n8n is a workflow automation platform that connects over 500 services and orchestrates AI agents, offered both as an EU-hosted cloud service and as source-available software you can run yourself. It targets technical teams who want visual building with the option to drop into code, and it is licensed under the Sustainable Use License rather than a standard open source licence.
Used this tool? Rate it
Used this tool? Rate it
n8n is a workflow automation platform. You build processes as a visual graph of connected nodes — a trigger fires, data flows through steps that transform it or call external services, and something happens at the end. What distinguishes it from the many tools that do roughly this is a combination of three things: it targets technically capable users rather than hiding code from them, it can be run on your own infrastructure, and it has repositioned around AI agents.
That last shift is visible in the current headline, which reads "AI agents and workflows you can see and control", with the supporting line describing an approach where you build visually, go deep with code, and connect to anything. The emphasis on seeing and controlling is a deliberate contrast with agent frameworks that operate as opaque black boxes.
The operating entity is disclosed precisely in the privacy policy: n8n GmbH, registered as HRB 212509 B with a registered address at Novalisstr. 10, 10115 Berlin, a German company subject to EU jurisdiction. TechCrunch reports the company was founded in 2019 by Jan Oberhauser, who serves as CEO. Funding has been substantial: a €55 million round led by Highland Europe in March 2025, followed by a $180 million Series C led by Accel at a $2.5 billion valuation, with NVIDIA's venture arm among the participants.
Vendor-reported scale figures include over 500 integrations, more than 10,000 workflow templates and 200k+ community members, plus a GitHub star count above 200,000. The homepage also cites a G2 score of 4.7 out of 5. These are self-reported on the vendor's own pages and should be read as marketing rather than audited metrics — the G2 figure in particular is the vendor quoting a third party, not a rating verified here.
This section exists because n8n is very widely described as open source, and that description is inaccurate in a way that matters legally.
n8n is licensed under the Sustainable Use License, not under MIT, Apache or any other OSI-approved open source licence. The authoritative text is the LICENSE.md file in the project's own repository, and its restrictions clause is explicit: "You may use or modify the software only for your own internal business purposes or for non-commercial or personal use." Distribution to others is permitted only free of charge and for non-commercial purposes.
In practice this means running n8n internally to automate your own company's processes is clearly permitted. Taking n8n, hosting it, and selling that hosted service to customers is not. If your business model involves offering n8n-based automation as a paid service to third parties, you need to read the licence carefully rather than assuming open source freedoms.
There is a second layer. Source files containing ".ee." in the filename or ".ee" in the directory name are excluded from the Sustainable Use License entirely — for those "you must hold a valid n8n Enterprise License". So the repository is not uniformly licensed; parts of it are commercial-only.
The cloud terms carry a parallel restriction on the hosted service: you can only use the Cloud Services for internal business purposes only and may not transfer, sell, distribute, lease, sublease, assign or licence to any third parties. So the same internal-use boundary applies whether you self-host the software or subscribe to the hosted product.
The ecosystem term for this model is fair-code, and its own definition site is refreshingly direct: "Fair-code is not a software license." It describes a model where software is generally free to use, has openly available source, can be extended by anyone, and is commercially restricted by its authors. That is a coherent and defensible position — but it is not open source, and conflating the two can lead an organisation into a licensing problem.
The practical takeaway is that "source-available and self-hostable" is accurate, while "open source" is not. For most internal users the distinction changes nothing. For anyone building a commercial offering on top, it changes everything.
The core interaction is a node canvas: triggers, actions, conditional branches and loops assembled visually. The differentiator is that when the visual abstraction becomes limiting, you can write JavaScript or Python in a code node rather than abandoning the tool. This is why the product describes itself as offering the flexibility of code with the speed of no-code, and why it appeals to people who find purely no-code tools constraining.
Over 500 integrations cover the usual SaaS surface — CRM, databases, messaging, spreadsheets, cloud storage — plus generic HTTP request nodes that let you call any REST API that lacks a dedicated node. In practice the generic HTTP node matters as much as the count, since it means the absence of a specific integration is rarely a hard blocker.
This is where the product has changed most. LangChain integration arrived in October 2023, and an AI Workflow Builder that constructs automations from natural language followed in October 2025. Agents can be given tools, memory and branching logic within the same visual canvas used for conventional automation — which is the actual argument for using n8n for agents rather than a code framework, since the agent's decisions and tool calls remain inspectable.
More than 10,000 published workflow templates function as both a learning resource and a starting point. For a tool with a genuine learning curve, this library is a meaningful part of the value.
The same software runs as n8n's EU-hosted cloud or on your own servers, with dedicated self-hosting documentation. Very few competitors offer a credible self-hosted path, and for teams with data residency or air-gap requirements this is often the deciding factor.
Internal business process automation is the bread and butter: syncing records between systems, routing approvals, generating reports, pushing notifications when conditions are met.
Data movement and transformation between systems that lack native integration is a classic fit, particularly where the transformation logic is too complex for simpler tools but does not justify a bespoke service.
