n8n self-hosted vs cloud: the true cost, including your hours
Self-hosting n8n is free to license and not free to run. Cloud prices as of August 2026, the real release cadence, and an honest hours-included comparison.
What you get from this guide
- n8n Cloud plan prices, executions and concurrency limits as of August 2026
- What the free Community edition includes, and the features n8n documents as paid
- The real release cadence, measured from n8n's own public releases
- An hours-included comparison instead of a server price against a subscription
- Where self-hosting is genuinely the right call - because often it is
The claim, and what is wrong with it
The comparison you usually see is a €20 subscription against a €5 server, followed by a conclusion. It is not wrong so much as incomplete, and the missing part is the only part that matters.
Self-hosting n8n is free to licence. It is not free to run. The cost is denominated in hours: installing it, putting SSL in front of it, upgrading it, backing it up, testing that the backup restores, noticing when it stops, and fixing it when it does. Those hours are real whether or not you write them down, and pricing them at zero is why so many self-hosting decisions get regretted six months later.
This page prices both sides properly. It is not an argument against self-hosting - for a lot of people it is clearly the right call, and the last section says exactly who. It is an argument against comparing a subscription to a server bill and calling that the answer. Every vendor figure below is taken from n8n's own pricing page, documentation and public releases, and dated August 2026.
What n8n Cloud costs as of August 2026
n8n prices Cloud on monthly workflow executions rather than per step, which the pricing page frames as making execution accounting independent of workflow complexity. That matters more than the headline price: a forty-node workflow and a three-node workflow cost the same per run, so you are never penalised for building the thing properly.
| Plan | Price | Executions / month | Concurrency |
|---|---|---|---|
| Starter | €20 / month | 2,500 | 5 concurrent executions |
| Pro | €50 / month | 10,000 | 20 concurrent executions |
| Business | €667 / month | 40,000 | Includes a self-hosted option |
| Enterprise | Contact Sales | Custom | Hosted by n8n or self-hosted |
All plans are advertised as including unlimited users and workflows and every integration, and unlimited active workflows. The page carries an annual toggle advertising a 17 percent saving; we could not reliably read distinct monthly-billed prices from the rendered page, so the figures above are the annual-billing ones and the monthly equivalents will be higher. Enterprise-listed features include SSO with SAML and LDAP, log streaming to external services, external secret store integration, unlimited shared projects, 200 or more concurrent executions, 365-day data retention, and dedicated support with an SLA.
Read the concurrency number, not just the execution count
What the free Community edition actually gives you
n8n's documentation is unusually generous here and worth quoting rather than paraphrasing. It describes the Community edition as suited to self-hosting with almost the complete feature set, and says you can stick with the basic Community edition for free, indefinitely.
| Edition | Cost | Designed for |
|---|---|---|
| Community | Free | Self-hosting with almost the complete feature set |
| Registered Community | Free | Community plus folders, debug in editor, and custom execution data. Unlock by registering your email |
| Business | Paid licence key | Teams needing paid features like SSO, environments, and projects |
| Enterprise | Paid licence key | Organizations needing the most advanced security, scaling, and governance |
The features n8n documents as needing a paid plan or edition:
- Custom variables
- Environments (dev, staging, prod)
- External secrets, and external storage for binary data
- Log streaming to external services - n8n notes standard logging is included
- Multi-main mode - n8n notes queue mode itself is included
- Projects, and workflow and credential sharing
- SSO via SAML and LDAP
- Version control using Git
Two of those deserve attention from a small team. Credential sharing being paid means a free instance run by more than one person gets awkward quickly. Git version control being paid means your change history lives in n8n's own workflow history rather than a repository, which is fine until the day it is not. Neither is a reason to avoid self-hosting; both are reasons to know what you signed up for. n8n itself notes that feature availability changes over time and that the pricing page is the definitive source, so check before you commit.
The server bill, honestly
n8n does not publish a hardware requirement, and we are not going to invent a benchmark. What we can do is list the line items honestly, because the usual comparison names one of them.
- The instance itself. A small VPS runs a light n8n workload comfortably. This is the number everyone quotes.
