n8n Guide (2026): Pricing, Self-Hosting and 7 Workflows

A strange thing has happened to the tool submissions we read at FeaturedLane over the past year. Founders used to describe their integrations as "works with Zapier." More and more of them now say "works with n8n" first, and some say nothing else at all. Developer tools, AI agents, form builders, lead finders: the n8n node, or at least a webhook that n8n can catch, has become the thing people mention.

That shift is not just vibes. n8n's main GitHub repository sits at over 210,000 stars, its public template library lists 14,216 workflows, and in May 2026 an investment from SAP put the company at a reported $5.2 billion valuation, according to n8n's press page, roughly double what its $180 million Series C (led by Accel in October 2025) valued it at. For a workflow tool that started as one developer's project in Berlin, that is a lot of momentum.

So if you have been meaning to look at n8n properly, this is the guide we wish we had: what it is, how the pricing really works, whether to self-host, what the licence lets you do, and seven workflows that pay for themselves quickly.

What n8n actually is

n8n is a workflow automation tool. You build workflows on a visual canvas by connecting nodes. Each workflow starts with a trigger (a schedule, a webhook, a new row in a sheet, a new email) and then runs a chain of steps: fetch something, change it, decide something, send it somewhere.

Three things make it different from the tools most people compare it with.

You can drop into code whenever you want. The Code node runs JavaScript or Python inside the workflow, and the HTTP Request node can talk to any API, even one n8n has never heard of. In practice this means you rarely hit a wall. If a service has an API, you can use it.

You can run it yourself. n8n offers a hosted cloud version, but the same software is on GitHub and you can run it on your own server. A lot of teams start on cloud and move to self-hosted once usage grows, or the other way round once they get tired of maintaining a server.

It is built for AI workflows now. n8n describes itself as a platform "to build and deploy AI agents and workflows," and it has an Agent node that can call models from providers such as OpenAI, Anthropic and Google. More on that below, because it is both the most exciting part and the easiest place to waste money.

The README claims 1,500+ integrations. Take that number loosely: it counts every node, including utility nodes like HTTP Request, and the repo's own summary line says 400+. The honest answer is that the popular apps are covered natively, and everything else is one HTTP Request node away.

How n8n pricing works (and why it surprises Zapier users)

This is the part people get wrong most often, so it is worth slowing down.

n8n charges by executions. In n8n's own words, "an execution is a single run of your entire workflow. It doesn't matter how many steps are in the workflow or how much data it processes."

Zapier charges by tasks, and a task is counted each time an action step runs successfully. A five-step Zap that runs 1,000 times uses around 5,000 tasks. The same workflow in n8n uses 1,000 executions.

That difference is why long, branching workflows are so much cheaper on n8n, and why short single-step ones often are not. If your automations are mostly "new form entry, send to Slack," the billing model barely matters. If they are "pull 200 records, loop over them, enrich each one, score it, write it back," it matters a lot.

Here are the cloud plans as listed on n8n's pricing page in November 2026. Prices are in euros, billed annually (paying annually saves 17%):

Starter: €20 a month, 2,500 workflow executions, 5 concurrent executions
Pro: €50 a month, 10,000 workflow executions
Business: €667 a month, 40,000 workflow executions, currently listed as self-hosted only
Enterprise: custom pricing

Every plan includes unlimited users, unlimited workflows and every integration. You are paying for runs, not seats, which is unusual and genuinely nice for small teams.

Then there is the self-hosted Community Edition on GitHub, which has no licence fee. You pay for the server instead. A small VPS is enough for most small-team workloads, and that is where the "n8n is free" claims you see on Reddit come from. It is free the way a vegetable garden is free: no one sends you a bill, but somebody has to do the weeding.

Cloud or self-hosted?

We get asked this more than anything else about n8n, so here is a straight answer.

Choose cloud if nobody on your team wants to own a server, your volume fits comfortably under 10,000 runs a month, and you would rather spend the time building workflows. Updates, backups and uptime are someone else's problem.

Choose self-hosted if you run high volumes (polling workflows burn executions fast), you handle data you do not want leaving your own infrastructure, or you need custom nodes and packages. You also get no execution cap, which changes how you design things. You stop worrying about whether a check every five minutes is "worth" it.

The costs of self-hosting are real but boring: you need backups of the database, a plan for updates (n8n ships often), HTTPS in front of it, and some monitoring so you notice when it falls over at 2am. If that sentence made you tired, use cloud.

One middle path we like: start on cloud while you figure out which workflows actually matter. Once you know what you are running and how often, the self-hosting decision makes itself.

What the licence allows

n8n is "fair-code," not open source in the strict sense. It is distributed under the Sustainable Use License, and the short version is:

You can use and modify it for your own internal business purposes, or for personal and non-commercial use.
You can build workflows for clients as a consultant or agency. n8n's FAQ explicitly says commercial consulting and support services, such as building n8n workflows, do not need a separate agreement.
You cannot host n8n and sell access to it, or let your customers connect their own accounts and build workflows inside your product. That needs a commercial agreement.

For 95% of the teams reading this, none of that is a problem. It only matters if you are building a SaaS product with n8n embedded inside it for your users. If that is you, talk to n8n before you build, not after.

Seven workflows worth building first

The best first automation is a boring one you already do by hand every week. Here are seven we see working well, with tools from the FeaturedLane directory that fit into each one.

