n8n Pricing: What It Costs and When It Pays Off

Short answer

• What n8n costs: in the cloud, EUR 20 a month on the Starter plan and EUR 50 on Pro, each billed annually (EUR 24 and 60 billed monthly). Business costs EUR 667 a month billed annually (EUR 800 billed monthly) and runs only on your own server. The Community Edition is free if you run n8n yourself and use it only internally. All prices as of 25 October 2026.
• What drives the cost: every paid plan has a fixed monthly price and a quota of executions, meaning runs of a whole workflow. Starter includes 2,500 executions a month. A schedule every five minutes produces about 8,600 to 8,900 a month according to n8n, more than Starter includes. Once the quota on Starter or Pro is used up, further runs fail until the next month or until you upgrade.
• Where the data goes: n8n Cloud stores data on servers in Frankfurt, and Germany is on the Swiss Federal Council's list of countries with adequate data protection. An AI step sends the content on to a model provider. That provider works for n8n when you pay for the AI step with n8n's Gateway credits (n8n's own AI billing). n8n lists the US as the place of processing for several of these providers.
• My view: I do not use n8n myself, so this rests on n8n's documentation and on what my own scheduled jobs cost. The plan costs little with two or three workflows that do not run every few minutes, and the time of whoever notices a failed run and fixes it costs more.
• When n8n is the better choice: when several people build and look after workflows, many applications are connected and the team should be able to trace every run.

Where your data goes with n8n

n8n links applications into automated sequences called workflows, which you put together from individual steps in a visual editor. n8n calls these steps nodes. You can rent n8n as a cloud service or run it on your own server, and its role in data protection terms differs between the two.

n8n's pricing page answers the data protection question far down in its FAQ, in one sentence that says hosted plans store data in the EU, on servers in Frankfurt. That sentence covers the stored data, while the answer for the AI features n8n sells on the same page sits in the list of sub-processors, the other companies that process customer data on n8n's behalf.

n8n Cloud: servers in Frankfurt, a provider in Berlin

The company behind n8n is n8n GmbH, based in Berlin. Its privacy documentation describes n8n as both controller and processor for the cloud. As controller it decides itself why and how it processes some data, and as processor it processes data on your behalf. In that second role n8n offers a data processing agreement as part of its terms of service, and that agreement governs how n8n may process personal data on your behalf. It includes Standard Contractual Clauses, the EU's model contract terms for data transfers. You as the client must make sure the provider guarantees data security, under Art. 9 of the Swiss Federal Act on Data Protection (FADP).

The server location is a smaller problem than the AI nodes. Germany is on the list of countries with adequate data protection in Annex 1 of the Data Protection Ordinance. Personal data may go there without additional safeguards (Art. 16 FADP).

AI nodes: the content travels on

AI nodes on the Starter and Pro cloud plans can run on Gateway credits, without an account of your own with the model provider. n8n then forwards the request to the provider through its own account. The provider receives the content, for example the prompt you send to a model. Any personal data inside that content reaches it too, but your identity does not.

The list of sub-processors names the US as the place of processing for several of these AI providers. Annex 1 of the Data Protection Ordinance counts the US as adequate only for organisations certified under the Swiss-U.S. Data Privacy Framework. The model you choose in the node decides which provider gets the content, and n8n's service pricing page lists each model under its provider. Look up a provider in the US on the official Data Privacy Framework list and check that the line for the Swiss-U.S. Data Privacy Framework says Active. If it does not, or if the provider processes data in a country that is not on the Federal Council's list, the disclosure needs other safeguards under Art. 16 para. 2 FADP, such as standard data protection clauses, so ask n8n which ones cover that provider. The duty stays with you even when the provider works for n8n, because n8n processes the data on your behalf. If you disclose data abroad, Art. 19 para. 4 FADP requires you to tell the people concerned which country it goes to.

