n8nworkflow-automationself-hostinglocal-aigdpr

Self-hosting n8n: licence, real cost and when not to

November 20, 202613 min readPixel Management

This article is also available in Dutch

The n8n licence costs nothing, and that's where most companies stop comparing, too early.

Self-hosting n8n is running the n8n workflow engine on your own server or in your own cloud, so workflows, credentials and execution logs sit with you rather than on n8n Cloud in Frankfurt. For internal use it's allowed and free. It's rarely cheaper: in the worked example below, self-hosting costs more than five times n8n Cloud Pro. What you're buying is control, not a saving.

n8n Community, Cloud or Business: what do you get?

n8n ships the same engine in four forms, and the differences sit in collaboration and administration. n8n's pricing page and its edition comparison showed this on 6 November 2026:

CommunityCloud ProBusinessEnterprise
Price€0 licence€50 per month, annual contract€667 per month, annual contractQuote
Runs per monthLicence states no limit10,00040,000Custom
HostingYoursn8n, FrankfurtYoursEither
SSONoNoYesYes
Git and environmentsNoNoYesYes
Shared projectsNone36Unlimited

A run in n8n is one complete execution of a workflow, however many steps it has. That counts differently from Make, which matters later.

For a company of a hundred people or more, one line in the Community edition carries real weight: you can't share workflows and credentials. Only the instance owner and the creator can reach them. If three people build automations, they work side by side, with no Git version control, no SSO and no projects. That holds until someone leaves.

Business is the obvious next step, but n8n describes that edition as being for companies with fewer than 100 employees. Above that, the page points you towards Enterprise, which has no published price. Ask before you budget.

What does the Sustainable Use License allow?

n8n isn't open source in the legal sense. The Sustainable Use License 1.0 permits use and modification "only for your own internal business purposes", plus non-commercial and personal use. n8n's licence FAQ spells out what that means, and it's broader than many people assume. Customers may see the result. The test is who controls the workflow logic.

SituationCommunity licence
Staff build workflows for your own processesAllowed
Customers receive the output, such as a quote or status updateAllowed
n8n runs invisibly as the backend of your productAllowed, if only your backend talks to n8n
Customers connect their own accounts to workflows you builtAllowed
An agency builds for clients on its own instance, no client accessAllowed
An agency installs and maintains n8n on the client's serverAllowed, if it doesn't also host
Customers build or change workflows, including via your UI, API, MCP or an AI agentNot allowed
Your product's value comes mainly from n8nNot allowed
Offering n8n without branding, or hosting it as a serviceNot allowed

Watch the AI agent line. Should customers assemble their own automations through a chatbot, that falls under "Allow external end users to build or configure their own workflows", and n8n says you'd need an Enterprise licence.

Two details are missing from most summaries. A breach ends the licence automatically, though if you stop within 30 days of a notice it's reinstated retroactively. And the text gives no firm answer for a group with several operating companies: the definition of "your company", which covers entities under common control, is only used in the patent clause, and the FAQ doesn't mention groups. If you run one instance across several Dutch BVs, put the question to n8n in writing.

What does self-hosting n8n cost per year?

Self-hosting n8n costs about €8,200 a year in our example, against €1,500 for n8n Cloud Pro. The licence is the smallest of the items that matter. Admin time decides the outcome.

The assumptions, all indicative. A maximum of 10,000 runs per month, so the three options stay comparable. Server prices from Hetzner Regular Performance on 6 November 2026, excluding VAT: a CPX32 (4 vCPU, 8 GB) for production at €35.99 a month and a CPX22 (2 vCPU, 4 GB) as a test server at €19.99. Backup storage and monitoring are estimated at €25 a month. Internal admin time is priced at €75 an hour, matching the €600 a day in our AI running-cost model. n8n's pricing page doesn't mention VAT. Building workflows isn't counted, because that work is the same everywhere.

