L'article ci-dessous est publié en anglais. Chaque chiffre qu'il contient est le chiffre publié par l'éditeur lui-même, avec la date à laquelle nous l'avons lu.
n8n is one of the two automation platforms in AI Agents, and the buying question it answers is unlike the one next door: this category's other profile is Zapier, which meters tasks. n8n meters executions, and the whole review is about what that choice commits you to, because the vendor splits its own product in two. Cloud is a subscription. Self-hosted is a license. The two have different prices, different meters, and different legal boundaries.
Cloud or self-host is the purchase decision
The pricing page makes the split explicit in the first two paragraphs: pricing is based on monthly workflow executions regardless of complexity, and every plan includes unlimited users, workflows and steps. The cards print Starter at 24€, Pro at 60€ and Business at 800€ a month with the Monthly toggle selected, and the same cards print 20€, 50€ and 667€ a month when the page defaults to Annually (Save 17%). The currency is euros, and the page does not offer a currency switcher, so a buyer outside Europe converts those figures themselves.
The Business card, which prints 800€ a month, carries the line Self-hosted only, and the pricing FAQ confirms it in plainer words: the Business plan is not yet available on n8n Cloud. Enterprise is the current hosted tier above Pro: its card reads Contact Sales and lists both Hosted by n8n and Self-hosted. The order of the decision is therefore fixed by the vendor: if you want Business features hosted by n8n, that option does not exist yet, and if you want them self-hosted you are subscribing to a license key for software you run.
The meter is the whole run, and it has published edges
The definition is one sentence on the pricing page: an execution is a single run of your entire workflow, and it does not matter how many steps are in the workflow or how much data it processes, it is still a single execution. The cards repeat the shape as workflow executions with unlimited steps. Starter allows 2.5k executions a month, Pro 10k, Business 40k, and Enterprise a custom number through sales. The Cloud docs break the Pro tier into Pro-1 (10k executions) and Pro-2 (50k executions) with different memory, which is one more shape of the same execution meter rather than a different plan family.
The edges are where the meter earns its honesty. Saved executions are capped separately from the meter: max saved executions of 2.5k with 2.5 GB of storage and 7 days of log retention on Starter, 25k with 25 GB and 30 days on Pro, 50k with 50 GB and unlimited days on Enterprise, and reaching a saved-execution limit does not stop workflows from running. Concurrency is its own cap: 5 concurrent executions on Starter, 20 on Pro, 200+ on Enterprise. And since n8n 2.38, released September 1, 2026, runs of error workflows no longer count toward the execution quota on any plan, and the changelog notes that this was the reason some people previously left error workflows out.
The credits are a second meter
AI Assistant is a preview feature included on all n8n Cloud plans except Enterprise, and the pricing FAQ gives it a monthly allowance in credits: 2,300 on Starter and 5,700 or 13,700 on Pro depending on the plan size. That range is what the Pro card prints as the headline Up to 13,700 AI credits/month, and the two figures are about plan size rather than a monthly choice. The consumption side is vague on purpose: credits are consumed as you use AI Assistant, and how much depends on the size and complexity of what you are building, with no published cost per credit.
The FAQ answers the two budget questions in the negative: unused credits do not roll over, and it is not yet possible to buy additional credits. Self-hosted n8n runs the same assistant in preview with a bring-your-own-key model, and the FAQ says no credit limits apply there. On Cloud there is a separate prepaid balance, Gateway credits, that pays for models and tool services used by workflow nodes; it is available on Starter and Pro, not on Enterprise and not on self-hosted, and it is kept apart from AI Assistant credits in the docs' own table.
The free option is a license, and the boundary is written down
The self-hosted Community edition is free and its terms are the Sustainable Use License, which n8n's own license FAQ says is officially the Sustainable Use License (SUL). The repository's LICENSE.md, read today, grants a non-exclusive license to use, copy, distribute and modify the software, with a limitation that carries the weight: you may use or modify the software only for your own internal business purposes or for non-commercial or personal use. The same file separates the .ee. source files and directories under the n8n Enterprise License, which is a different document.
The open source question is answered by the vendor in its own FAQ: because open source licenses, under the Open Source Initiative's definition, cannot include limitations on use, n8n does not call itself open source, and it describes the model as fair-code. The license FAQ then draws the practical boundary. You can build automations for your clients on your instance as long as your clients cannot create or edit them, charge for workflow creation, setup and maintenance, and charge for training and consultancy. You cannot host n8n as a service where your clients build workflows, or let external end users build or configure workflows through your product. There is no published revenue or user-count threshold anywhere in the FAQ; the boundary is written in behavior, and n8n invites license@n8n.io questions when a use case is not clearly covered.
