n8n Self-Hosted vs Cloud: What Each Really Costs a Small Business, and When Free Is Not Free

The free version costs about $14 a month, plus a job you did not have

Every n8n tutorial eventually says the same thing: why pay for Cloud when you can self-host it for free? So you rent a server to save 20 euros a month. A real server with backups runs about $14, so your savings are roughly the price of a sandwich, and you are now the person responsible for security patches on a system holding the keys to your Gmail, CRM and phone provider.

Sometimes that trade is right. At high volume it can be very right. This guide does the actual math, shows what you lose when you leave Cloud, and gives a plain rule for choosing. If you are still deciding between n8n and other tools, start with this site's comparison (builtwithagents.ai/blog/n8n-vs-make-vs-zapier-ai-agents). This post assumes you already chose n8n.

What each option really costs

n8n Cloud's bill is simple. As of November 2026, Starter is listed at 20 euros a month billed annually for 2,500 executions, and Pro at 50 euros for 10,000. One execution is one workflow run, however many steps it has. n8n's pricing page says manual test runs, sub-workflows and error workflows do not count, and that your workflows keep running when you reach the limit.

The self-hosted Community edition has no license fee and no execution count from n8n. You pay for the server. DigitalOcean lists a 1 GiB Droplet at $6 a month and 2 GiB at $12, and weekly backups add 20 percent. So a sensible setup is $14.40 a month. Hetzner is another option, with a 2 GB server in its EU locations listed at 11.99 euros and a US one at 17.99 euros, though its page marked those plans as currently unavailable when checked for this article. Add a domain if you do not already have one.

The AI model bill is identical on both. A workflow using Claude or OpenAI pays the same per token either way. One build on this site does run at $0 a month using local models on its own server, which is real but a different project: you need the hardware and you accept weaker models (builtwithagents.ai/strategy/self-hosted-ai-automation-n8n-docker-local-llms-zero-cost).

The cost people leave out is your time. These are estimates, since it varies by person. Budget 3 to 6 hours for a first install if you are comfortable in a terminal, and a day or more if you are not. Then an hour a month for updates, checks and the occasional surprise. Value that hour at even $50, and self-hosting at low volume costs about $64 a month against Starter's 20 euros. Once the volume climbs, the picture changes, as the examples below show.

What you give up on the Community edition

The missing features matter more for teams than for solo operators. n8n's docs list these as absent from Community: projects, sharing of workflows and credentials, SSO, environments, Git version control, custom variables, external secrets and log streaming. Free registration gets you folders, debug in editor and custom execution data.

Sharing is usually the one that hurts. Two people working on the same automations will feel that gap in the first week. A smaller annoyance is Google. On Cloud you click Sign in with Google. Self-hosted, you create your own OAuth client in Google Cloud Console, and n8n warns that consent and tokens expire after seven days if your app is External and in Testing status. That produces a very confusing failure one week after launch. Set the app to Internal if you are on Google Workspace, or publish it.

What you take on

Security is the real price. n8n's own docs say it recommends self-hosting for expert users, because mistakes can lead to data loss, security issues and downtime. That is not boilerplate. In January 2026, a flaw called Ni8mare (CVE-2026-21858) was rated 10 out of 10 for severity, letting attackers take over certain exposed instances through webhooks without logging in. The fix arrived in version 1.121.0. TechRadar, citing Shadowserver, reported nearly 60,000 internet-connected instances still vulnerable on January 11. A self-hosted install only gets a fix when someone applies it.

Think about what your n8n server holds: credentials for email, CRM, payments and phone systems. If you cannot promise to apply a critical patch within a day or two of hearing about it, you should not be running this on the open internet.

There is also a licensing limit. The Community edition's Sustainable Use License allows use for your own internal business purposes. If you are an agency planning to host n8n and charge clients for it, read the license and talk to n8n first, or put each client on their own Cloud account.

Three businesses, three answers

These are illustrations with assumed numbers.

A roofing company runs missed-call text-back, a review request and a lead alert, at 800 executions a month. Cloud Starter costs 20 euros, and no one has to touch anything. Self-hosting saves perhaps $10 a month and adds a job nobody there wants. Choose Cloud, and it is not close.

A small marketing agency runs eight internal workflows at 9,000 executions a month, and it employs one person who already manages servers. Pro at 50 euros against a $14.40 server is a saving of around $40 a month, and that person already has the skills. Self-hosting is defensible, but the team loses shared workflows, which may decide it the other way. Most teams should stay on Pro. Teams that work out of one account can make the move.