AI agent deployment is the fastest-growing use. Teams building agents that call tools, query internal data and act on results get an inspectable execution graph rather than opaque framework logs.
Enrichment and lead workflows in sales and marketing — capturing an inbound record, enriching it from external APIs, scoring it and routing it — map naturally onto the node model.
DevOps and internal tooling suits the self-hosted option well: alert routing, scheduled maintenance jobs and incident notification pipelines that must run inside your own infrastructure.
Conversely, n8n is a poor fit for teams with no technical capability whatsoever, for use cases needing sub-second latency guarantees, and for anyone wanting to resell n8n-powered automation as a service without addressing the licence.
Exploit the whole-workflow execution model. Because a run costs the same regardless of how many steps it contains, consolidating related logic into one workflow is dramatically cheaper on cloud plans than splitting it across several that each trigger separately. This single fact is the most useful cost lever the platform offers.
Beware high-frequency triggers. A workflow polling every minute consumes roughly 43,200 executions a month on its own — more than the Business tier's allowance. Use webhooks instead of polling wherever the source system supports them.
Model your real volume before choosing a tier. Compare the annual cost of a cloud plan against the cost of a small server plus the engineering time to maintain it; the answer varies enormously by team.
Read the licence before building a product on it. This is worth repeating in the practices section because the mistake is usually made early and discovered late.
Version-control exported workflows. Workflows are JSON and can live in Git, which gives you review, rollback and change history that the UI alone does not.
Store credentials in the credential system rather than hardcoding them into nodes, and keep least-privilege scopes on connected accounts.
Test with production-shaped data. Most automation failures come from data variation the builder did not anticipate rather than from logic errors.
Technical teams are the core audience — developers, ops engineers and technically-minded operations staff who want to move faster than writing bespoke integrations but do not want a tool that hides the mechanics.
Organisations with data residency or privacy constraints have an unusually strong reason to choose it, since self-hosting keeps workflow data entirely inside their own infrastructure and the cloud offering is EU-hosted by default.
Teams building AI agents who want observability over their agent's behaviour are the fastest-growing segment.
Cost-sensitive high-volume users benefit from the self-hosted path, where per-execution billing does not apply.
It is a weaker fit for non-technical business users who would be better served by simpler consumer automation tools, for teams with no capacity to maintain infrastructure who also cannot afford cloud tiers, and — critically — for anyone wanting to build a commercial resale offering without a compatible licence arrangement.
n8n runs as a web application, either on n8n's cloud or on infrastructure you control. Cloud data is stored within the EU, on servers located in Frankfurt, Germany, which is a concrete commitment rather than a vague assurance and matters for GDPR-conscious buyers.
Self-hosting is supported through a dedicated self-hosting documentation site at docs.n8n.io/hosting, with the usual deployment routes for a Node.js application including Docker. Self-hosting is available broadly, while the cloud plans reserve certain self-hosted enterprise arrangements for Business and Enterprise tiers.
Connectivity is the real platform story: 500+ integrations plus generic HTTP nodes, webhook endpoints for inbound triggers, and an API for programmatic control. The footer also exposes the enterprise track and security documentation.
Cloud pricing is tiered by execution volume, and understanding what an execution means is essential — the definition is unusually favourable and is stated plainly: "An execution is a single run of your entire workflow. It doesn't matter how many steps are in the workflow or how much data it processes."
The published annual-billing rates are Starter at 20€/mo with 2,500 executions, Pro at 50€/mo with 10,000 executions and Business at 667€/mo with 40,000 executions, with Enterprise quoted individually. Note the very large jump between Pro and Business, which reflects Business being positioned as an organisational rather than team purchase.
Other limits scale with tier: 5 concurrent executions on Starter and 20 on Pro, with 200+ on Enterprise; execution history retention of 7 days on Starter and Pro, 30 days on Business and 365 days on Enterprise. Importantly, there are unlimited active workflows on every tier, so you are constrained by runs rather than by how many automations you build.
Self-hosting changes the economics entirely: there is no per-execution charge, and your cost is infrastructure plus maintenance. For high-volume workloads this is frequently far cheaper, provided you have the capability to run it — and provided your use stays within the licence.
Zapier is the mainstream comparison: far easier for non-technical users, with a larger integration catalogue, but priced per task rather than per workflow run, which becomes expensive at volume, and with no self-hosting.
Make (formerly Integromat) sits closer to n8n in visual sophistication and is generally cheaper than Zapier, though it is also cloud-only.
Activepieces and Windmill are the closest alternatives for teams that specifically want self-hosting, with Activepieces notable for using a more permissive licence.
Temporal and Airflow address a different problem — code-first durable execution and data orchestration respectively — and are better choices when the requirement is engineering-grade reliability rather than accessible automation.
Writing custom code remains a legitimate alternative for a small number of stable, high-value integrations, where a workflow platform adds a dependency without adding much.
The licence is the most consequential limitation, and its practical shape is worth restating: internal use is fine, commercial resale is not, and the enterprise-only .ee files mean the repository is not uniformly licensed. Teams that assumed open source freedoms can find this out uncomfortably late.
