n8n Workflow Automation: Build AI Agents Without Code

n8n Workflow Automation: Build AI Agents Without Code

Every office runs on small, repetitive jobs. A new order arrives and someone copies it into a spreadsheet. A customer writes in and someone pastes the same reply. A form is filled in and someone forwards it to three different tools.

n8n is a tool that takes those jobs away. You draw your work as a flow chart: when this happens, do that, then do the next thing, then tell someone it is done. Each step is a small block on a canvas, and the blocks are connected by lines. When the flow runs, data moves along the lines from one block to the next, and the whole job happens without anyone touching it.

The numbers behind it are large. The repository listed as n8n-io/n8n has about 211,000 stars as of 16 November 2026, making it one of the most-starred automation projects in the world, and it ships a new release almost every day — the newest, n8n@2.47.7, arrived on 13 November 2026. But one fact must be understood before anything else: n8n is fair-code, not open source. Its source code is public to read, but the Sustainable Use License v1.0 limits how you may use it, and you cannot host n8n and sell access to it without a commercial agreement.



What Is n8n, in Plain Words?

n8n — pronounced "n-eight-n" — is a workflow automation platform. A workflow is a chain of steps that runs automatically: a trigger starts it, and each step, called a node, does one job. A trigger might be a new email, a new row in a spreadsheet, a message in a chat app, a form submission, or a schedule that fires every morning at nine. The nodes in between can read data, change it, send it somewhere else, or ask an AI model a question. The last node usually delivers the result: an email sent, a record updated, a message posted, a file saved.

Here is a simple example. A small shop receives orders through a web form. In n8n, the form submission is the trigger. The next node checks whether the customer already exists in the shop's records. If not, a node creates the customer. Five nodes, one flow, and nobody in the shop ever copies an order by hand again.

That is the whole idea, and it is why n8n has grown so fast. It sits between two older kinds of tool. On one side are simple connectors that move data from one app to another but cannot make decisions. On the other side are programming frameworks that can do anything but need a developer for every change. n8n gives non-programmers a canvas they can understand and programmers a tool they can extend with code when a node is not enough.

The second reason for its growth is the AI agent builder. Since 2025, and more fully in 2026, n8n has added nodes that turn a workflow into an agent: a chat trigger receives a question, a chat model node thinks about it, a memory node remembers the conversation, and tool nodes let the agent look things up, send messages and update records.

The Licence Warning: Fair-Code, Not Open Source

This is the most important section of this post, and the one most other introductions skip. n8n is fair-code, not open source.

Fair-code means the source code is public and you can read it, download it and run it yourself. It does not mean the code is open source under the Open Source Definition. n8n is released under the Sustainable Use License v1.0. That licence lets you use and modify n8n for your own internal business purposes, or for non-commercial and personal use. The restriction is the part that matters: you cannot host n8n and sell access to it, and you cannot let your customers connect their own accounts and build workflows inside your own product. If you want to do either of those things, you need a commercial agreement with n8n.

Put plainly, three uses are fine and one is not. Running n8n on your own server to automate your own company's work is fine. Running it at home for personal projects is fine. Giving it away free to others for non-commercial use is fine. But building a paid service on top of it — a platform where your customers log in, connect their apps and run their own workflows, while you charge them — is not allowed under the free licence.

Two smaller details are worth knowing. First, files with ".ee." in the name need an Enterprise licence; they are the commercial-only parts of the product. Second, commercial consulting work does not need a separate agreement: an agency that builds workflows for clients using n8n, and charges for its time and expertise rather than for access to n8n itself, is covered by the licence, according to n8n's own frequently asked questions.

One more behaviour follows from the licence and surprises new users: telemetry is on by default. n8n sends back anonymous usage information, including the shape of every workflow saved — which nodes you used and how they connect, not the data inside them.

How the Visual Canvas Actually Works

Open n8n and you see a blank canvas with a small menu of nodes. You drag a trigger onto the canvas, then drag the next node beside it, and a line connects them. Click a node and a panel opens where you set it up: which account to use, which fields to read, what to do with the result. Run the workflow once and you can watch the data move along the lines, node by node, with the output of each step shown on screen.

Every workflow has the same shape: a trigger, one or more actions, and an output. Common triggers are webhooks (a web address other apps can call), schedules, new emails, new database rows, new form submissions, and chat messages.

The nodes in the middle are where the work happens. There are nodes for reading and writing spreadsheets, databases and files; nodes for sending email, chat messages and notifications; nodes for calling other apps; nodes for transforming data from one shape into another; and nodes that run small pieces of JavaScript code when nothing else fits. The public template library holds 14,216 ready-made workflows you can import and change.

Executions are the unit everything is counted in. One execution is one full run of a workflow, no matter how many nodes it has. A workflow that runs a hundred times a day uses a hundred executions; a workflow with fifty nodes that runs once uses one.



The AI Agent Builder: When the Flow Starts Thinking

The AI Agent node is the part of n8n that changed most in 2026. In an agent workflow, you give the agent a goal, a set of tools, and a chat model — and the agent decides the steps itself, one at a time, until the job is done or it runs out of turns.

A typical agent workflow has four parts. The chat trigger receives the user's message. The chat model node — connected to a model from OpenAI, Anthropic, Google or a local model — does the thinking. The memory node remembers the conversation so the agent does not ask the same question twice. And the tool nodes give the agent hands: a search tool, a calendar tool, a database tool, an email tool, or a connection to another n8n workflow.

