n8n is free in its self-hosted edition and paid in its cloud edition. The Community edition installs on your own server with no licence to pay, unlimited executions and access to every node. The vendor-managed cloud plans start at 20 euros per month on annual billing, a price checked on the official pricing page on 7 August 2026. Between those two options, the question that matters is not the advertised price but the total cost, and that cost is decided by three lines that appear nowhere in the pricing grid.
Tandem is an n8n partner agency and deploys the tool across small and mid-sized companies, from Doctolib to InfraVia, PayFit, Nomination and KparK. This article gives the dated official prices, what the free edition really covers, the calculation that lets you estimate your execution volume, and the hidden costs our teams price during framing.
Is n8n really free?
Yes, provided you host it yourself. The n8n Community edition is free forever, with no execution limit, no limit on active workflows and no restriction on the node catalogue. The official vendor documentation states that it holds almost the complete feature set of the product. That word "almost" covers nine functions reserved for paid plans.
- Single sign-on through SAML or LDAP, essential as soon as an IT department mandates a central directory.
- Workflow and credential sharing. In the Community edition, only the instance owner and the author of a workflow can access it, which blocks teamwork.
- Separate projects and environments, which let you keep a development space apart from a production space.
- Version control through Git, external storage for binary data, external secrets and log streaming to a third-party tool.
One nuance deserves to be known, because it often goes unnoticed. By registering your email address you receive a Registered Community licence key, free as well, which unlocks folders, debugging in the editor and custom execution metadata. The move takes two minutes and costs nothing.
How much does n8n Cloud cost in 2026?
Four cloud plans coexist, all with unlimited users and active workflows. The prices below were checked on the official n8n pricing page on 7 August 2026, on annual billing. Monthly billing costs 17 percent more.
| Plan | Price | Executions / month | Concurrent executions | Hosting |
|---|---|---|---|---|
| Community | 0 euros | Unlimited | Depends on your server | Yours |
| Starter | 20 euros / month | 2,500 | 5 | n8n Cloud |
| Pro | 50 euros / month | 10,000 | 20 | n8n Cloud |
| Business | 667 euros / month | 40,000 | Not published | Self-hosted under licence |
| Enterprise | On quote | Custom | 200 and above | Cloud or self-hosted |
Three elements of this grid deserve your attention before signing. The Business plan requires self-hosting, so it does not bring the peace of mind of a managed service, it brings enterprise features. Companies with fewer than 20 employees can apply for the Start-up plan and get 50 percent off Business. Finally, the free trial lasts 14 days and asks for no credit card on the Starter and Pro plans.
n8n cloud data is stored within the European Union, on servers located in Frankfurt, Germany. That settles part of the compliance objections, but not all of them, as we explain in our complete guide to n8n.
An n8n execution is not a step
The definition of an execution is the most misunderstood point of n8n pricing, and it is the one that decides your bill. n8n counts one full workflow run as a single execution, whatever the number of steps crossed and the volume of data handled. A three-node scenario and a thirty-node scenario consume exactly the same thing.
Legacy automation tools bill per operation or per task, meaning per step. A twelve-node workflow run 10,000 times in the month represents 10,000 executions on n8n and 120,000 operations elsewhere. The practical consequence is twofold. Your bill becomes predictable, since it depends only on trigger frequency, and you stop compressing your scenarios to save steps.
How many executions will your company consume?
n8n documents a simple estimation method, and it holds a nasty surprise for many teams. For a workflow triggered by a schedule, count how many times the schedule runs in the month. A daily trigger produces 30 or 31 executions, a trigger every 5 minutes produces 8,600 to 8,900. For a webhook-triggered workflow, count the real frequency of the event. For a conversational agent, multiply the number of conversations by the average number of messages exchanged.
| Scenario | Executions per month | Minimum plan |
|---|---|---|
| A sales report sent every morning | About 30 | Starter |
| A monitoring workflow running every 5 minutes | 8,600 to 8,900 | Pro |
| A support agent at 400 conversations and 6 messages | About 2,400 | Starter, with no headroom |
| Four business workflows in production | 10,000 and above | Pro or self-hosting |
The real cost of self-hosting
The free n8n licence does not make the project free. An entry-level server is plenty to host n8n at the start, since the vendor advertises up to 220 executions per second on a single instance in its performance documentation. At OVHcloud France, the VPS-1 is billed 3.81 euros excluding tax per month for 2 vCores, 4 GB of memory and 40 GB of disk, a price checked on 7 August 2026. The hardware cost of a self-hosted n8n therefore sits around 50 euros a year.
