n8n — the complete guide for businesses

This guide is for people considering n8n in a business, not for hobbyists automating chores at home. It therefore concentrates on what matters inside a company: what the tool does and does not do, what it really costs, what the free edition allows, where the data goes, and when choosing n8n is a mistake. Every price and licence fact comes from n8n’s own documentation and pricing page as of 5 September 2026; where something may change, we say so.

What n8n is — and what it is not

n8n is a workflow automation tool: you connect services that have an API (e-mail, a CRM, an accounting system, a spreadsheet, a language model) into a chain of steps that runs automatically on an event or on a schedule. You assemble it on a visual canvas from ready-made blocks, and where a block is missing you write code. It was created in Berlin, is developed by n8n GmbH, and is released under a fair-code model — the source is public, but the licence is not an open-source licence in the strict sense (more on that below).

Three things set n8n apart from the rest of the market, and they usually decide the choice:

• You can run it yourself. The same product works as n8n’s hosted service or on your own server, in your own data centre, on your own terms.
• It bills per execution, not per step. One run of a workflow is one unit, however many nodes it has.
• AI nodes are built in. An agent, a language model, a vector store and conversation memory are ordinary blocks in the same editor as the e-mail or spreadsheet node.

What n8n is not: it is not a system for very high event throughput (thousands per second), it is not a database, and it is not where logic that your accounting or billing depends on should live. It is a glue layer — a very good one, as long as it is treated as a glue layer.

Five concepts you need to know

The whole of n8n’s documentation rests on five words. It pays to have them straight before opening the editor.

A workflow is one process: a graph of nodes joined by arrows along which data flows. A company typically has anything from a handful to a few dozen workflows, each with one job.

A node is a single step: fetch an e-mail, ask a model, write a row to a spreadsheet, test a condition. n8n ships several hundred nodes for specific services and a dozen or so logic nodes — IF, Switch, Merge, Loop Over Items, Wait — plus an HTTP Request node that talks to any API and a Code node in which you write JavaScript or Python.

A trigger is the first node, the one that starts the workflow: an incoming webhook, a new e-mail, a change in the CRM, a schedule (hourly, every Monday), a message in a chat window. A workflow without a trigger can only be run by hand.

An execution is one run of a workflow from trigger to end. It is the unit n8n Cloud bills and the unit n8n records in its history — together with the data that flowed through the workflow. That second fact returns in the GDPR section.

Credentials are stored accesses to services: API keys, OAuth tokens, passwords. n8n encrypts them with a single instance key; if you self-host, that key is the one thing you must never lose; set it yourself rather than rely on the value generated automatically on the server.

n8n Cloud or self-hosting

This is the first decision, and it deserves to be made deliberately, because it changes the cost, the responsibility and the legal position.

n8n Cloud is the same application, operated by the vendor: you receive an address, log in and build. You manage no server, no updates and no backups. You pay a subscription that depends on the number of executions per month.

Self-hosting is n8n running on your infrastructure — most often as a Docker container on a virtual machine (Hetzner, OVH, Azure, your own server room). The product itself is then free in the Community edition, but running it is your job.

n8n paid-plan pricing (as of 5 September 2026)

For a self-serve Cloud buyer, then, n8n Cloud means Starter or Pro; Enterprise can also be hosted by n8n, but is priced individually. Business is a plan for your own installation — you pay for a licence key that unlocks the features for larger organisations, and the server is yours. Active workflows are unlimited on every plan. “Concurrent executions” is how many workflows may run at the same moment — on Starter, five is a real ceiling. A “project” is a workspace with its own permissions, not a deployment. The price jump from Pro to Business comes from Business being a licence with features for larger organisations, not merely a higher execution allowance.

One thing in that table matters more than the prices: the definition of an execution. The vendor puts it like this: an execution is a single run of your entire workflow, however many steps it has or however much data it processes. A thirty-node workflow that handles a hundred invoices in one run is one execution. It cuts both ways, though: if the trigger starts the workflow separately for each e-mail, a hundred invoices are a hundred executions. The definition is favourable when you process data in batches. That matters for the comparison with Make below. Prices change — check the current pricing on n8n.io before deciding.

Self-hosting: what you really need

The recommended route is Docker Compose; there is also a one-line installer for Linux and macOS and ready-made guides for Hetzner, DigitalOcean, AWS, Azure and Google Cloud. Installation through npm has been marked deprecated from n8n 3.0, so do not start there.