The self-hosted subscription is a license key, not a download
Host your own n8n and the docs say it runs as the free Community edition without a license key. What a paid Business or Enterprise subscription buys is a license key that unlocks the features of that edition, and one FAQ entry states that a key can be applied to an unlimited number of instances, with the combined usage from all of them counting toward the quota.
The telemetry half is worth reading before you self-host with a business in mind. The docs on monitoring say the license key must ping the n8n license server daily to stay active, and the ping includes the number of production executions. The same page says telemetry data is collected by default but can be disabled. Neither statement is a criticism; both are the visible price of the daily ping model, and they are published in the FAQ rather than buried.
Hosted data and the exit
The security answer is one paragraph: hosted plans store data within the EU, on servers located in Frankfurt, Germany, and self-hosted plans store it wherever you decide to host n8n. The docs also publish outbound IP ranges for Cloud, on shared infrastructure, with the caveat that an allowlist alone does not prove a request came from your instance, and that inbound webhook traffic does not use those addresses.
Portability is documented in two horizons. During a trial, you have 90 days after it ends to download your workflows, and the docs add the sharp edge: after that, all workflows are permanently deleted and are unrecoverable, and the trial page says n8n deletes your workspace when the trial expires. On a paid account, the Self-Serve Terms say Customer Content is usually deleted six months after account deactivation, and that n8n contacts you by email 30 days before deactivating and deleting. The download path for Cloud runs through the admin dashboard Export tab, and the docs' export page prints the workflow file contents: exported workflow JSON includes credential names and IDs, and the CLI has separate export commands for workflows and credentials.
What the score rests on
The five components and their weights are published on our methodology page, so here is what this review put under each of them.
Fit for the stated job, 30 percent. n8n is an automation platform with AI nodes, and its answer to the buying question is legible: one meter for the run, a second for the assistant, a documented license for self-host, and a FAQ that answers quota and data questions in writing. Held back on the hosting axis, where the Business plan, at 800€ with SSO and environments, is self-hosted only, so the buyer who wants that feature set hosted by n8n has no published option.
Cost to get started, 25 percent. The entry is a 14-day trial with Pro features and a hard 1,000-execution ceiling, and the paid entry is 24€ a month for Starter, the figure the monthly toggle prints, with the annual toggle's 20€ and the vendor's own Save 17% label quoted from the page rather than recomputed. The score is stronger because the meter is predictable: step count and data volume do not change the price. It is docked because the assistant's consumption is unspecified, because the only overage figure published applies to Business, because the page prints no currency switcher, and because the trial ceilings are not on the cards.
What the vendor commits to, 20 percent. A pricing FAQ and docs that answer quota behavior, credit rules, license boundaries, hosting region and deletion windows in writing, plus a changelog entry that removed error workflows from the quota. Docked because saving executions, concurrency and credits are three separate limits a buyer must assemble into one budget, and two of them have no published conversion to cost.
Portability, 15 percent. Ninety days of workflow download after a trial, a six-month data deletion window on a paid account, and a self-host edition whose workflows and data are yours by definition. Docked because the trial deletion window is a hard stop rather than a grace, because the fair-code boundary limits how far someone else's portability tooling may serve clients, and because moving credentials between instances is a separate CLI step from moving workflows.
Documentation quality, 10 percent. The docs are thorough and dated, with a changelog that names versions and release dates, a community license FAQ that answers the open source question in one place, and a pricing FAQ that publishes quota and overage behavior. Docked because several docs URLs in the footer and the legal pages no longer resolve, and because the credit pricing and the savings math live in prose across two pages rather than in one table.
Weighted at the published weights, that comes to 8.3.
What we could not settle
There is no score here for running the product. We did not install the Community edition, we did not run workflows through either edition, and we did not measure build time, stability or output quality, so none of those appear in the score. What this review compares is what n8n publishes: prices, meters, credits, license text and docs, each read today and each listed with its date in the ledger.
We also did not settle how many AI credits a realistic month of assistant use consumes, because the pricing FAQ describes consumption in terms of size and complexity without a number. There is no published cost per credit from the pages this review read. None of that is a criticism stated as absence; it is a list of figures the vendor does not print, and the review says so where it matters.
The n8n tool profile carries the same figures with the same dates, the AI Agents hub is where the category comparing n8n with Zapier lives, and our methodology is where the 30/25/20/15/10 weights are published.