A lead-generation builder runs a scraper and enrichment flow that fires 60,000 times a month, and the runs take several minutes. That is far beyond Pro's 10,000 and Starter's 5-minute limit. n8n does sell larger tiers, with Business listed at 667 euros a month billed annually for 40,000 executions, self-hosted. For a technical solo operator, self-hosting is the obvious answer here: start on a 4 GiB server at $24, move the database from SQLite to Postgres, and the savings pay for the upkeep many times over.

How to decide and move this week

Step 1: Count your executions

If you are on Cloud, read the number off your usage page. If you are planning, multiply: runs per day for each workflow, times 30. A polling trigger that checks every minute and finds nothing does not start a run, so it costs less than you would fear. Busy webhooks and inboxes are what add up.

Step 2: Start on Cloud regardless

Build your first workflows on Starter. You will learn n8n without also learning Docker, Caddy and OAuth at the same time. You can move later, because workflows export as JSON files from the editor. Credentials you will need to set up again on the new server, which takes minutes each.

Step 3: Run the four-question test

Self-host only if you answer yes to all four. Are you over 10,000 executions, or do you need runs longer than Pro's limit? Is there a named person who is comfortable with SSH and Docker? Will that person patch within two days of a critical advisory, including during vacations? Can your business survive the automations being down for a day while someone restores a backup? One no means stay on Cloud. Data control can be a good reason to self-host too, but remember that if your workflow calls an outside AI model, your data still leaves the server.

Step 4: Install using n8n's own guide

Use n8n's DigitalOcean guide rather than a random video. It has you create a Droplet from the Docker marketplace image, log in as root and create a non-root user with SSH keys, clone the n8n-docker-caddy repository, create the caddy_data and n8n_data volumes, add a DNS A record for a subdomain like n8n.yourcompany.com, open ports 80 and 443 with ufw, edit the .env file and Caddyfile, and start it with sudo docker compose up -d. Pick the 2 GiB size, as 1 GiB leaves little headroom once AI workflows handle large payloads. Set a strong owner password as soon as it loads.

Step 5: Back up, then prove the backup works

Turn on Droplet backups. Then export on a schedule by running n8n export:workflow --backup and n8n export:credentials --backup inside the container, and copy those files off the server. Protect the encryption key in the n8n data directory, because encrypted credentials are useless without it. Then do the step almost everyone skips: restore onto a throwaway server and confirm a workflow runs. An untested backup is just a hopeful file.

Step 6: Put patching on a calendar

Subscribe to n8n's security advisories on GitHub so critical fixes reach your inbox. Then schedule a monthly update. Take a snapshot ($0.06 per GB a month on DigitalOcean), and run docker compose pull, docker compose down and docker compose up -d. Check that a real workflow still works. Update on a Tuesday morning, not on a Friday night.

Step 7: Watch it from the outside

One operator documented here replaced a $150 a month uptime service with a free n8n workflow (builtwithagents.ai/strategy/12-n8n-workflows-replaced-3000-month-manual-work). That is a good trick for watching your other sites, but it cannot tell you n8n itself is down, since a monitor on the same server dies along with it. Use an external uptime checker pointed at your n8n URL, and add an error workflow in n8n that alerts you when a run fails. Silent failure is the worst outcome here, since your leads just stop being answered and nobody notices.

What breaks

Disks fill up first. Execution logs pile up on a small server, so turn on execution pruning in your configuration. Updates sometimes change node behavior, which is why you snapshot first. Memory runs short when workflows handle big files or long AI responses, and the symptom looks like random failures. And the person who set it up leaves, taking the knowledge with them. Write one page describing where the server is, who pays for it and how to restore it.

Cloud has failure modes as well. Starter's 5-minute limit will stop a slow AI workflow, its five concurrent runs can queue during a burst of leads, and seven days of logs is short if you discover a problem late. Hitting those limits is the honest reason to move up a plan, and moving from Starter to Pro is usually cheaper than doing it yourself.

When not to self-host

Do not self-host because a video said it was free. Do not do it if you are under 2,500 executions and nobody on your team has used a terminal. Do not do it as your first n8n project. Do not do it for automations touching customer payments or health records unless someone qualified owns security. And if you are avoiding Cloud to save 30 euros a month, look at what your last hour of work earned you.

Frequently Asked Questions