There is a genuine learning curve. The product is explicitly aimed at technical people, and non-technical users often struggle with data structure mapping, expression syntax and debugging — the same depth that makes it powerful.
Cloud execution costs can escalate unexpectedly, mostly through high-frequency triggers rather than workflow complexity. The whole-workflow billing model is generous per run, which makes it easy to underestimate how many runs a polling trigger generates.
Self-hosting shifts rather than removes cost. You take on updates, database maintenance, monitoring and backup — real engineering time that should be counted when comparing against cloud pricing.
Integration depth varies. Five hundred integrations is a large catalogue, but individual nodes differ in how completely they cover the underlying API; the generic HTTP node is the fallback and requires reading vendor API docs.
Automation is fragile by nature. Upstream API changes, credential expiry and rate limits break workflows, and without deliberate error handling failures can go unnoticed.
On third-party validation, evidence is positive but limited in volume: a 4.8 out of 5 rating from 73 reviews on Product Hunt. That sample is small enough that it should be quoted with its size attached. The G2 figure that appears on the homepage is the vendor citing a third party and was not independently verified here.
For an automation tool this section matters more than usual, because workflows routinely carry an organisation's most sensitive data between systems.
The disclosed collection scope is proportionate. n8n collects contact data, account and usage data including "pseudonymized User and workspace identifiers, account settings, authentication events, workflow configuration", and traffic and device data. Note that workflow configuration is collected — the structure of your automations — which is distinct from the payloads flowing through them.
The strongest commitment concerns AI training, and it is unusually clear: n8n states it does not use personal data processed in connection with the website or services "for developing, improving, or training AI and/or ML models." At a moment when most platforms reserve broad training rights, an explicit exclusion is a meaningful differentiator for enterprise buyers.
A qualification belongs alongside it. The policy also states that "We may derive de-identified data sets from your Customer Content" for service improvement. De-identified derivation is a narrower permission than model training on personal data, but organisations with strict contractual confidentiality obligations should read both statements together rather than only the reassuring one.
GDPR rights are fully enumerated — withdrawal of consent, access, correction, deletion, portability and objection — which is expected from a German controller but still worth confirming. Subprocessors are not listed in the policy body; a separate subprocessors page is referenced.
Two structural points strengthen the privacy position. Cloud data resides in Frankfurt within the EU, and self-hosting removes n8n from the data path for workflow payloads entirely. For organisations handling regulated data, that second option is the strongest privacy argument the product has.
Not in the OSI sense, and this is the most common misconception about it. n8n uses the Sustainable Use License, under which you may use or modify the software only for internal business purposes or non-commercial use, and may distribute it only free of charge for non-commercial purposes. The source is publicly available and self-hosting is permitted, so "source-available" and "fair-code" are accurate descriptions — but MIT or Apache-style open source freedoms do not apply. Files marked with .ee are excluded entirely and require an Enterprise Licence.
Running it internally to automate your own business is permitted. Reselling n8n-based automation as a hosted service to third parties is not covered by the Sustainable Use License and requires a separate commercial arrangement. If your plan involves offering n8n to customers, read the licence text before building.
Cloud plans billed annually are 20€ per month for Starter with 2,500 executions, 50€ for Pro with 10,000 executions and 667€ for Business with 40,000 executions; Enterprise is quoted individually. Self-hosting has no per-execution fee — you pay for infrastructure and maintenance instead.
One complete run of one workflow, regardless of how many nodes it contains or how much data passes through. A twenty-step workflow processing a large dataset costs the same single execution as a two-step one. This makes consolidating logic into fewer, larger workflows the main cost optimisation on cloud plans.
It can, and this is the most common billing surprise. A workflow polling once a minute uses roughly 43,200 executions per month by itself, exceeding even the Business allowance. Prefer webhook triggers over polling wherever the upstream service supports them, and lengthen intervals where you can.
The privacy policy explicitly states that personal data processed in connection with the website or services is not used for developing, improving or training AI or ML models. The policy separately notes that de-identified data sets may be derived from customer content for service improvement. If confidentiality obligations are strict, self-hosting removes n8n from the data path altogether.
Cloud data is held within the EU on servers in Frankfurt, Germany, and the controlling entity is a German company, n8n GmbH in Berlin. Full GDPR rights are enumerated in the privacy policy.
Only partly. The visual builder is approachable, but the product deliberately targets technical people and the depth that makes it powerful — expressions, data mapping, code nodes, debugging — is where non-technical users typically stall. Teams without any technical capacity are usually better served by simpler consumer automation tools.
Both, and agents are now the headline use. LangChain integration arrived in October 2023 and an AI Workflow Builder that generates automations from natural language followed in October 2025. The distinctive benefit is that agent behaviour runs on the same inspectable canvas as ordinary workflows rather than inside an opaque framework.
Choose cloud if you want zero maintenance, are comfortable with EU hosting and your execution volume fits a tier's allowance. Choose self-hosted if you have high volume where per-execution pricing becomes expensive, if data residency or air-gap requirements rule out hosted services, or if you have the engineering capacity to run it. Whichever you pick, confirm your intended use fits the licence.
