n8n is a workflow automation tool: you connect apps and services on a visual canvas so that something happening in one (a form submitted, an email arriving, an invoice due) automatically triggers steps in others. Businesses use it for AI automation because it lets you drop language-model steps and AI agents into those workflows, call almost any API, run custom code when you need it, and host it on your own servers if you want your data to stay with you.

It sits in the same family as Zapier and Make, but it's built more for technical users and for workflows with real logic in them. This post explains how n8n works in plain terms, what its "fair-code" model and self-hosting option mean, how its AI features fit in, the kinds of workflows businesses build with it, the precautions, and who it suits. For a side-by-side with the alternatives, see our comparison of n8n, Make and Zapier for AI automation.

How does n8n work, in plain terms?

Think of n8n as a flowchart that actually runs. Every workflow has the same basic parts:

  • A trigger: what starts the workflow. A new row in a spreadsheet, a webhook from your website form, a new email, a schedule ("every weekday at 8 am"), or a message from a chat app.
  • Nodes: the steps. Each one does a single job: read from a CRM, filter items, transform data, send an email, call an API, run a bit of code, or ask an AI model to do something.
  • Connections: the lines between nodes that pass data from one step to the next.
  • Credentials: stored logins and API keys that let nodes access your accounts, kept separately from the workflow itself.
  • Executions: a record of each run, showing what data went through each step. This is what you look at when something breaks.

n8n ships with a large library of integrations for common apps, plus generic building blocks (HTTP requests, webhooks, code, databases) that let you connect to anything with an API even if there's no ready-made node. Node names and the library itself change over time, so check the current docs for specifics.

What does an n8n workflow look like on a normal day?

For illustration, here's a lead-handling workflow for a small B2B company:

  1. Trigger: someone submits the website contact form.
  2. Check: look the email up in the CRM. Existing customer? Route to account management and stop.
  3. Enrich: pull public company information from the email domain.
  4. AI step: a language model reads the message and returns a structured result: what they want, company size, urgency, a short summary.
  5. Decide: if the lead is a good fit, create a deal in the CRM and notify sales in the team chat. If not, send a polite reply with useful resources.
  6. Human checkpoint: a salesperson reviews hot leads before any personalized follow-up goes out.
  7. Log: everything that happened is written to the CRM.

That's a few minutes of work per lead that now happens in seconds, with a person still deciding who gets a call. We go deeper on this pattern in n8n lead capture and qualification with AI.

What makes n8n different from Zapier and Make?

n8nTypical alternatives
HostingCloud service or self-hosted on your own serversMostly cloud-only
Licence model"Fair-code": source available, with conditionsProprietary
LogicBranches, loops, error paths and code built inVaries; simpler tools are more linear
Custom codeJavaScript (and in some setups Python) inside workflowsLimited or none in some tools
AI featuresAI agent and LLM building blocks, memory, tools, vector storesGrowing AI steps, depth varies
Learning curveSteeper; rewards some technical comfortOften gentler for non-technical users

"Fair-code" means n8n's source code is publicly available and you can self-host it, but the licence puts conditions on some commercial uses. It isn't the same as a traditional open-source licence. If you plan to embed n8n in a product or offer it to clients, read the current licence terms or ask counsel.

None of this makes n8n better in general. Zapier and Make are often faster for simple app-to-app automations and for teams without technical help. n8n tends to win when workflows get complex, when you want to self-host, or when AI agents are central.

How does n8n handle AI and AI agents?

n8n's AI features let you do two kinds of things.

Single AI steps. A node sends text to a language model with instructions and gets a result back: classify this email, summarize this call, extract these fields from this invoice, draft this reply. These are predictable and easy to test.

AI agents. An agent step gives a model a goal plus a set of tools it can choose to use, such as looking up a customer, searching documents, or creating a ticket. The model decides which tools to call and in what order. n8n also offers building blocks for memory (so an agent remembers the conversation) and vector stores (so it can search your documents, the pattern known as RAG).

