Short answer
n8n is a workflow automation platform you can run on your own infrastructure, built around nodes you connect visually with a code escape hatch when the interface runs out. It suits companies with real run volume, data they would rather not route through a third party, or a technical person who will own it. It is the wrong choice if nobody will maintain the server, or if you need a working automation this afternoon with no technical help available.
What n8n is, and what it is not
n8n is a node-based automation tool: each node does one thing — receive a webhook, query a database, call an API, transform a payload — and you wire them into a flow. That much it shares with every tool in the category.
Two things distinguish it. You can run the whole thing on your own server, which changes both the pricing model and where your data goes. And when the visual nodes cannot express what you need, you can write JavaScript or Python inline rather than abandoning the workflow. Together those make it feel less like an appliance and more like a toolkit.
What it is not is a shortcut past understanding your own process. The failure mode is identical to every other automation platform: a workflow built quickly, never documented, owned by someone who has since moved on. The official documentation is unusually good, which helps with the tool and not at all with the process.
The licensing question, answered properly
This one causes real confusion, so it is worth stating carefully. n8n's source code is public and you may run it yourself, but it is not distributed under a conventional open-source license. It uses a sustainable use license, which permits use for internal business purposes and restricts making n8n available to your own customers for them to build workflows on.
For an ordinary company automating its own operations, this is a non-issue. It becomes a real question the moment you plan to host workflows on behalf of clients as a product. If that is the plan, read the current license terms directly rather than relying on any summary, including this one — the terms have been revised before.
Self-hosted or hosted: the real trade
The pricing difference is what draws people to self-hosting. It is rarely the thing that decides whether it was a good idea.
Self-hosting gives you three things that matter: run volume that does not carry a per-execution cost, data that stays on infrastructure you control, and no ceiling imposed by someone else's plan tiers. The hosting documentation covers the setup thoroughly, and the setup is not the hard part.
What it costs you is ownership. Upgrades that occasionally change node behavior. Backups of the database where all your workflows live — the failure that ends companies' automation programs is losing that database with no recent copy. Certificate renewal. And monitoring that tells a human when the instance is down, because a stopped automation is silent by nature.
Our rule of thumb: self-host when someone specific has agreed to own it, or when data residency requirements leave you no choice. Otherwise pay for hosted and spend the saved attention on the workflows themselves.
Where n8n is the right choice
- High run volume. Per-execution pricing on other platforms is comfortable at low volume and can become the dominant cost at scale. Self-hosting flattens that curve.
- Data you would rather not route outside. Client records, health data, anything where a processing agreement is a genuine obstacle rather than a formality.
- Workflows that need real logic. Iteration over nested structures, conditional retries, custom transformations. The code node means you stop fighting the interface.
- An in-house technical owner. One person who understands servers changes this from a risk into an asset.
Conversely, if none of those apply, a more constrained hosted tool will get you further faster, and there is no prize for using the harder tool. We have written the two head-to-head comparisons that come up most often — n8n versus Zapier and n8n versus Make — laid out by what actually differs rather than by feature checkboxes.
Five mistakes that are hard to undo
- No backup of the workflow database. Everything you build lives there. Losing it means rebuilding from memory. Back it up before the second workflow, not after the twentieth.
- Credentials pasted into nodes. Use the credential store. Anything typed inline gets exported, shared and eventually leaked.
- One giant workflow. Past a certain size nobody can reason about it. Split by responsibility and call sub-workflows.
- No error branch. The default behavior on failure is to stop. Silently. Add an error path that notifies a person on every workflow that matters.
- Upgrading without reading the release notes. Node behavior does occasionally change between versions. Test on a copy before upgrading the instance everything depends on.
Using it as the spine for AI work
The most common serious use we see is n8n as the orchestration layer around a model rather than as pure integration plumbing: receive a document, extract fields with a model call, validate against your own rules, write to the system of record, notify a person if confidence is low.
That shape works well because the deterministic parts stay deterministic. The model handles the one step that genuinely needs judgment, and everything around it is ordinary, testable workflow. It is also considerably easier to debug than a fully agentic design — the distinction we draw in what AI agents actually do.
Keep the model call narrow and the surrounding logic explicit, and you get most of the benefit with a fraction of the failure modes.
Frequently asked questions
Is n8n open source?
Not in the strict sense, and the distinction matters if your legal team is involved. n8n publishes its source code and lets you run it yourself, but it is distributed under a sustainable use license rather than a standard open-source license. In practice you can self-host it for your own business without difficulty. What the license restricts is reselling n8n itself as a hosted service to others. Check the current terms before building a commercial offering on top of it.
Do I need to know how to code to use n8n?
No for the common cases, and it helps more than it does elsewhere. Most workflows are built by connecting nodes and mapping fields. The difference with n8n is that when you hit a case the interface does not cover, you can drop into a code node and write a few lines rather than being stuck. That escape hatch is the main reason technical teams prefer it — and the main reason it feels harder at first than the more constrained tools.
What does self-hosting actually require?
A small server, a database, and someone who will notice when it stops. The installation itself is well documented and typically an afternoon. The ongoing part is what people underestimate: upgrades, backups of the workflow database, certificate renewal, and monitoring that alerts a human. If nobody owns that, hosted is the cheaper choice regardless of the sticker price.
How does n8n compare to Zapier and Make?
Roughly: Zapier optimizes for breadth of integrations and getting a non-technical user to a working automation fastest. Make sits in the middle with a visual model that handles branching and iteration comfortably. n8n optimizes for control — self-hosting, arbitrary code, and pricing that does not scale linearly with run volume. The right answer depends far more on your run volume and who maintains it than on any feature comparison.
We build and hand over n8n systems for companies of 10 to 100 people, including the boring parts — backups, alerting, documentation. Book a call if you want it done once, properly.
