the same workflow, which is the pattern worth noting: you can put a human-approval step right after the agent's output, so nothing it decides ships without a check, and you can drop back into a plain code node whenever an if-else is simpler than asking a model.
What to know before adopting it
n8n's license is the Sustainable Use License plus an Enterprise License, not a fully permissive open-source license: you can self-host and read the source, but reselling n8n as a competing hosted service is restricted. The core is written in TypeScript, and custom-code nodes can run JavaScript or Python if a built-in node doesn't cover what you need. Hosting is either self-managed, most commonly via Docker, or n8n Cloud if you'd rather not run the infrastructure. As of 2026-09-27 the repository shows 206,100 stars and 60,900 forks, and the project ships more than 9,000 workflow templates you can start from instead of building a workflow from a blank canvas.
Where it sits next to similar tools
n8n's angle is breadth of integration plus a visual canvas, not a minimal code library. If you want an agent embedded in your own Python or JavaScript codebase with a small footprint, a library-first tool is a better fit than a full workflow platform. If your problem is mostly about connecting an agent to dozens of existing business tools, and you want non-engineers on the team able to see and edit the workflow, n8n's node-based approach earns its place. Teams that already maintain a large collection of automation-ready procedures also tend to keep a broader set of ready-made skills on hand. The AI Ultimate Skill Collection, for instance, packages 3,029 ready-made skills in 12 packs that your agent follows step by step, which is worth knowing about if you're giving an agent methods rather than writing every one yourself.