- The database. SQLite is the default and costs nothing extra. Postgres, which is what you want once the instance matters, is either another container on the same box - cheap, and a single point of failure - or a managed database, which is not cheap.
- Redis, if you move to queue mode with workers. Another container or another managed service.
- Backup storage, offsite. Small, but not zero, and worthless if it is on the same machine.
- Monitoring or uptime checking. Free tiers exist and are adequate for a single instance.
- A domain and TLS. Certificates are free; the domain is not.
- Headroom. Execution data grows, binary data grows faster, and an n8n instance that fills its disk stops in a way that looks like a bug.
None of these are large. Added up, a properly built small deployment is not €5 a month, but it is plausibly in the same range as the Starter plan. Which is precisely the point: on infrastructure alone the two options are close enough that infrastructure is not the deciding factor. The deciding factor is the next section.
The hours nobody counts
Here is the part that gets left out. These are the recurring jobs of running your own instance, with what each one involves according to n8n's documentation.
Initial setup
Install by Docker or npm, put a reverse proxy and SSL in front of it, point a domain at it, set the environment variables, set a custom encryption key rather than accepting a generated one, configure SMTP so user management emails work, and take a first backup. n8n documents all of this properly, which makes it tractable rather than instant. Anyone who has done it before will move quickly. Anyone who has not will spend an evening, and possibly a second one on the reverse proxy and webhook URLs.
Upgrades, at the real cadence
n8n documents updating a Docker deployment as pulling the image, stopping and removing the container, and starting it again; with Docker Compose it is docker compose pull, down, then up -d. Mechanically simple. The cost is not the command, it is the cadence and the vigilance.
The documented Docker Compose update
cd /path/to/your/compose/file/directory docker compose pull docker compose down docker compose up -d # Simple. The cost is doing this on a policy, # reading release notes for breaking changes, # and having a backup you can roll back to.
How often? We counted n8n's own public releases. In the four weeks to 4 August 2026, minor versions 2.31, 2.32, 2.33 and 2.34 shipped at roughly weekly intervals, with patch releases landing several times a week across supported lines; the stable release on 4 August 2026 was n8n@2.34.0. You need not take every one, but you need a policy, and reading release notes for breaking changes before a major version is not optional. n8n publishes migration guidance for major versions - the v10 guide is where MySQL and MariaDB deprecation is documented - which tells you these transitions are real events.
Backups, and the restore you have never tested
Two things must survive: the database and the encryption key. n8n documents setting a custom encryption key and, separately, rotating the data encryption key that protects credentials, which tells you how central that key is - a database restored without it gives you your workflows and none of your credentials. Test the restore into a fresh instance at least once, and preferably after any change to how you run it.
Monitoring, and the question behind it
n8n documents a Prometheus metrics endpoint, OpenTelemetry tracing configuration, and queue health check variables, so the instrumentation is there. Wiring it up, and wiring an alert to a human, is your job. The question monitoring forces you to answer is not technical: when the instance stops at two on a Saturday morning and the missed-call automation stops answering leads, who finds out, and how fast?
Security work
n8n publishes a real security surface for self-hosters: SSL setup, security audits you can run against your instance, disabling the public API, blocking specific nodes, SSRF protection, hardening task runners, verifying user emails, and rotating encryption keys. That list is a credit to the project. It is also a list of decisions that a Cloud user never makes, each of which takes an hour to understand and a few minutes to apply.
Incident response
The unbudgetable one. Disk fills, a migration fails, an upgrade changes node behaviour, the reverse proxy certificate expires. It is rare and it is not zero, and it always happens at an inconvenient time. Put a small annual allowance against it rather than pretending it is zero, because assuming zero is what makes the first incident feel like a betrayal.
The comparison with hours included
Rather than publishing invented totals, here is the arithmetic to run with your own numbers. Fill in your hourly rate and your honest hour estimates - and be honest, because the estimate is where this goes wrong.
