What Is n8n? The AI Workflow Automation Platform Explained
n8n (pronounced n-eight-n, short for "nodemation") is a workflow automation platform with native AI. You build automations by dragging nodes onto a visual canvas, wiring them together, and letting data flow through the graph. What sets it apart from most no-code tools is that you can drop into real code whenever the canvas runs out of road. The vendor describes it as offering "the flexibility of code with the speed of no-code," and for technical teams that framing is accurate.
Jan Oberhauser founded n8n in 2019 in Berlin. It is built on Node.js and TypeScript, and under the hood every workflow is a directed graph of nodes: a trigger fires, data passes from one node to the next, and each node transforms, routes, or acts on it. You can run n8n as a hosted service (n8n Cloud) or self-host it on your own infrastructure. This breakdown covers what n8n actually is, how its licensing works (the honest version), the self-host versus Cloud decision, its integrations and AI capabilities, and who should reach for it.
What Is n8n?
At its core, n8n connects the apps and services you already use so that work moves between them without a human copying and pasting. A new row in a database can trigger a Slack message, a webhook can enrich a lead and write it to your CRM, an incoming email can be summarized by an AI model and filed into the right project. You define this logic visually: each step is a node, and connections between nodes describe the path your data takes.
The visual editor is the front door, but the Code Node is what keeps n8n from hitting the ceiling that pure no-code tools eventually hit. The Code Node runs custom JavaScript or Python anywhere in a workflow and supports npm packages, so when a built-in node does not do exactly what you need, you write the few lines that do. Most automations never need it, but knowing it is there changes how far you can push the platform.
Practitioner note: The mental model that helps most is "a workflow is a graph, not a script." Triggers start the run, data items flow through the nodes as a list, and each node operates on that list. Once you stop thinking in linear steps and start thinking in items moving through a graph, n8n's branching, merging, and looping behavior stops feeling surprising.
Fair-Code vs Open Source: The Honest Version
n8n is frequently called "open source," and you should know that this is not strictly accurate. n8n is fair-code. Its source is licensed under the Sustainable Use License (SUL) plus a separate n8n Enterprise License. The project adopted this arrangement in March 2022, replacing its earlier setup of Apache-2.0 with a Commons Clause. The distinction matters enough that it deserves a plain statement rather than a marketing gloss.
Here is what fair-code gives you in practice. The source code is visible on GitHub, you can self-host it, and you can modify and extend it. That covers most of what people actually mean when they reach for the phrase "open source." Where the SUL departs from an OSI-approved open-source license is in its restrictions: it permits use, modification, and redistribution but limits use to internal business purposes, and certain advanced features require a Business or Enterprise license key to enable.
If your requirement is a license approved by the Open Source Initiative (for example, for procurement or policy reasons), n8n does not meet that bar. The SUL is source-available with usage restrictions, which is a different category from MIT or Apache-2.0.
The SUL restricts use to internal business purposes, so you cannot, for instance, repackage n8n and sell it as your own hosted automation service. Advanced enterprise features sit behind a Business or Enterprise license key rather than in the freely self-hostable Community edition.
None of this is a knock on n8n. Fair-code is a deliberate choice that lets the company fund development while keeping the platform self-hostable and inspectable. It simply is not the same thing as open source, and a clear-eyed evaluation starts by naming the model correctly.
Self-Host or n8n Cloud?
n8n gives you two deployment paths, and the choice mostly comes down to how much infrastructure you want to own. Self-hosting runs n8n on your own servers; n8n Cloud is the managed, hosted version where the company runs everything for you. Both reach the same product, so this is a question of operations, not features.
To self-host, you deploy via Docker or npm (an npx command gets you a local instance), and the editor opens at localhost:5678. Run it without a license key and you get the free Community edition; add a Business or Enterprise key and those editions activate. When a single instance is not enough, queue mode splits execution across worker processes coordinated by a broker so you can scale out.
The tradeoff is straightforward. Self-hosting gives you maximum privacy and control, which matters when data residency or air-gapped environments are non-negotiable, but it asks you to bring server, scaling, and security expertise. n8n itself recommends Cloud for teams that are not infrastructure experts. If you would not be comfortable patching, backing up, and hardening a Node.js service in production, Cloud is the honest recommendation. For AI builds specifically, a self-hosted AI Starter Kit exists to get a working stack running quickly.