You can usually connect several LLM providers, or run a local model if you self-host and have the hardware. If AI agents are new to you, our explainer on what agentic AI is covers the concept.

The practical advice: use single AI steps wherever the job is well defined, and reach for agents only when the path genuinely varies. Agents are more flexible but harder to test and more expensive to run.

Should you use n8n cloud or self-host it?

Both are real options:

  • n8n's cloud service: n8n runs it, updates it and keeps it available. You pay a subscription, and plans differ in limits; check current pricing.
  • Self-hosted: you run n8n on your own server or cloud account. You control where data lives and avoid some limits, but you own updates, backups, security and uptime.

The choice usually comes down to data residency, budget and whether someone on your team (or a partner) will look after the server. We cover the trade-offs in self-hosting n8n vs. the cloud version.

What do businesses actually build with n8n?

  • Lead capture and routing from forms, WhatsApp and email into the CRM.
  • Inbox triage: classifying incoming emails, extracting details, drafting replies for approval.
  • Invoice and payment reminders driven by your accounting system.
  • Customer support assistants on WhatsApp or website chat.
  • Reporting digests: pulling numbers from several tools into one morning summary.
  • Document processing: reading PDFs, extracting fields, filing results.
  • Internal ops: onboarding checklists, approvals, syncing data between systems that don't talk to each other.

What precautions apply?

  • Credentials: n8n holds keys to your CRM, email and finance tools. Limit who can edit workflows and see credentials, and use least-privilege API keys.
  • Errors: a workflow that fails silently is worse than no workflow. Set up error alerts from day one.
  • AI cost: an agent stuck in a loop can make many model calls. Set spending limits at your LLM provider and cap steps per run.
  • Human checkpoints: anything that sends money, changes records customers rely on, or messages customers should have an approval step until it has earned trust. See human-in-the-loop AI agents.
  • Ownership: someone in your business must own each workflow, including what it does and what to do when it breaks.

Who is n8n right for?

A good fit: businesses with someone technical in-house or a partner, workflows with real branching logic, AI agents at the center, data that needs to stay on your own infrastructure, or high volumes where per-task pricing elsewhere gets expensive.

Probably not the first pick: a non-technical owner who wants three simple app-to-app automations this afternoon. A simpler tool will get you there faster, and you can move later.

Time to go beyond it: when workflows become a core product, need complex testing and deployment, or outgrow what a visual tool makes maintainable. That's when custom code earns its cost.

Frequently Asked Questions

Is n8n free?

The self-hosted version can be run without a subscription under its fair-code licence, though you pay for your own server and maintenance time, and some features may be limited to paid plans. The cloud service is paid. Check n8n's current pricing and licence terms for your use.

Do I need to know how to code to use n8n?

No, many workflows can be built visually. But n8n rewards some technical comfort: understanding data structures, APIs and occasionally writing a small code snippet makes a big difference for anything beyond simple automations.

Is n8n secure enough for business data?

It can be, if it's set up well. On the cloud service, n8n handles the infrastructure; when self-hosting, security is your responsibility, including updates, access control, encryption and backups. Either way, limit who can see credentials and workflows.

Can n8n build AI agents?

Yes. n8n has building blocks for language-model steps, agents with tools, memory and document search. They're a practical way to build agents that work inside your existing systems, with the usual need for guardrails, testing and human checkpoints.

Is n8n better than Zapier or Make?

It depends on the job. n8n suits complex logic, self-hosting and AI agents; Zapier and Make are often quicker for simpler automations and less technical teams. Our comparison post covers when each makes sense.

Wondering whether n8n fits your business?

If you're considering n8n for AI automation, or you've started and hit the limits of what you can build alone, tell us about the workflow you want to automate. We'll tell you whether n8n, a simpler tool or custom code is the right fit. See our AI automation service.