Self-hosted n8n with Docker: the complete guide.

The three reasons that tip the scale.

n8n comes in two versions: the official cloud (n8n.cloud, run by the publisher) and self-hosting on your own server. Same product, same workflow editor, same integrations. Three differences tip the balance.

The cost model. Cloud bills by execution tiers: €20/month for 2,500 executions at the first tier, rising with volume from there. Self-hosted costs the price of your server, full stop. A workflow running every 15 minutes already generates 2,900 executions a month on its own: as soon as you're automating seriously, the cloud meter becomes the topic, while the VPS doesn't move.

Data location. Your workflows handle leads, invoices, customer data. Self-hosted, everything flows through your server, at the European host you chose. Your data-processing register gets simpler, and so do your enterprise clients' vendor security questionnaires.

No cap. No execution limit, no limit on active workflows, full access to technical features (code, HTTP requests, queues). Your instance grows with your needs, not with a pricing grid.

To place n8n against the alternatives: neither Make nor Zapier offer self-hosting. Our full comparison n8n vs Make vs Zapier breaks down the match tool by tool. For a full overview of the tool (use cases, pricing, AI agents), see the n8n guide.

Install n8n with Docker: the official commands.

n8n's official documentation, checked on September 27, 2026, is clear: Docker is becoming the only way to install n8n. From n8n 3.0, announced for October 2026, n8n is distributed only through Docker, and the old npm install n8n / npx n8n approach no longer works for a new install.

To try it on your machine: one command

With Docker installed and running (Docker Desktop on Mac or Windows, Docker Engine on Linux), n8n provides an install script:

curl -fsSL https://get.n8n.io | sh

The script checks that Docker is running, creates an n8n folder with the configuration (compose.yml, and a .env with generated secrets), downloads n8n and starts it on http://localhost:5678. It is safe to run again: if n8n is already set up in that folder, it changes nothing. If you would rather read a script before running it, the documentation suggests downloading it first (curl -fsSL https://get.n8n.io -o get-n8n.sh). On Windows, run it from WSL.

This install uses SQLite, a built-in database stored in a file. Perfect to explore, not for a team.

For production: Docker Compose, a domain and HTTPS

On a server, the documentation describes a Docker Compose install in a few steps: a project folder, an .env file, a local-files folder to exchange files with n8n, and a compose.yaml that runs n8n behind Traefik, a proxy that handles the HTTPS certificate and renews it on its own. The .env fits in four lines:

DOMAIN_NAME=yourdomain.com
SUBDOMAIN=n8n
GENERIC_TIMEZONE=Europe/London
SSL_EMAIL=you@yourdomain.com

Then sudo docker compose up -d, and n8n answers on https://n8n.yourdomain.com, over HTTPS only. Copy the full compose.yaml from the Docker Compose page of the n8n documentation: it is the reference version, kept up to date by the publisher.

Two production settings not to forget, according to the same documentation:

• PostgreSQL rather than SQLite as soon as several people work on the instance or workflows run around the clock. SQLite data is not carried over automatically: better to start on PostgreSQL from day one.
• The stable version, not beta. n8n ships a new minor version most weeks; as of September 27, 2026, stable is 2.40.7.

n8n says it itself: self-hosting is meant for people comfortable with servers, and a mistake can cost data or uptime. That is what the steps below are for.

License, limits, prerequisites: the honest picture.

The fair-code license

n8n isn't open source in the strict sense: it's fair-code (Sustainable Use License). In practice: internal use at your company is free, including commercial use. The restriction: you can't resell n8n itself as a service. For an SMB automating its own operations, the license costs nothing and constrains nothing.

What stays reserved for paid tiers

The free self-hosted version covers the essentials: every integration node, unlimited workflows, custom code, webhooks. Some team features stay reserved for the publisher's paid tiers: SSO, advanced per-project permissions, log streaming, externalized secrets management. For most SMBs, nothing blocking; at the scale of a mid-market company with several teams, it becomes a real question worth costing out.

