n8n: What It Is and When It Stops Being Enough (2026)
n8n is a fair-code workflow automation platform: self-host it for the cost of a server, or use n8n Cloud from €20 a month. We explain what it does, who it is for, the GDPR edge of self-hosting, and where a workflow ends: exactly where a decision and an exception begin. You start with a free process scan.
n8n is a fair-code workflow automation platform: the source is open, you can self-host it for the cost of a server, or run it on n8n Cloud from €20 a month. It connects apps and executes fixed steps beautifully. It ends where a decision and an exception begin: that is where you need an agent with boundaries, escalation, and a trail, from €6,000 net.
What is n8n?
n8n is a workflow automation platform: it connects your apps, data, and AI models so that repeatable processes run on their own. You build them visually, dragging blocks (nodes) onto a canvas and wiring them with lines. A trigger in one app sets off a chain of actions across others: fetch data, transform it, branch on a condition, send it onward.
Who is it really for? Teams that have at least one technical person. n8n is more powerful and cheaper at scale than Zapier or Make, but it expects more knowledge: that is a feature, not a flaw. It has 400+ integrations and native AI built on LangChain (n8n GitHub, 2026). If you want a tool someone configures once and forgets, n8n is the wrong road. If you want control and flexibility, it is the right one.
Is n8n free? Fair-code, open source, and self-hosting
Short version: n8n is free to self-host, but it is not classic open source. The source is open on GitHub under a fair-code license (the Sustainable Use License, which n8n introduced in 2022), and you can run it, modify it, and use it internally for free. The license restricts one thing: reselling n8n as your own hosted service without a commercial agreement (n8n Docs, 2026). For everyday business automation the distinction barely matters, but it is why n8n does not call itself open source.
In practice you have two roads:
- Self-hosting (Community Edition): free in license terms. You run n8n on your own server via Docker or npm and pay only for that server (for example a VPS) and the time to maintain it.
- n8n Cloud: the managed version. n8n hosts, updates, and keeps it available for a monthly fee. A sensible start for a team with no technical backbone.
n8n and GDPR: why self-hosting can be an edge
Self-hosting has one concrete advantage that companies handling personal data feel most: the data never leaves your infrastructure. You run n8n on your own server, choose an EU region, and keep data residency in the Union. Per n8n's documentation, for self-hosted versions n8n is neither a Controller nor a Processor, because it does not manage your data at all (n8n Docs, Privacy, 2026). That shortens the list of processors for which you need a data processing agreement.
Honestly, the other side of the coin. Since the data is on your side, you are responsible for deleting it, for retention, and for server security. n8n recommends self-hosting for technical people and warns plainly that a configuration mistake can end in data loss or a security hole (n8n Docs, 2026). n8n Cloud lifts that burden, but then the data passes through the vendor: n8n becomes the Processor, with a data processing agreement (DPA) and Standard Contractual Clauses (SCCs). The choice between the two is not religion, it is arithmetic: who has the time and skill to maintain a server, and whether the data has to stay with you.
Where n8n ends: workflow vs decision
n8n executes the steps you designed beautifully. An if/else condition handles the branch you anticipated. The problem starts with a case the rule did not anticipate, or with a step that needs judgment rather than a condition.
A simple test: a workflow ends where a decision and an exception begin. As long as the process is "if A, do B," n8n is enough. When "it depends" shows up (on the tone of an email, on the customer's context, on an ambiguous situation), you need something that makes that judgment within boundaries and leaves a trail. That is an AI agent.
Run it on your own numbers before you decide on the layer. This is your substitution, not our promise:
Is a workflow alone enough?
share of cases that are an exception or need judgment
x number of cases per month
= how many times per month you step in by hand anywayIf the result is near zero, an n8n workflow alone is enough and there is no point paying extra for an agent. If someone steps in by hand dozens of times a month, that is exactly where an agent works: it recognizes the exception, acts within the boundaries you set, escalates the hard case to a human, and records why it made a decision. If that decision touches a customer, the trail stops being a convenience and becomes a requirement: from 2 August 2026, Article 50 of the EU AI Act requires informing people that they are dealing with an AI system, unless it is obvious from context (as of July 2026).
n8n or an AI agent: one table
This is not an either/or in every company. An AI agent often uses n8n as the execution layer underneath. But you are buying two different things, so it is worth separating them. The key row is "who decides."
| Dimension | n8n (workflow) | AI agent |
|---|---|---|
| What it does | Executes fixed steps: fetches, transforms, and passes data between apps. | Runs a case to a result: assesses the situation and picks the next step. |
| Who decides | You, up front, by designing the flow and its conditions (if/else). | The model, on the fly, within the boundaries you set. |
| Exceptions | A case outside the rule halts the flow or ends in an error. | Recognizes the exception and escalates it to a human with context. |
| Trail | A run log: what executed and when. | A decision trail: what it did, why, on what basis, and at what cost. |
| When to choose | The process is stable, the rules are known, no discretionary decisions. | Cases are ambiguous, exceptions are frequent, judgment and boundaries are needed. |
How much does n8n cost?
The self-host license is €0; you pay only for the server. n8n Cloud is billed by workflow executions, where one execution is one full run of a flow, regardless of the number of steps (n8n Blog, August 2025). The n8n.io price list, billed annually (as of July 2026):
- Starter: €20/mo, 2,500 executions, 5 concurrent, 1 shared project,
- Pro: €50/mo, 10,000 executions, 20 concurrent, 3 projects,
- Business: €667/mo, 40,000 executions, self-hosted deployment, SSO and Git-based version control,
- Enterprise: custom pricing, cloud or self-host.
Monthly billing costs more. On top of that comes the cost of AI models if you use AI blocks: you calculate it as everywhere, as cents per call times volume.
And on our side, when a workflow is not enough and an agent enters: process automation from €3,500 net, an agent that runs a process from €6,000 net, full implementations €6,000–35,000 net. These are two different purchases: n8n is a tool someone has to operate; our implementation is a process with boundaries, escalation, a trail, and maintenance. The full price list is on the Syntalith pricing page.
When n8n is entirely enough
Honestly, and against what most "agent" pitches want to sell: very often you do not need an agent. A well-designed n8n workflow is enough, and the difference in cost and maintenance is large. n8n is enough when:
- The process is stable and deterministic. Moving data between systems, notifications, syncing, recurring reports, backups. Each step follows from the last, with no judgment.
- There are no discretionary decisions. Every case can be described by a condition, and an exception either does not happen or can calmly wait for a human.
- You have someone technical. Someone maintains the self-host, or deliberately picks Cloud and watches the execution limits.
If that is your case, do not buy an agent. And if AI is in play anyway (classification, summaries, drafts), the best setup is often n8n plus one narrowly bounded AI block, not a full agent. Before you add autonomy, check whether the process needs it at all: we break that down in the guide on what an AI agent is.
How to start
The cheapest sensible first step is to count the process, not to buy the tool.
- Name one process and count the share of cases that are an exception or need judgment (the formula above).
- Decide the layer: a stable process with no decisions is n8n (self-host or Cloud); many exceptions and judgments is an agent with boundaries.
- Book a free process scan: 30 minutes with an engineer plus a written takeaway in two business days. We will say plainly whether n8n is enough for you or you really need an agent.
What comes next depends on who is going to build it. If you want your team to build in n8n properly, that is what our AI-Native course and team training (from €1,200 net per day) teach: building with boundaries, not just clicking flows together. If you want someone to build and run it in production, that is AI automations.
Book a free process scan | AI-Native course | AI automations | What an AI agent is