n8n or Custom Code for Automation? What We Learned Running Both

For most business automation, start with n8n and reach for custom code when you hit a specific limit. That is the same advice you will find elsewhere. What follows is the part that is harder to find: the licence detail most articles get wrong, and what running it has actually cost us.

We run our own content pipeline on n8n. This comes from operating it, not from reading about it.

The licence, stated correctly

This page used to describe the self-hosted community edition as free and open-source. The first half is right and the second is wrong, so we have corrected it.

n8n ships under the Sustainable Use License, which is not an OSI-approved open source licence. It grants you the right to use, copy, modify, and distribute the software for your own internal business purposes or for non-commercial use, free of charge. What it does not grant is the right to provide it to others commercially — hosting n8n as a service, reselling it, or building a commercial offering on top of it requires a separate enterprise licence.

For a business automating its own operations, this changes nothing at all. You are squarely inside the internal-business-purposes grant and you owe nothing. It matters in exactly one situation: if your plan is to run n8n on behalf of clients as a product, you need to read the licence before you build the business, not after. We are flagging it because the widespread claim that it is open source could lead someone to make a commercial decision on a false premise.

What self-hosting actually costs

Free to licence is not free to run, and the gap is mostly time rather than money.

The server itself is minor — a small VPS handles a modest workload. The real cost is that you now own a piece of infrastructure. It needs updating. It needs securing, particularly because your n8n instance holds credentials for every service it connects to, which makes it a more attractive target than its traffic volume suggests. And when it stops at an awkward hour, someone has to fix it.

This page previously put a figure of $200 to $500 a month on that. We have removed it because it was an estimate presented as though it were a finding. The honest version: the licence is free, the server is cheap, and the maintenance burden is a real recurring cost that depends entirely on whether you already have someone who runs servers. If you do not, n8n Cloud is often cheaper once you count the hours properly.

Where n8n genuinely wins

The usual argument for visual automation is that it is faster to build. That is true and it is not the reason we still use it.

The reason is debugging. When a workflow fails, n8n shows you the run as a diagram with the failed step highlighted and the exact data that entered it. You can change one thing and re-run from that step rather than from the beginning. For a pipeline that touches five external services, being able to see which one returned something unexpected — without adding logging and redeploying — is worth more than the build-time saving.

• Failure is visible. The diagram tells you where it stopped and what data it stopped on.
• Re-run from the failure point. You are not replaying the whole pipeline to test a fix at step seven.
• Someone else can read it. A colleague can understand a canvas without being able to read your code.
• Code is still available. JavaScript and Python nodes mean the visual model is a default, not a ceiling.

When to move to code

The threshold is not volume, which is what most comparisons say. It is readability.

A visual workflow is easier to understand than code right up until it isn't. Past a certain amount of branching, the canvas stops being a picture of the logic and becomes a puzzle — you are tracing lines to work out what happens when a condition is false, which is precisely the job that written code does better. We hit this on one of our own workflows: it had grown conditional branches over several months, and the point where we should have rewritten it in code was several months before we admitted it.

The other genuine triggers are narrower than they sound. Very high throughput where per-execution overhead matters. A step that must be as fast or as cheap as possible. Logic that needs real tests, version control, and code review because it is business-critical enough that a silent failure is expensive.

The hybrid, concretely

The pattern that holds up is n8n as the orchestrator and code for the parts that need it. n8n owns the trigger, the schedule, the connections to external services, the retry behaviour, and the visibility. Where a step involves logic that is genuinely intricate, that step calls out to your own code and gets a result back.

You keep the thing n8n is uniquely good at — knowing what happened and where — while the complicated logic lives somewhere it can be tested and reviewed. This is what our own pipeline looks like, and it is not an architecture we designed up front. It is what it became after the parts that did not belong on a canvas were moved off it.

Building this kind of practical automation is a core part of our AI and automation work, and the code-heavy end of it overlaps with our custom software development. If you are weighing this up for your own operation, we are happy to look at the specific workflow and tell you which side of the line it falls on — including when the answer is that you do not need us.