What is n8n in 2026: what it automates, what its licence allows, and what its cloud costs against your own server

In n8n you draw the job on a canvas: a node that starts it (an email arriving, a time of day, a call to a web address) and behind it the nodes that do things with that data. Each node is an app or an operation, and the whole thing is called a workflow. Each time the workflow runs, that is an execution, and that word is what sets the price.

That is the difference n8n puts at the centre of its own pricing page: “Pricing based on monthly workflow executions, regardless of complexity”. A fifty-step workflow costs the same as a two-step one. Tools that charge per step or per task do not work that way. The name, by the way, is short for nodemation and is pronounced n-eight-n, according to its own repository.

The second thing that sets it apart is that it installs on a machine of yours. That edition is called Community, it is free and, in the words of its own documentation, it is “Self-hosting with almost the complete feature set”. What changes is not what you can automate: it is who holds the credentials of every app you connect and where your data passes through.

“Which one is cheaper, n8n or Make?” is one of the questions Google UK shows under n8n searches, and the answer is not in either price list: it is in the unit each one charges for. n8n counts one execution per complete run of a workflow, whatever its length. Make Make view › counts credits, and its own pricing page defines them: “Each module action in your scenario, like adding a Google Sheet row or fetching Gmail account data, counts as one credit.”

Put a ten-step workflow to run 3,000 times a month. In n8n that is 3,000 executions, inside the Pro plan. In Make it is 30,000 credits, thirty times its free plan of 1,000 credits a month. The shorter the workflow, the closer the two get; the longer it is, the more the count favours n8n. What Make gives back is that there is nothing to install, update or keep alive.

n8n's documentation has its own step-by-step guide for installing it on a Hetzner Cloud server, with Docker Compose and Caddy as a reverse proxy, and it gives the size: “For most usage levels, the CPX11 type is enough”, and if you plan to add the n8n Assistant, “at least 4 GB RAM / 2 vCPU”. The CPX11 it names is now listed only in Hetzner's two US locations; in Germany and Finland the Regular Performance line starts with the CPX12, with 1 vCPU and 2 GB, and the 4 GB size the guide asks for with the Assistant is the CPX22. The same guide is candid about who it is for: “n8n recommends self-hosting for expert users”, and anyone without experience of running servers is pointed to n8n Cloud.

Shared web hosting, the kind that runs a WordPress site, is not an option: it gives no administrator access and does not let a process run day and night, which is exactly what n8n needs. It has to be a virtual server. Some hosts package n8n for theirs: Hostinger's UK tutorial on n8n, for example, says installing it from its dashboard “takes just a few clicks”. On Hetzner you follow n8n's own guide.

Neither route bills in pounds. n8n's pricing page shows its plans in euros only, and the data behind the page carries no other currency. Hetzner lets you choose between euros and dollars and lists its prices without VAT; the VAT table on its own site sets 20% for the United Kingdom, so the 8 GB server at €35.99 comes to about €43.19 a month with that VAT added.

So a UK card pays in euros and the bank sets the exchange rate, with whatever foreign-currency fee your card charges. The euro figures on this page are the ones you will be billed; the pounds depend on your bank and on the day.

Moving data between apps that do not talk to each other. A form that lands in a spreadsheet, the spreadsheet that sends a message, the message that opens a task. It is the simplest kind of workflow and the one that uses the least resources.

Watching something and raising the alarm. A web address checked every hour, an email arriving with a word in it, a new file in a folder. Here the execution counter climbs fast: a workflow triggered every five minutes is 8,640 executions a month on its own, more than three times the Starter plan's 2,500 and most of the Pro plan's 10,000.

Calling a language model inside the workflow. Sorting emails, summarising documents, drafting replies. It is the use that decides whether the server needs 8 or 16 gigabytes, because a model on the same machine takes up real memory.

Acting as glue with an API that has no integration. The HTTP request node talks to any service that exposes an address, so the limit is not the catalogue of integrations, which its own repository puts at more than 1,500.

Putting a language model inside the workflow changes two things, and neither is n8n's price. The first is memory: a four-bit model with 8 billion parameters takes roughly 5 GB as a file and about 10 with the conversation and the system around it, so on an 8 GB server it does not fit next to n8n and on a 16 GB one it does. On Hetzner that is the CPX42, at €69.99 a month without VAT. The second is speed, which on a server without a graphics card, like the cloud servers in this guide, is set by the bandwidth of its memory and not by the number of cores.

If that part is the important part of your project, the decision stops being about hosting and becomes about the machine, and it is worth looking at the numbers before signing up for anything.