Cost per yearCloud ProCommunity self-hostedBusiness self-hosted
Licence€600€0€8,004
Production serverno separate cost€432€432
Test serverno separate cost€240€240
Backup and monitoringno separate cost€300€300
Admin12 hours: €90096 hours: €7,20096 hours: €7,200
Total, indicative€1,500€8,172€16,176

Those 96 hours are eight a month. Two updates with a test on the second server (three hours), reading advisories (one hour), operating system and monitoring (one hour), a quarterly restore test (a little over an hour on average) and some room for incidents. That isn't generous: according to its changelog, n8n releases "a new minor version most weeks".

Cut admin time in half to 48 hours and Community still costs €4,572, more than three times Cloud Pro. You can check the break-even yourself: €1,500 minus €972 of hosting leaves €528, or just over 7 hours of admin per year. Spend more than that and you're paying more than Cloud Pro.

Above 10,000 runs a month, the comparison changes. n8n then mentions possible overage charges or moving up a tier, whereas the Community licence has no meter. At high volume, self-hosting can win on money, but you can only put a figure on that with an n8n quote in hand. For how to set total costs side by side fairly, see build vs buy.

Service

Learn more about business automation?

View service →

How secure is a self-hosted n8n?

A self-hosted n8n is as secure as your patching rhythm. n8n's advisory page on GitHub listed 237 advisories on 6 November 2026. Of those, 223 date from 2026, with 19 rated critical and 97 high. Plenty of advisories also means n8n discloses them, so don't read this as a league table. It does tell you about the workload.

Three advisories show what lands on your desk:

  • CVE-2026-21858, published on 7 January 2026, scored CVSS 10.0: access to files on the server without logging in. n8n itself limits it to active workflows with a Form Submission trigger and a Form Ending node returning a file. The fix had been available since 18 November 2025 and Cloud was already updated. When you self-host, that step is yours.
  • CVE-2025-68613, code execution through expressions, has been in CISA's catalogue of actively exploited vulnerabilities since 11 March 2026.
  • CVE-2026-27577 does need an account. But anyone allowed to create or edit workflows could run commands on the server on unpatched versions.

That last one has a practical lesson: edit rights in n8n are effectively server rights, and that server holds the credentials for your ERP, your mail and your CRM. Put the editor behind a VPN or SSO proxy, expose only the webhooks you really need, and give n8n its own machine.

Keep the encryption key separately as well. Credentials sit encrypted in the database, and according to the backup documentation a restored database can't be decrypted without it. Store the two apart, and test them together.

Running n8n with a local model so data stays in-house

Running n8n with a local model means the steps where a language model reads text run on your own hardware, so quotes, emails and client files never touch an external API. It's the strongest reason to self-host. A cloud instance can only reach a local model if you open that model up to the internet.

The connection itself is small. n8n has built-in Ollama credentials with http://localhost:11434 as the default address. If n8n and Ollama run in separate containers, put them on the same Docker network and use http://ollama:11434. What we check first in these setups:

  • No open port. Ollama listens on 127.0.0.1 by default, and the Ollama FAQ describes no built-in authentication. So don't publish port 11434.
  • Cloud features off. Ollama also offers cloud models. OLLAMA_NO_CLOUD=1 turns them off, so nobody picks an external route by accident.
  • Telemetry off. A self-hosted n8n sends data to n8n by default. Its privacy page says that excludes workflow data but includes usage, error information and the domains your HTTP Request nodes call. The isolation guide sets N8N_DIAGNOSTICS_ENABLED, N8N_VERSION_NOTIFICATIONS_ENABLED and N8N_TEMPLATES_ENABLED to false. Note that switching off version notifications means tracking updates yourself.
  • Starter kit isn't production. The Self-hosted AI Starter Kit bundles n8n, Ollama, Qdrant and PostgreSQL, and n8n itself labels it "For testing only".

A closed network clashes with the paid editions. According to the pricing page, Business and Enterprise licences must reach the licence server daily, including the number of production runs, and the privacy page says that reporting can't be disabled. n8n doesn't list that requirement for the unregistered Community edition.