The minimum that in practice means “production” rather than “trial”:

1. PostgreSQL instead of the default SQLite. The default database is fine for experiments; with real traffic and an execution history you switch n8n to PostgreSQL (DB_TYPE=postgresdb plus the connection details).
2. Your own encryption key (N8N_ENCRYPTION_KEY), set before the first credential is saved and stored off the server. Without it, the credentials in a restored backup cannot be decrypted — a database backup without a copy of the key is incomplete.
3. Backups of the database and of that key, tested by restoring, not by assuming.
4. Updates. n8n releases very frequently. With Docker an update is three commands (docker compose pull, docker compose down, docker compose up -d), but someone has to run them and read the release notes, because some nodes change behaviour between versions.
5. Queue mode (EXECUTIONS_MODE=queue, with Redis and separate worker processes) only once a single process is no longer enough. It is not needed on day one; multi-main mode (several main processes) is already a paid feature.

If nobody in the company will maintain this, self-hosting is not “free”. It is a deferred invoice for administration — and that is the most common reason we recommend Cloud to small businesses despite the higher price on paper.

The licence: what the free edition allows

n8n is not open source in the OSI sense. It is released under the Sustainable Use License, with files marked .ee. in their name — the Enterprise features — under a separate commercial licence. In practice, for a business this means:

• You may use, copy and modify n8n for your own internal business purposes and for non-commercial and personal use. Automating your own processes, with no limit on executions, users or time — yes.
• You may provide n8n to others only free of charge and for non-commercial purposes.
• You may not sell n8n itself: host it for clients for a fee, offer it as a service, or embed it in a product you sell. That needs a separate commercial licence; confirm with n8n which variant applies to your case.

The boundary most service firms ask about: may a contractor build workflows for a client? The licence text does not settle this explicitly. Our reading is that it is allowed — as long as the workflow runs in the client’s own instance (their Cloud account or their server) and the client uses it for their own internal purposes; before you build a service on it, confirm with n8n or a lawyer. What you may not do is stand up one n8n and sell several companies access to that one instance. The licence text is short and worth reading yourself: LICENSE.md in the n8n repository.

The Community edition without a key has almost the full feature set. Registering an e-mail address unlocks folders, debugging in the editor and custom execution data, free of charge. What stays paid — in a self-hosted installation — are the features that only matter in a larger organisation (on Cloud some of them, projects for instance, are part of the plans): projects and sharing workflows and credentials across teams, SSO (SAML, LDAP), environments (dev/test/prod), version control through Git, external secret stores, log streaming, custom variables, multi-main mode and external storage for binary files. The vendor notes that feature availability changes over time — another reason to rely on the pricing page rather than on articles.

n8n and GDPR: where the data goes

This is the chapter most guides leave out, and inside a company it is the most important one. The general rule: n8n is neither GDPR-compliant nor non-compliant — your use of it is one or the other. Three questions organise the subject.

Where is the data physically? On n8n Cloud, data is stored in the European Union, on servers in Frankfurt. The operator (n8n GmbH) is then your data processor and you need a data processing agreement with it — check, before you sign, whether and on what terms n8n enters into one. With self-hosting, that processor disappears from the list, but the hosting provider takes its place unless the machine sits in your own server room.

Where does the workflow send data? This question matters more than the first. Every node that calls an external API — a language model, OCR, a cloud CRM, a messaging tool — passes data to that provider. If a node sends the text of a customer’s e-mail to a language model, the model provider is a recipient of personal data regardless of where n8n itself runs. Self-hosting settles the question of n8n’s operator; it does not settle the question of the providers. Running the model locally (n8n has an Ollama node) removes only the model provider from that list, at the cost of quality and hardware; the OCR, CRM or messaging services called from other nodes remain recipients.

What does n8n remember? The execution history stores the data that flowed through a workflow. Self-hosted, n8n prunes finished executions by default after 14 days or once there are more than 10,000 of them (EXECUTIONS_DATA_MAX_AGE and EXECUTIONS_DATA_PRUNE_MAX_COUNT); running, waiting and annotated executions stay, and the limit can be raised with a single variable. On Cloud, check what retention your plan applies. In a company whose workflows carry invoices, CVs or customer requests, that is where, with the default execution-save settings, a complete copy of the last two weeks of those documents sits — longer if someone has raised the limit. Set execution retention deliberately on day one and record it in your retention policy rather than relying on the default. How we approach these three questions in our own projects is described on the data security and GDPR page.

n8n versus Make and Zapier

All three tools do roughly the same thing, and the choice between them comes down to four differences.

