n8n is a workflow automation platform that connects applications, databases and AI models inside a single visual flow. It started in Berlin in 2019, built by Jan Oberhauser, and went from a personal project to automation infrastructure running inside companies worldwide.
The name comes from "nodemation", a contraction of "node" and "automation" that the founder found too long to type into a command line. It became n8n: one n, eight letters, another n.
The history is not the interesting part. What matters is what n8n does differently from the tools it competes with.
What It Actually Is
An n8n workflow is a sequence of nodes wired together on a canvas. Each node performs one action: read an email, call an API, run a SQL query, invoke a language model, write to a spreadsheet.
The difference from pure no-code tools is that n8n does not trap you inside its integration catalogue. When no ready-made node exists for the system you need, you use the HTTP Request node and talk to the API directly. When the logic is too specific, you drop a code node into the middle of the flow and write JavaScript or Python.
The catalogue is substantial anyway: the official repository lists more than 1,500 integrations and over 9,000 workflow templates ready to adapt.
Fair-Code: The Detail Almost Everyone Gets Wrong
Plenty of articles describe n8n as open source. It is not, and the distinction has practical consequences.
Since March 2022 the code has been distributed under the Sustainable Use License, a model the company calls fair-code. The source is visible and modifiable, but it carries usage restrictions that a certified open source licence could not impose.
Translated into what matters to a business:
- You can install n8n on your own servers and automate as many internal processes as you like, with no licence fee;
- You can modify the code and build your own nodes;
- You cannot resell n8n as a hosted service to third parties.
An SME automating its own operations will never come near that boundary. A consultancy wanting to host instances for clients is a different conversation, and it runs through an OEM agreement. Worth having legal classify this correctly before anything gets signed.
Running It On Your Own Server Changes the Data Conversation
The Community Edition is free and self-hostable, with no limit on workflows or executions. You pay only for infrastructure, which for light workloads sits somewhere around five to ten euros a month on a modest server.
The real gain, though, is not the price. It is that the data never leaves your infrastructure.
Automate invoice processing and your suppliers' details stay exactly where they already were. Handle customer requests and their contents never cross a third-party cloud. For anyone working under GDPR, with clinical information or with contracts, this stops being a technical preference and becomes a compliance requirement.
The official documentation covers Docker installation and scaling with PostgreSQL and Redis.
Pricing Explains Half the Adoption
Here is the difference that converts into money fastest.
Zapier charges per task. Every step in a flow burns a credit. n8n charges per execution: an entire workflow counts as one unit, whether it has three steps or thirty.
Run the numbers on a concrete case. A ten-step workflow running a thousand times a month:
- Zapier: 10 steps × 1,000 runs = 10,000 tasks consumed
- n8n: 1,000 executions
Same flow, same outcome, orders of magnitude apart on the invoice. n8n Cloud's entry plan sits around €20 a month for 2,500 executions; Zapier's is roughly $20 for 750 tasks.
Self-host and the cost per execution drops to zero.
The honest caveat: for one or two-step automations that advantage disappears and Zapier gets you moving faster. Execution-based pricing pays off once flows carry real logic.
n8n 2.0 and the Agent Layer
Version 2.0, released in January 2026, is what earns the word "reshaping" in this article's title.
n8n stopped being an integration tool with a few AI nodes bolted on and became an agent orchestration platform. It ships around 70 dedicated AI nodes, with native LangChain integration, memory that persists across executions, vector stores for RAG, and support for MCP, the protocol that became the standard way of handing tools to a model.
The practical consequence is this: an Agent node receives a goal and a set of tools, which can be any other n8n node, then works out the sequence of steps by itself, with retries and fallbacks when something fails.
That opens the door to processes rule-based automation never solved. Classifying complaints by substance. Pulling data out of invoices with no fixed layout. Reading contracts. Answering requests that fit no category anyone anticipated.
The pattern that settled in 2026 is not replacing everything with agents. It is hybrid: a deterministic core handles validation, routing and transformation, where reliability is mandatory, and the agent takes the ambiguous parts. Wrapping a deterministic task inside an agent does not make it smarter. It makes it slower, costlier and harder to debug.
Revolution or Hype?
The market has answered. In October 2025 n8n raised $180 million in a Series C led by Accel, with NVIDIA's venture arm participating, at a $2.5 billion valuation. By May 2026 that valuation had climbed considerably further, with SAP joining the cap table.
More telling than the valuation is the growth pattern. The company reported six times the users and ten times the revenue within a single year, with half its base in the United States. This is not developer curiosity. It is enterprise adoption.
What n8n Does Badly
No tool is right for everyone, and this is worth knowing before you invest the time.
The learning curve is real. The node canvas and JSON handling demand more technical comfort than a guided step-by-step interface. A team with nobody technical will move faster elsewhere.
Self-hosting transfers the security burden to you. In January 2026 critical vulnerabilities were patched that had exposed instances to remote takeover. Anyone on the managed version received the fix automatically. Anyone self-hosting who had not updated stayed exposed. Your own server means your own patching routine.
Not every integration exists. The catalogue is large, but a niche SaaS dependency may still mean building the connection by hand against its API.
Where to Start
Install the Community Edition on a small server, or open a cloud account on the free tier. Pick a process you already understand in detail, ideally a boring repetitive one, and rebuild it in n8n.
The first flow takes longer to assemble than doing the job by hand. The tenth takes minutes. That is the point at which the internal conversation about automation stops being about cost and starts being about priorities.
Conclusion
n8n became relevant through a combination that is hard to replicate: visible source, the option to run on your own infrastructure, pricing per execution rather than per task, and an AI agent layer that arrived ahead of the competition.
It is not the easiest tool on the market. It is the one that imposes the fewest limits on anyone who knows what they want to automate.
The useful question is not whether n8n beats Zapier. It is whether your processes carry enough complexity for that difference to pay you back.