The model server isn't in the table above. The Hetzner servers in the example have no GPU, so hardware comes on top: see AI hardware for business. Our cost model lands at €15,000 to €40,000 a year for a production server, a different order of magnitude from the n8n server. Ollama itself is covered in Ollama for business.

GDPR with n8n: self-hosting moves the responsibility

Self-hosting takes n8n out of the chain, not the GDPR (the AVG, as Dutch regulators call it). For workflow data on your own server, n8n's privacy page calls itself "neither a controller nor a processor", and adds: "If you self-host n8n, you are responsible for deleting user data." On n8n Cloud, n8n is a processor. So you're trading a supplier with a data processing agreement for duties of your own.

Four things land with you:

  1. Execution logs contain personal data. Every saved run keeps the data that passed through the workflow. The default settings prune after 336 hours (14 days) or 10,000 runs. Whether that fits the storage limitation in GDPR Article 5 depends on the purpose, and successful runs often needn't be kept at all.
  2. Recovery has to be demonstrable. Article 32 asks for timely restoration after an incident and for regular testing of your measures. The quarterly test in the cost example covers that.
  3. Your hosting provider is still a processor. If the server sits with a provider, Article 28 applies even with n8n out of the picture. What to check is covered in data processing agreements for AI.
  4. Every external API stays external. If one workflow step sends text to a cloud model, that text leaves your server, however the rest is set up.

When does n8n outgrow a single server?

n8n on one server handles the editor and the runs in the same instance. When peaks or long runs start slowing the editor down, queue mode is the next step, and it's included in Community. You add Redis as a queue, move the database to PostgreSQL (SQLite isn't supported in this setup) and start separate workers with EXECUTIONS_MODE=queue and the same encryption key as the main instance.

There's a trap with files. Queue mode doesn't support filesystem storage for binary data. That leaves Community with the database, because S3 is only in Business and Enterprise. If you process lots of PDFs or scans, PostgreSQL grows fast, and your backups grow with it. High availability with multiple main instances is Enterprise only.

Plan for the major upgrade too. On 6 November, the documentation listed n8n 3.0 as early access. That version requires Docker and drops npm installs, cuts the code task timeout from 300 to 60 seconds and changes the storage folder path. n8n provides a Migration Report under Settings. Run that first and test on the second server: this isn't an automatic update.

When is n8n the wrong choice?

n8n is the wrong choice if nobody in your organisation owns the platform, or if the process fits better in a tool you already pay for. The comparison of Make, Zapier and n8n covers features. These are the cases where we'd advise against self-hosting n8n:

SituationBetter choiceWhy
Everything runs on Microsoft 365, SharePoint and DynamicsPower AutomateSits in your tenant and admin, including a gateway to local systems
Marketing or finance builds it, few steps per runMakeNo server, no patches
Up to 10,000 runs, no data that must stay in-housen8n Cloud ProSee the worked example
Customers must assemble their own automationsCustom software or EnterpriseThe Community licence rules it out
It's your core process, with testing and audit requirementsCustom softwareCode in Git with tests and review, not a workflow in a database
Nobody can spend time on admin every weekCloud or outsourced adminAn unpatched n8n is worse than no n8n

Power Automate Premium costs €13 per user per month, billed annually and excluding VAT, so ten builders cost €1,560 a year. According to Microsoft Learn, each user gets 40,000 actions a day, and the on-premises gateway is included in Premium. That licence is less than the admin hours alone on a private n8n server.

Make charges per action, not per run. According to Make itself, it starts at $9 a month for 10,000 credits, and there's no self-hosted version. A 25-step workflow running 10,000 times a month, uses about 250,000 credits against 10,000 runs in n8n.

Custom software wins once the logic is more code than workflow. If most of your steps are Code nodes, you're paying for an editor you don't use, and you're missing the tests and version control that software gets by default. Which processes to tackle first is covered in automating business processes.

Curious how much time you could save?

Request a free efficiency audit. We'll analyze your processes and show you where the gains are, no strings attached.