The billing model. n8n counts executions (one run of a workflow), Make counts operations (every module run for every bundle of data), Zapier counts tasks (usually every successful action step). On a simple three-step workflow the differences are cosmetic. On a twenty-step workflow processing hundreds of items a month, the difference on the bill can be large in n8n’s favour — and conversely, with thousands of very short runs it narrows.

Self-hosting. Of the three, only n8n can be run on your own infrastructure. For companies with requirements about data location, that often ends the comparison.

Code inside. n8n allows JavaScript and Python in the Code node and expressions in every field; Make has a more modest expression language, Zapier a code step with limits. The more unusual the process, the more this favours n8n.

The entry threshold. Zapier is the simplest, Make has the most readable visual editor, n8n demands the most understanding of data. That is not a flaw — it is the price of the freedom — but it must be counted honestly.

AI in n8n

n8n ships a built-in set of AI nodes: a chat trigger, an AI Agent node, language-model nodes (OpenAI, Anthropic, Google, a local Ollama and others), conversation memory, vector stores and nodes for extracting information from documents. They are built like everything else — on the canvas, from connections.

Two uses that make sense in a business straight away:

• Extracting structured data from e-mails and documents: the model receives the text and a schema and returns JSON with fields that go on into a system. This is the most common and the safest first step, because the result can be checked.
• Classifying and routing requests: the model assigns a category and a priority, and the workflow sends the request to the right person.

The AI Agent node, which decides for itself which tools to use, is tempting and very easy to over-engineer. When an agent makes sense instead of an ordinary workflow, and when it is merely a dearer and less predictable way of doing the same thing, is covered on the AI agent implementation page.

A first workflow step by step: an invoice from e-mail to the accounting system

The best first workflow solves one dull, repetitive problem and has a person at the end. Supplier invoices arriving by e-mail are the classic case. Here is that process in n8n, node by node:

1. An e-mail trigger (Gmail Trigger, Microsoft Outlook Trigger or IMAP) starts the workflow for every new message in a mailbox such as invoices@yourcompany.com.
2. IF checks whether the message has a PDF attachment. Without one, the workflow ends with no further action.
3. Extract from File pulls the text out of the PDF. Scanned documents need a separate OCR step.
4. Information Extractor (an AI node) receives the text and a schema: invoice number, supplier tax ID, net amount, VAT, gross amount, due date. It returns JSON.
5. IF compares net plus VAT with gross. If they disagree, the message goes to a “check manually” channel and the workflow ends.
6. Send and Wait sends the responsible person a summary with an approve button and waits for the answer.
7. HTTP Request, after approval, posts the document to the accounting system through its API.
8. An e-mail node (Gmail or Outlook, reply operation) sends the result back into the thread the invoice arrived in — without this step nobody learns that the document was booked.

Add two things that separate business automation from a demo: an Error Trigger in a separate workflow, set as the error workflow on each production workflow, that reports failed production executions (manual test runs are not covered), and execution retention set deliberately rather than left at the default fourteen days. More processes of this kind, with numbers, are collected in our post on examples of AI automation.

When n8n is the wrong choice

An honest guide has to say this plainly. n8n is the wrong choice when:

• Nobody will maintain it. A workflow built by someone who leaves becomes an invisible process the company learns about when it stops working.
• The process is not yet defined. An automation tool does not decide who approves what and what happens in the exceptions. The company decides that before opening the editor — we cover this in how to implement AI in a business.
• Volume is very high or the requirements are transactional. Thousands of events a second, exactly-once guarantees, settlements — that is not work for a workflow tool.
• A feature in a system you already have is enough. Many CRMs and ERPs have their own automation rules. If they do what you need, an extra layer is extra risk.

What next

n8n is one of the tools we build business automations with — alongside Make, Power Automate and plain code where code is simply better. The choice of tool is the last decision in a project, not the first; before that, you need to choose the process, name the number that is supposed to change, and check the data. What such a project looks like from diagnosis to maintenance is described on the AI automation for businesses page.