Self-hosted: data with you, telemetry to n8n

n8n is neither controller nor processor for your workflow data when you run it yourself. The data stays on your infrastructure. n8n still sends telemetry by default, including which node types you connect, the domains of the services you connect, error messages, execution counts and the truncated IP address of your server. n8n says the data that flows through your workflows and your credentials are not part of it.

You cannot switch telemetry off in the interface, only through environment variables, settings that someone with access to the server has to set. With a paid licence, your n8n always reports its usage counts. By default n8n deletes the records of past executions after 14 days and keeps at most 10,000 of them, and both values can be changed through environment variables. Deleting personal data is your responsibility, and n8n recommends shortening the period to a few days with the variable EXECUTIONS_DATA_MAX_AGE.

What n8n costs

Source: n8n.io/pricing, read from Switzerland on 25 October 2026. Francs at the ECB reference rate of 23 October 2026 (1 euro = 0.9227 francs). Max run time is the longest a single run may take.

All plans allow unlimited users and workflows, but how much those users can share depends on the plan. n8n's help pages say that every further execution that starts automatically fails at once when the quota on Starter or Pro is used up. That lasts until the counter resets at the start of the next calendar month or you move to a bigger plan, and n8n does not re-run the failed executions. The pricing page says Business workflows keep running and extra costs may apply.

n8n aims Business at companies with fewer than 100 employees and lists single sign-on, separate environments and version control with Git among its features. Business is not yet available in the cloud, and n8n keeps a waiting list for it. Pro also comes in a bigger tier with 50,000 executions, and Enterprise has no public price. Companies with fewer than 20 employees and less than EUR 5 million in funding can pay half the Business price on n8n's start-up plan.

What counts as an execution

An execution is one whole run of a workflow, however many steps it has. The n8n documentation says only production runs count. Test runs in the editor do not count, and neither do sub-workflows or error workflows. A sub-workflow is one that another workflow starts, and an error workflow runs when another workflow fails.

How often it counts depends on the trigger. A schedule counts every time it fires, even when there was nothing to do. A polling trigger, such as the one for Google Drive, checks a service for new data at set intervals and counts only when it finds new data.

Worked example: one workflow every five minutes

Since 28 September 2026, my own online shop has sent an automatic acknowledgement when a customer withdraws from a purchase. A withdrawal cancels the purchase under the 14-day withdrawal right that EU law gives consumers who live in the EU. It runs without n8n, which I do not use myself. A schedule at the cloud provider Cloudflare starts a check of the shop's mailbox every five minutes. That is 288 runs a day, 8,640 in 30 days.

If I rebuilt this workflow on n8n Cloud with a schedule as the trigger, it would not fit into Starter with its 2,500 executions. The quota on Starter would be used up after just under nine days, so the workflow would need Pro. A polling trigger that counts only when new mail arrives would make the plan depend on how much mail comes in. Starter would be enough for my shop, because by July 2026 the mailbox held 1,072 messages going back to October 2024, fewer than Starter allows in a single month.

Community Edition: the licence is free, running it is not

n8n states that the Community Edition has almost the full feature set. Missing features include workflow and credential sharing, projects, single sign-on, separate environments and version control with Git. The Sustainable Use License limits use to your own internal business purposes or to non-commercial or personal use, which is why n8n does not call itself open source.

Servers, backups and updates are up to you. n8n has published 196 security advisories on GitHub since January 2026, 19 of them rated critical (as of 11 October 2026), so updates are not a formality.

In November 2025, email and websites of many SMEs hosted at a Swiss provider stopped working, as 20 Minuten reported (in German). Those customers at least had a provider whose job it was to get things running again. When n8n runs on your own server, you are the hosting provider and that work falls to you. Before the first workflow, settle who is responsible for updates and backups, and who takes over when the person who set everything up leaves.

What my scheduled jobs cost without n8n

My website has a page that compares AI models. A scheduled job that needs no workflow platform fetches the data for it twice a day. It runs at GitHub, where the website's code also lives.