The calculation, with your numbers
SELF-HOSTED, annual server + database + backups + monitoring = A setup hours (one-off, amortised year one) = B x your hourly rate upgrade hours (per year, at your policy) = C x your hourly rate backup + restore testing hours per year = D x your hourly rate monitoring + security review hours = E x your hourly rate incident allowance hours = F x your hourly rate TOTAL = A + (B+C+D+E+F) x rate CLOUD, annual plan price x 12 = G TOTAL = G The two lines are usually much closer than expected. If B+C+D+E+F is small because you already run servers, self-hosting wins clearly. If it is not, it does not.
| n8n Cloud | Self-hosted Community | Managed self-hosted | |
|---|---|---|---|
| Licence cost | €20 to €667 / month published | Free | Free licence, paid operator |
| Infrastructure | None | Yours to buy and size | Yours, run by someone else |
| Setup hours | Effectively none | An evening to a weekend | None to you |
| Upgrade hours | None - n8n handles it | Ongoing, at your chosen policy | None to you |
| Who answers at 2am | n8n | You | The operator |
| Data boundary | n8n's infrastructure | Entirely yours | Entirely yours |
| Custom nodes / full control | Limited to available options | Full control | Full control |
| Paid features | By plan | Need a licence key | By plan |
The row that decides it for most people is not a cost row. It is who answers at 2am. If the answer is you and you are content with that, self-host. If the answer is you and you are not, that discomfort is a price and you should pay it in euros instead.
When self-hosting is the right call
This page is not an argument for Cloud. Self-hosting is genuinely correct in several situations, and n8n itself says so - its documentation recommends self-hosting for anyone who needs full control over deployment, has customised use cases, has infrastructure and technical resources, or wants to run n8n for free, while recommending Cloud for anyone who wants to start quickly, does not have technical expertise, or does not want to manage infrastructure. That is a fair summary from the vendor with the most to gain from selling you Cloud.
- You already run servers. If Docker, a reverse proxy and a backup routine are things you maintain anyway, the marginal cost of one more container is genuinely small and self-hosting is close to free in practice.
- Data cannot leave your boundary. A regulatory, contractual or client requirement about where data lives settles the question regardless of the arithmetic. This is common in health, legal and public-sector work.
- Volume. Execution-based pricing is fair and predictable, and at high volume a fixed server cost beats it clearly. Run the numbers at your actual volume rather than assuming either way.
- You need custom nodes or unusual configuration. n8n lists custom nodes as a self-hosted capability, and if you need one, the decision is made.
- The workflows are the product. If automation is what your business sells, owning the runtime is a strategic position rather than a cost decision.
And the counter-case, stated as plainly: if the reason you are self-hosting is to avoid a €20 monthly subscription, and nobody in the business enjoys server maintenance, you are trading a small predictable cost for an irregular unpredictable one and a risk you have not priced. That is the mistake this page exists to prevent. Related reading: AI automation cost for the wider budget picture, what an automation build costs for the one-off side, and consultant versus agency for who should own it. On the build side, the Jobber and n8n integration guide shows what you do with the instance once it exists, missed call text-back with n8n is the workflow most people build first, and the Housecall Pro to Sheets sync is a good test of whether your instance can be trusted with something that runs unattended. If you run field service work, the HVAC and plumbing page sets out what to build in what order. Useful terms: self-hosting, total cost of ownership, vendor lock-in and execution vs task vs operation, which is the pricing unit this whole comparison turns on.
FAQ
- Is self-hosted n8n really free?
- The licence is. n8n's documentation describes the Community edition as free, self-hosting with almost the complete feature set, and says you can stick with the basic Community edition for free, indefinitely. What is not free is everything around it: the server, the database, backup storage, and your hours for setup, upgrades, monitoring and incident response. The honest framing is that self-hosting swaps a subscription for infrastructure plus labour. For plenty of people that trade is worth making. It is just not the same thing as free.
- What does n8n Cloud cost?
- As of August 2026, n8n's pricing page shows Starter at €20 per month with 2,500 executions and 5 concurrent executions, Pro at €50 per month with 10,000 executions and 20 concurrent executions, and Business at €667 per month with 40,000 executions and a self-hosted option, all quoted billed annually with an annual toggle advertising a 17 percent saving. Enterprise is contact-sales with a custom number of executions and hosting by n8n or self-hosted. All plans are described as including unlimited users and workflows and every integration, with pricing based on monthly workflow executions rather than per-step charges. We could not reliably read separate monthly-billed prices from the rendered page, so treat these as the annual-billing figures and check the pricing page for the monthly equivalents.