The technical prerequisites

• A VPS from a European host: Hetzner, OVHcloud, Scaleway. 1 vCPU and 2GB of RAM to start, 2 vCPU and 4GB for comfort.
• A dedicated subdomain (for example n8n.yourdomain.com) pointing to the server.
• A Docker install, the only method from n8n 3.0: clean, isolated, easy to update.
• One person comfortable with a terminal, in-house or at your partner. Nothing more is needed.

n8n in production in 5 steps.

The sequence below is what we deploy for our clients. The commands are in the previous part; here, what matters for deciding is the real effort of each step. The guiding principle never changes: secure access and backups from day one.

Total time for a clean install: half a day, security and backups included. For the step-by-step technical detail, n8n's official documentation (docs.n8n.io) covers every variant, command by command.

Under €30/month, maintenance included.

The three-year comparison makes the point: a self-hosted instance costs around €900 over 3 years (VPS + monitoring), while the same workload on per-operation billed platforms runs into the thousands of euros. As your automations multiply, the gap widens: every new workflow is free on self-hosted, billed by usage elsewhere.

This microscopic infrastructure cost is also what makes more ambitious systems viable: a 24/7 AI agent orchestrated by n8n runs on the same machine, and the full budget breakdown for an agent is in How Much Does an AI Agent Cost.

Your data stays with you.

This is the deciding argument for many of our clients, and it's simple to understand: an automation workflow sees everything it processes pass through it. Incoming leads, supplier invoices, HR data, customer exchanges. The question "where does this data actually go" deserves a precise answer.

Self-hosted, the answer fits in one line: on your server, at the European host you chose, under contract with you. You choose the location, you keep the list of subprocessors, you control the logs and retention periods. When an enterprise client sends you their security questionnaire, every answer is in your hands.

Three practices round out the setup, and we apply them systematically:

• Encrypted credentials, a key kept safe. n8n encrypts the login credentials to your tools; the encryption key is kept off the server.
• Named accounts and a secure connection. One account per person, never shared, and the interface accessible only over HTTPS.
• Controlled execution retention. Execution logs contain the data that was processed: we configure automatic purging to keep only what's necessary.

For balance: n8n Cloud is a serious offering from a European publisher (German), with its own commitments. Self-hosting doesn't win by default against the other option; it wins through the level of control it gives you.

When n8n Cloud is still the right choice.

Honesty first: self-hosting isn't the right answer for everyone. Three situations where we recommend the cloud without hesitation.

• No one to run the server. No terminal skills in-house, no partner: the cloud removes the question entirely. A poorly maintained instance is worse than a subscription.
• Small volume, fast start. A handful of workflows, under 2,500 executions a month: €20/month with zero management is a good deal for validating use cases before industrializing.
• Need for managed team features. SSO, separate environments, vendor support under SLA: if these requirements arrive before the infrastructure skills do, the managed paid tier answers directly.

The typical trajectory we see: start on cloud to validate, switch to self-hosted once volume and data stakes rise. The migration is simple (workflows transfer as-is, credentials get recreated) and takes half a day to a full day.

What AUTOMATE ALL runs in production.

n8n is one of the pillars of our orchestration stack, alongside custom Python and PostgreSQL. Across 25+ systems in production, our deployment standard has settled into:

• Hardened instance: access only over an encrypted connection, nothing exposed unnecessarily, named accounts, minimal firewall.
• Tested daily backups: volume + database, externalized, restore verified.
• Active monitoring: alerts on workflow failures, disk space, certificates. Grafana dashboards when volume justifies it.
• Separated environments: a staging space, a production one. A workflow never ships to production without a dry run.
• 100% client ownership: server, workflows, credentials, documentation. You keep everything if our paths ever separate.

The method for deciding what to automate first is detailed in How to Automate a Business Process in 2026, and the full set of our disciplines is on the Services page.

Self-hosted n8n: what people ask us.