1. Form submissions that go somewhere useful

Most websites send form entries to an inbox where they rot. Point your form at an n8n webhook instead, then clean the data, drop obvious spam, add the lead to your CRM and post a short summary in Slack.

If your site has no backend, FormSubmit handles the form endpoint without server code and can pass submissions on through integrations. For surveys and longer forms, Rowform is a cheaper alternative to the enterprise form builders. Either way, n8n becomes the layer that decides what happens next.

2. Transactional emails from your own domain

Welcome emails, "your report is ready," trial-ending reminders. n8n can send email through SMTP, but deliverability is better through a proper sending API. Sendaro sends from your own domain with one API call and is free for 3,000 emails a month, which covers a small product comfortably. Call it from an HTTP Request node and keep the email copy in one place.

3. Monitoring alerts that reach the right person

Monitoring is only useful if the alert lands where someone will act on it. Use a dedicated monitor such as DNS Notify, which watches DNS records, SSL certificates and domain registration changes. Then use n8n to route alerts: certificate expiring in 14 days goes to a ticket, a nameserver change at 3am pages someone. The value is in the routing rules, not the alert itself.

4. Lead lists that enrich themselves

Mantis Ai builds a list of businesses matching your ideal customer description. A simple n8n workflow can take that list, check each website is live, pull a few public details with an HTTP Request node, score each lead, and write the good ones to your CRM. Keep a human step before anything gets emailed. Automated cold outreach without review is how domains end up on blocklists.

5. Social and market data on a schedule

If you track competitors or mentions, a scheduled workflow that pulls data and writes it to a sheet beats checking dashboards by hand. SocialCrawl gives you structured data from many social platforms through one API with one response format, which saves you from writing a different parser per network. Ecommerce teams can do the same thing with Amazon product and keyword data using Ecommerce Data API.

6. Web pages turned into clean input for AI steps

Feeding raw HTML to a language model is slow, expensive and messy. Siteprint turns rendered pages into structured JSON, Markdown or HTML snapshots that an AI step can actually work with. A useful pattern: fetch a competitor's pricing page weekly, convert it, ask a model what changed, and only notify you when something did.

7. Reports you can share with a link

At the end of a reporting workflow, you usually want a page someone can open rather than a spreadsheet attachment. BrewPage publishes HTML or Markdown to an HTTPS link, free and without signup. Have n8n build the report and publish it, then post the link in your team channel.

If you want ready-made starting points, StackLedge keeps a hand-reviewed directory of automation tools and n8n workflows, and n8n's own template library is huge. Treat templates as a starting point, though, not finished work. Most need their error handling rewritten.

AI agents in n8n: where they help and where they burn money

n8n's documentation describes an agent as "a chain that knows how to make decisions." That is a good way to think about it. A normal workflow follows the path you drew. An agent gets a goal and some tools, and chooses which tool to call and when.

That flexibility is great for messy inputs: support emails that could be about anything, documents with no fixed format, research tasks where the next step depends on what the last one found. It is a bad fit for anything you can already describe as rules. If "orders over €500 go to the manager" is the logic, write an IF node. It is faster, free, and it never decides to be creative.

A few habits that keep AI workflows sane:

Keep the agent's job narrow. One agent that triages, with clear categories, beats one agent that triages, replies and updates the CRM.
Watch model costs per run. An agent that loops can make a dozen model calls on a single execution. AICost is a handy way to compare model prices before you pick one, and a gateway like UnoRouter, which is OpenAI-compatible with one key for 290+ models, makes it easier to swap a cheaper model in later. If you want open models on your own GPUs, Runcrate puts 200+ open models behind one API.
Put a human in the loop for anything public. Drafting a reply is fine. Sending it without review is how you end up in a screenshot.
Treat agents in production like production systems. Logging, alerts, versioning. Services like Nestack Agent Care exist because this gets hard once you run more than a few. If your team also runs local AI tools, AUTOMATER.ai is a free tray app that tracks sessions across 41 of them.

Mistakes we see over and over

No error workflow. n8n has an Error Trigger node that fires when another workflow fails. Set up one error workflow that messages you, and attach it to everything. Silent failures are the reason people stop trusting automation.

Polling when a webhook exists. A schedule that checks an API every minute uses about 43,000 executions a month on its own. If the service offers webhooks, use them.

Credentials pasted into nodes. Use n8n's credential store. It keeps keys out of exported workflow JSON, which matters the first time you share a workflow with someone.

No retries. Networks fail. Most nodes have a "Retry On Fail" setting. Turn it on for anything that calls an external API.

Testing against live data. Pin sample data on a node while you build, so you are not sending forty test emails to a real customer while you get a field mapping right.

n8n vs Zapier in one paragraph

Zapier is still the easier tool for non-technical teams, with more polished app connections and less to think about. n8n wins when workflows get long or branchy, when you want code, when data has to stay on your own servers, or when you want AI agents with real control over each step. Plenty of companies run both: Zapier for marketing's simple handoffs, n8n for the workflows engineering cares about.

Where to start this week

Pick one task you do by hand every week. Build it in n8n with an error workflow attached from day one. Run it for two weeks before adding the next one. Small, reliable automations compound. A dozen half-finished ones just give you a new thing to babysit.

If you are building an automation or AI tool yourself, have a look at the AI and automation category to see what else is out there, and submit your tool when it is ready. A person reads every listing before it goes up.