- What features do I lose on the free Community edition?
- n8n documents that some features require a paid plan or edition, including custom variables, environments, external secrets, external storage for binary data, log streaming (it notes standard logging is included), multi-main mode (noting queue mode itself is included), projects, SSO via SAML and LDAP, workflow and credential sharing, and version control using Git. There is also a free Registered Community tier: registering your email unlocks folders, debug in editor, and custom execution data. For a solo builder or a small team running production automations, the notable absences are credential sharing, Git-based version control and environments - and n8n's documentation is explicit that feature availability changes and the pricing page is the definitive source.
- How often do I actually have to upgrade?
- More often than people expect. Looking at n8n's public GitHub releases in the four weeks to 4 August 2026, minor versions shipped at roughly weekly intervals - 2.31, 2.32, 2.33 and 2.34 - with patch releases landing several times a week across supported lines. The current stable release on 4 August 2026 was n8n@2.34.0. You do not have to take every release, and a sensible policy for a small deployment is monthly plus anything flagged as a security fix. But you do have to have a policy, because an instance nobody updates for a year becomes an instance nobody dares to update.
- SQLite or Postgres for a self-hosted instance?
- n8n's documentation says n8n uses SQLite by default and also supports PostgreSQL, and that MySQL and MariaDB support was deprecated in v1.0. SQLite is genuinely fine for a light single-instance workload, and the SQLite settings include a pool size option that switches the file into WAL mode, which the documentation describes as much more performant and reliable than rollback journal mode. Postgres is the right choice once the instance matters, because it makes backups, restores and any future move to queue mode with workers straightforward rather than fiddly. Choosing Postgres on day one costs an hour; migrating later costs an evening.
- Does self-hosting scale better than cloud?
- It scales differently. n8n documents queue mode as included in the free edition, with Redis-backed workers configured through QUEUE_BULL_REDIS variables, so horizontal scaling is available to self-hosters without a licence. Multi-main mode, which is what you need for a highly available control plane, is listed as requiring a paid plan. So the ceiling on a free self-hosted deployment is high on throughput and low on availability, which is exactly the trade you would expect. If your concern is executions per month, self-hosting wins on unit cost at volume. If your concern is uptime without a person, it does not.
- Which should a small business choose?
- If nobody in the business enjoys running servers, take Cloud and stop thinking about it - n8n's own comparison recommends Cloud for anyone who does not have technical expertise or does not want to manage infrastructure, and that recommendation is fair. Self-host when you already run infrastructure, when a compliance or data-residency requirement makes it necessary, when you need custom nodes or full control over the environment, or when execution volume makes the unit economics obvious. Both are legitimate. The failure is choosing self-hosted to save €20 a month and then spending six hours a quarter on it.
- Can somebody else run it for me?
- Yes, and it is the option people forget. Managed self-hosting sits between the two: your own instance, your own data boundary, someone else carrying the upgrades, backups and pager. That is what we do when a client needs self-hosting for a real reason but has nobody to own it. It is worth pricing as a third column rather than treating this as a binary, because the reason self-hosting goes wrong is almost never the technology - it is that the person who set it up moved on.
Done for you
Want the instance run for you, or just an honest answer on which way to go?
We build automations on n8n either way, and we will tell you plainly whether Cloud or self-hosted fits your volume, your data requirements and how much of your own time you want to spend. If self-hosting is right and nobody wants to own it, we can run it. The audit that scopes it is free.
The consult is free, and you keep whatever we build.
Keep building
The rest of the cluster
Jobber + n8n Integration: The Full Build
OAuth against Jobber's GraphQL API from the HTTP Request node, webhook verification, and one complete workflow.
Read the guideMissed Call Text-Back with n8n (Free Build Guide)
Twilio trigger, business-hours guard, AI qualification, calendar booking, and the A2P 10DLC registration step.
Read the guideHousecall Pro to Google Sheets, For Real
The MAX plan gate, the webhook events worth syncing, an upsert-safe sheet, and what breaks at scale.
Read the guideFree AI Automation Audit Template
Score every recurring task and rank the top three worth automating before you build anything.
