You keep seeing n8n next to AI agent threads. Half the takes treat it like Zapier. The other half treat it like a weekend self-host project.
So, what is n8n? It’s a workflow automation tool built from nodes and connections. Run it on n8n Cloud or self-host the free Community Edition.
We started n8n 2.46.6 on our box and ran a Manual Trigger → Set workflow. The catch most beginners miss is how executions get counted.
What n8n is (and what it isn’t)
n8n runs workflows: a trigger starts a run, nodes do the work, and data moves as JSON items.
Disclosure: links to n8n and Make below are affiliate links. If you buy through them, we may earn a commission at no extra cost to you. It does not change what we write.
You connect APIs, databases, chat apps, spreadsheets and custom code without rewriting the same glue every month.
This page is the plain explainer. For scores and deeper judgement, read our n8n review.
It is also not OSI “open source,” even though the code is public. n8n ships under a Sustainable Use License (fair-code / source-available).
Self-host for your own internal work. Do not turn it into a hosted editor for other people’s workflows without a commercial deal.
Against the big two names:
- Zapier (no affiliate link here): simplest path for light zaps. Details in Zapier vs n8n.
- Make: visual scenarios many marketers already know. See Make vs n8n.
n8n’s pitch is control: branching, code nodes, self-hosting, and AI/agent nodes on one canvas. That power creates the learning curve.
How an n8n workflow actually runs
A workflow is a directed graph. A trigger starts it. Nodes transform data or call services until the run finishes or errors.
Trigger → nodes → items
A trigger starts the run: Manual, Schedule, Webhook, app event, chat, and more. Everything after it is a node with a job.
Between nodes, n8n moves items. Each item is a small JSON document, like {"greeting":"hello from lab","checked_at":"…"} from a Set node, or one item per API row from HTTP.
Need the webhook pattern first? Our what are webhooks guide covers it.
What counts as an execution
On n8n Cloud, pricing tracks monthly workflow executions. An execution is one start-to-finish run, whatever the node count or data size.
Unlike tools that meter every step, a 30-node workflow that runs once is still one execution. A two-node workflow that fires every minute can burn thousands a month.
Our take: map how often each trigger fires before you pick a Cloud tier. Frequency beats node count.
What we ran on the box
On 6 November 2026 we installed n8n 2.46.6 with npx [email protected] start. The editor answered on localhost:5678. Health check returned ok.
We built a tiny workflow named Lab Manual Set:
- Manual Trigger
- Set node that wrote
greeting = "hello from lab"andchecked_at = {{ $now.toISO() }}
A manual REST run produced execution id 1, status success, mode manual, wall time about 86 ms (started 2026-11-06T11:43:09.210Z, stopped …09.296Z).
Both nodes lived inside that single execution. The Set node emitted one item whose json held the greeting and timestamp, which is the item model in miniature.
What we did not test: Cloud UI, Assistant or AI credit burn, queue mode, SSO, or a live production webhook.
Cloud vs self-hosted: which lane fits you?
Two lanes: n8n Cloud (hosted) or Community Edition (free self-host under the Sustainable Use License for your own work).
List prices checked on 6 November 2026 (euro, billed annually on the cards we saw):
| Plan | List price | Where it runs |
|---|---|---|
| Starter | 20€/mo | Hosted by n8n |
| Pro | 50€/mo | Hosted by n8n |
| Business | 667€/mo | Cloud or self-hosted option shown |
| Enterprise | Contact sales | Cloud or self-hosted |
| Community Edition | Free | You host it |
Plans advertise unlimited users, workflows and integrations. You still pay against monthly executions and concurrent caps (Starter lists 5 concurrent; Pro goes higher). Confirm the live monthly allowance for your tier before you budget.
For side-by-side cost maths against Make and Zapier, use our automation platform pricing breakdown.
Pick Cloud if…
- You want an editor online today without Docker homework.
- You accept paying for executions and concurrent limits.
- You want hosted extras (shared projects, insights, Assistant credits) without running the stack.
Pick self-host if…
- Credentials and execution data must stay on your machines.
- You already run containers or VMs and can patch them.
- Your use fits the Sustainable Use License (your org’s workflows, not a multi-tenant editor for clients).
Self-host skips Cloud fees, not ops. Backups, upgrades, proxies and monitoring stay on you.
What n8n is good at (and bad at)
Good at: branching, HTTP and code nodes, mixing AI/agent steps with classic automation, and self-hosting. If you think in APIs and JSON, the canvas feels natural.
Bad at: being the friendliest first tool for “when form submits, send Slack.” Zapier still wins that lane for many people, and license limits matter if you sell automation as a product.
Weighing agents versus fixed workflows? Start with AI agents vs traditional automation and what is AI automation.
Who should start with n8n (and who shouldn’t)
Start with n8n if you want one canvas for APIs, code and AI nodes, and you accept Cloud execution billing or can run Community Edition cleanly.
Skip it for now if you need the simplest zap builder, you refuse JSON shapes, or your product lets outside customers design workflows on your instance. That last case needs a commercial talk with n8n.
Next steps that actually help
- Read the n8n review if you want our scored take before you pay.
- Check automation platform pricing if cost is the blocker.
- Build something small with our build an AI agent in n8n tutorial once the mental model clicks.
Or open n8n, recreate the Manual Trigger → Set pattern, and watch one execution. That screen beats another feature list.
Frequently asked questions
Is n8n free?
Community Edition is free to self-host for allowed uses under the Sustainable Use License. n8n Cloud is paid, with list prices from 20€/mo annually for Starter as of 6 November 2026. There is also a Cloud trial with execution limits. Free does not mean zero ops work on self-host.
Is n8n open source?
No, not in the OSI sense. The source is public under a fair-code Sustainable Use License with limits on commercial multi-tenant hosting. n8n’s own docs say they do not call the product open source. Say source-available or fair-code instead.
What is an n8n execution?
One execution is one full run of a workflow from trigger to finish. Node count and data volume do not multiply the execution by themselves. A chatty schedule can still spend thousands of executions even with few nodes.
Is n8n better than Zapier?
For complex branching, code, self-hosting and AI nodes, n8n often fits better. For the simplest no-code zaps, Zapier is still easier for many beginners. Pick from the job, not the hype. Our Zapier vs n8n comparison walks the trade-offs.
Can I use n8n for AI agents?
Yes. Current Cloud plans advertise Agents, MCP and AI nodes on the same canvas as normal automation. That does not mean every workflow should be an agent. Start with a fixed flow when the path is known.
Does n8n work without coding?
You can build plenty with triggers, no-code nodes and templates. The moment you need custom transforms, you will meet expressions, JSON items or a Code node. Expect a steeper ramp than pure zap builders.
Can agencies build client workflows on Community Edition?
Building and maintaining workflows for clients on your instance is generally allowed. Letting those clients log in to create or edit their own workflows as a hosted product is the restricted case. Read n8n’s license FAQ before you productise the editor.
Where should a beginner start with n8n?
Run one Manual Trigger workflow on Cloud or local Community Edition and inspect the execution. Then either read our n8n review for a scored verdict or follow the AI agent tutorial when you want a guided build.
The bottom line
What is n8n? A node-based automation tool for Cloud or self-hosted workflows. Items carry JSON. Executions bill the whole run.
Use it when you want control. Skip it when you only need the simplest zap.
Next step: open n8n, run one Manual Trigger workflow, then decide between the review, the pricing maths or a guided AI agent build.