The same job also runs once a week for another fetch. From 1 to 10 October 2026 there were 22 runs, 20 on the twice-daily schedule, one weekly and one started by hand, with 40 minutes of compute time in total. GitHub rounds each job up to whole minutes, which brought the total to 48 minutes. GitHub's list price for a Linux runner is USD 0.006 per minute, as of 25 October 2026. A runner is the machine the job runs on. At that price the 48 minutes come to about 29 US cents for those ten days, or just under a dollar a month, against EUR 20 a month for Starter.

The two daily runs would be about 60 executions a month on n8n, far below the 2,500 executions in Starter's quota.

The runs are scheduled for 06:00 and 18:00 UTC, and GitHub's own documentation says scheduled runs can be delayed under high load. All 20 twice-daily runs from 1 to 10 October started 2.3 to 6.9 hours late.

A delay like that does not matter for a benchmark page, but it would for an acknowledgement a customer is waiting for. My shop's withdrawal acknowledgement is scheduled at Cloudflare.

If a source fails during the fetch, the job saves what did come in and then marks the run as failed on purpose. The outage shows up that way, and someone still has to see the failure and fix the cause.

Vercel, where my website runs, also charges by usage, for the compute time it takes to rebuild the website after each change to it. I would pay for these builds with or without n8n. In the first eight days of a billing month, those builds came to USD 15.22, and part of that went on building the same version several times, which nobody needed. A needless execution on n8n uses up quota in the same way.

Since 7 October, I have had Vercel build only the versions of my site that go live or that I want to look at first.

With a handful of workflows that do not run every few minutes, the time of the person who fixes a failed run matters more than the price of the plan. Choose the setup that person already works in, a script if they read code and n8n if they already build workflows there.

When n8n is the right choice

n8n pays off most once more than one person works on the workflows. In the executions list, each person sees the runs of the workflows they have access to, and an error workflow can send a message when a run fails. n8n also comes with ready-made connections to many applications, and for AI agents you can require approval by a human before the agent uses a tool.

I think Pro gives you a lot of visibility for little money once you run ten workflows across several applications and three people look after them. In the cloud, Starter includes one shared project for the team and Pro includes three. A project is a shared space for workflows and credentials. Self-hosted, sharing of workflows and credentials comes only with Business, according to n8n.

When a script or your existing scheduler is enough

A few fixed workflows, one person who reads code

n8n is hardly worth it when all you have is a few fixed workflows that rarely change and one person who reads code. Often a scheduler you already have is enough: the one built into the service that stores your code, such as GitHub, the one at your hosting provider, or the one on your server.

Does a simple workflow need AI?

Readers in the comment sections of Swiss news sites keep writing that simple workflows need no AI, and for many workflows that is true. My shop's automatic withdrawal acknowledgement needs no AI. It consists of a fixed template and a check for a marker word that the withdrawal form puts into each message. An AI step on Gateway credits costs money per request and sends the content to one more provider.

The first run is the most dangerous

Before the first live run of an n8n workflow that reads mail, check how its trigger treats old mail. My first plan had the same withdrawal acknowledgement go through every unread message in the shop's mailbox. At that point the mailbox still held 1,072 messages going back to October 2024. Of those, 550 were unread, nearly nine in ten of them automatic reports, platform notifications and advertising. When I later sorted the mailbox, only about 50 of the 1,072 messages turned out to be genuine customer and supplier mail.

That first run, as planned, would have treated all 550 unread messages as new mail, however long ago they had arrived, and any old withdrawal among them would have got its automatic acknowledgement long after the customer had sent it. I marked the planned first run as a hazard to fix before anything was sent, and it never ran in that form. The workflow therefore only processes messages that arrived after a fixed starting point in the mailbox, and in the live test it left all the old messages untouched.

Your workflows decide the plan

For each workflow, write down how often it runs, which data it processes and who repairs it when it fails. That list is what the choice between n8n Cloud, your own server and a simple script rests on. The AI automation page shows how I automate processes like these. You can send me your list through the contact page.

Frequently asked questions