Integrations: How Many, Really?
n8n's value scales with the number of apps it can talk to, and the headline number depends on where you look. The GitHub repository and homepage cite 400+ integrations, the homepage in places also says 500+, and an earlier Wikipedia snapshot (December 2025) put it at 350+. These are vendor-reported figures, so treat the range as "several hundred and growing" rather than a precise count.
What matters more than the exact count is the escape hatch. When n8n does not have a dedicated node for a service, the HTTP Request node lets you call any REST or GraphQL API directly, which means "is it supported?" is rarely a hard no. Between the built-in catalog and custom HTTP and GraphQL calls, the realistic answer for most stacks is that you can connect what you need.
AI-Native: The AI Agent Node
AI is not bolted onto n8n as an afterthought. The platform is AI-native and built on LangChain, so you can assemble retrieval-augmented generation pipelines and multi-agent setups using the same node canvas you use for everything else. The centerpiece is the AI Agent node.
The difference between the AI Agent node and a plain language-model node is the difference between doing and describing. A plain LLM node generates text. The AI Agent node performs goal-oriented task completion: given an objective, it selects and executes actions across multiple steps, calling tools and chaining its own reasoning rather than returning a single block of prose. n8n supports several agent patterns, including Conversational, ReAct, Plan and Execute, and SQL agents.
On the supply side, you wire models in as sub-nodes. Chat-model options span OpenAI, Anthropic, Google Gemini, Mistral, DeepSeek, Groq, and locally hosted models through Ollama, among others. For retrieval, vector stores such as Pinecone, Qdrant, and PGVector plug in the same way. AI features are available on both Cloud and self-hosted deployments from version 1.19.4 onward, and the self-hosted AI Starter Kit packages a working stack so you are not assembling it piece by piece.
Practitioner note: Because the agent and its tools are just nodes, an n8n AI workflow can reach straight into the hundreds of integrations the platform already has. That is the quiet advantage: your agent's "tools" are not a separate ecosystem you have to build, they are the same Slack, database, and HTTP nodes you would use in any other automation.
Templates: A Running Start
You rarely have to start from a blank canvas. n8n ships a large library of pre-built workflow templates (vendor-reported as 800+, with the homepage in places citing 900+, so treat it as a several-hundred-plus library that keeps growing). You can browse them in-app or on the n8n.io Workflows page and clone one as a foundation to adapt.
The examples are practical rather than toy demos: an AI agent chat workflow, scrape-and-summarize for web pages, a Telegram bot, JSON-to-Excel conversion, company enrichment, backing up your workflows to GitHub, standing up an API endpoint, and joining datasets. For someone learning the platform, working backward from a template that is close to your goal is usually faster than building from scratch and teaches the node patterns along the way.
Who n8n Is For
n8n's own framing, "the flexibility of code with the speed of no-code," is also its best targeting statement: it is built for technical teams. If your automations occasionally need a snippet of real logic, an API that has no pre-built node, or the ability to self-host for data control, n8n fits. If you want a purely point-and-click tool and never want to see a line of code, the fit is looser, though plenty of users stay entirely on the visual canvas.
Frequently Asked Questions
Is n8n free?
Yes, with a caveat. The self-hosted Community edition is free to run from GitHub. n8n Cloud and the Business and Enterprise editions are paid, and some advanced features require a license key. So you can use n8n at no software cost if you self-host the Community edition and supply your own infrastructure.
How does n8n bill its paid plans?
By execution. One execution is a single run of an entire workflow, regardless of how many steps it contains or how much data it processes. This contrasts with tools that charge per task or per operation, and it can be more predictable for workflows that fan out into many steps. Pricing is fast-moving, so verify the current tiers and execution limits on the n8n pricing page before committing.
What is n8n built on?
Node.js and TypeScript. Workflows are modeled as directed graphs of nodes, the visual editor runs in the browser, and a self-hosted instance serves the editor at localhost:5678.
Does n8n compete with Zapier and Make?
They occupy the same broad category of workflow automation, but the models differ. n8n bills per execution and can be self-hosted; Zapier bills per task and is cloud-only. Make uses a per-operation model. The deeper comparisons live in their own articles, but the short version is that n8n leans toward technical teams that value code access and self-hosting.