The true cost of self-hosting is counted in hours, not in euros. A production instance requires a PostgreSQL database, a reverse proxy, an SSL certificate, a custom encryption key for credentials, tested backups and monitoring. Then comes routine upkeep, namely version upgrades, key rotation and database pruning.
That pruning illustrates the kind of topic that lands on the technical team. By default n8n keeps finished executions for 336 hours, meaning 14 days, and at most 10,000 executions, whichever limit is reached first triggering deletion of the oldest ones. Without tuning those variables, a busy instance watches its database grow until the disk fills up. The problem is solved in one configuration line, provided you know it exists.
The three costs missing from the pricing page
Three spending lines escape the n8n pricing grid, and they are the ones that derail budgets.
- Calls to artificial intelligence models. A workflow that queries a large language model on every run pays for those calls to the model vendor, outside any n8n subscription. Louis Graffeuil documented this in his Optimia edition on the autonomous assistant OpenClaw, set up on a dedicated server at Tandem. The verdict was blunt, the performance came from the premium model used by default, and the bill climbed fast. On an agent under constant load, this line often exceeds the automation tool subscription itself.
- Operations hours. Half a day of setup, then a few hours per quarter for updates and incidents. Valued at internal cost, those hours quickly erase the gap between a server at 50 euros a year and a Starter subscription at 240 euros a year.
- Quota overages. On the Business and Enterprise plans, workflows keep running beyond the quota, but overage charges apply, billed 4,000 euros per bucket of 300,000 extra executions and invoiced 45 days after the fact. AI Assistant credits, for their part, cannot be topped up mid-month and do not roll over to the next one.
That workflow is broken down step by step in our article on automating a newsletter with AI. It illustrates the rule Tandem applies during framing, namely that the budget of an automation project is computed on the triptych of licence, infrastructure and models, never on the licence alone.
Does the free edition allow commercial use?
Yes for your own operations, no for reselling the tool. n8n is not open source in the sense of the Open Source Initiative, it ships under the Sustainable Use License, a fair-code licence that sets three limitations. You may use and modify the software for your own internal business purposes or for personal, non-commercial use. You may redistribute it only free of charge and for non-commercial purposes. You may neither remove nor obscure the licensing and copyright notices.
In practice, a company running its own automations on a self-hosted instance is perfectly compliant, even when those automations generate revenue. Charging for consulting or for building n8n workflows is allowed too, since the vendor lifted that restriction when it changed licence. The boundary lies elsewhere. Hosting n8n for your customers, reselling it white-labelled or making it the core value of a product you sell requires a commercial agreement with the vendor.
Cloud or self-hosted, how to decide
The choice between n8n Cloud and self-hosting is not settled on price, because the two options converge very quickly once the hours are counted. It is settled on three criteria, the sensitivity of your data, the skills available in house and the expected execution volume.
| Your situation | The right choice |
|---|---|
| You are testing, with no dedicated technical team | Cloud Starter, migrating later if needed |
| Your workflows handle sensitive customer data | Self-hosting, in France if required |
| You go past 40,000 executions a month | Self-hosting, cloud gets expensive |
| You need SSO and several teams | Paid plan, cloud or Business depending on infrastructure |
| You have nobody to watch a server | Cloud, without hesitation |
A small company starting out almost always gains from beginning on the cloud, validating two or three use cases, then deciding with full knowledge. That progression matches what we observe across the four levels of AI automation, where the vast majority of profitable projects stop at the deterministic workflow rather than the autonomous agent.
n8n free or paid, what to remember about pricing
n8n offers the best value for money of any automation tool on the market for a European small or mid-sized company, and that holds whether you pick the free edition or the cloud. Execution-based billing makes the budget predictable, the self-hosted edition removes the licence, and the European location of the cloud clears part of the compliance objections.
The real mistake is to reason on the "licence" line alone. An automation project costs the price of its subscription, plus its infrastructure, plus its model calls, plus the hours of its teams. The advertised price is visible. The real cost has to be framed.
Tandem prices those four lines before the first workflow, as part of an AI audit, then installs and maintains its clients' instances. The details of that service are on our n8n agency page.