This is powerful, and it is also where costs hide. Every step the agent takes calls the model again, and model providers charge per token. A simple question can turn into ten model calls if the agent keeps checking tools. On the Cloud plans, the AI assistant credits are limited — 1,600 a month on Starter, 4,000 on Pro — and on a self-hosted setup you pay the model provider directly.

The agent builder also connects to MCP, the Model Context Protocol. An MCP client tool lets an n8n agent use tools from any server that speaks MCP, so the agent is not limited to n8n's own nodes.

Integrations: 400+ Apps, and a Number to Read Carefully

n8n connects to more than 400 applications, according to its own integration list — Google Sheets, Slack, Gmail, Notion, Airtable, Stripe, Telegram, Discord, HubSpot, Salesforce, Postgres and MySQL, and hundreds more. The project's main description claims more than 1,500 integrations, and both figures are honest in their own way: the smaller number counts the built-in nodes, and the larger number counts every app reachable through those nodes and generic connections. This post uses the 400+ figure because it is the one you can count on the integration list, and it says so plainly rather than merging the two numbers into one.

One integration is worth a special mention for readers of this blog. n8n can call a local AI model through the approach in the Ollama post published on Tech Know on 13 November 2026 — the same local-models idea that lets you run a model on your own computer instead of paying a cloud provider per token. Pairing n8n with a local model is one of the cheapest ways to run an AI agent: the automation is free under the Community Edition, and the model costs only electricity.



 

What Does n8n Cost? The Full Price Picture

Pricing is where most people make their first mistake with n8n, because the free option and the paid options answer different questions. The self-hosted Community Edition costs nothing for the licence and has no limit on executions or users — you pay only for the server it runs on, which can be a five-dollar virtual machine for light use. The catch is that you are the operator: you install it, update it, back it up, keep it secure, and fix it when it stops.

PlanPrice per month (billed annually)Executions per monthWhere it runsWho it fits
Community EditionFreeUnlimited (limited by your server)Your own serverTechnical users, testing, internal use
Starter (Cloud)202,500n8n CloudA first department workflow
Pro (Cloud)5010,000n8n CloudSeveral teams, workflows in production
Pro, larger tier (Cloud)About 12050,000n8n CloudMid-volume without a sysadmin
Business (self-hosted)800 (667 euro on the euro page)40,000Your own serverCompanies needing SSO, Git and environments
EnterpriseCustom quoteCustomCloud or self-hostedLarge organisations

The prices above are from n8n's pricing page as checked on 2 November and again on 10 November 2026 by independent pricing guides, and they had not changed between those two checks. Two details deserve attention. First, the price you see depends on where you open the page: n8n shows dollars to visitors in the United States and euros to visitors in Europe, and the numbers are the same in both currencies for Starter and Pro, while Business is 800 dollars against 667 euro when billed annually. Second, there is no free Cloud plan, only a trial. Paying monthly instead of annually costs about 17 percent more: Starter is 24 a month, Pro is 60, and Business is 960.

On Starter and Pro there are no overage fees — when you reach the limit, runs fail until the month resets, which is kinder than a surprise bill but can stop a business process on the last day of the month. The hidden cost is not in the table at all: independent estimates put the real cost of self-hosting at roughly 1,272 to 4,560 euro a year once you count the server and the maintenance hours.

Self-Hosted or Cloud: Which One Should You Pick?

The decision comes down to four questions. First, can your data leave your building? If you handle health records, financial data or customer data under a contract that forbids third-party processing, self-hosting is the only option — Cloud runs on Microsoft Azure in the European Union, which satisfies many companies but not all. Second, how much do you want to operate? Cloud removes backups, updates and uptime from your list. Third, how many executions will you really run? Fourth, do you need the features that only paid plans have? The Community Edition has no projects, no sharing of workflows and credentials, no single sign-on, no environments, no Git version control, no custom variables, no external secret stores and no log streaming.

Installing the Community Edition is deliberately simple. The standard route is Docker: one container, running on port 5678, with SQLite as the default database and PostgreSQL as the option for heavier use. The command is short, the container starts in under a minute, and the first screen asks you to create an owner account. In January 2026 a severe flaw known as Ni8mare, rated 10 out of 10, let attackers take over certain exposed instances through webhooks without logging in, and the fix arrived in version 1.121.0. The lesson is permanent: put n8n behind a password, keep it updated, and never expose the editor to the public internet without protection.

Updating matters more with n8n than with most tools because it ships so often — 237 stable releases in the past year, a new release on most days, and a security update on 13 November 2026 that fixed five vulnerabilities. The minimum patched versions from that update were 1.123.96 on the old line, 2.47.6 on the current line, 2.48.3 on the beta line and 3.2.2 on the version-3 line. If you self-host, put the update check on a weekly calendar, and treat a security update as a same-day job.

n8n Against Zapier and Make: An Honest Comparison

Most readers will compare n8n with Zapier and Make, the two best-known hosted automation tools. All three can automate the same office jobs, and the differences are about price shape, control and who does the operating.

n8nZapierMake
Cheapest paid plan (annual billing)20 a month, 2,500 executionsAbout 20 dollars a month, 750 tasks9 dollars a month, 10,000 operations
How usage is countedPer execution (one full workflow run)Per task (one step that succeeds)Per operation (one step that runs)
Free self-hosted optionYes, Community EditionNoNo
AI agents on the canvasYes, AI Agent node and Agents builderYes, in newer plansYes, in newer plans
Data stays on your serverYes, if self-hostedNoNo
Learning curveSteeper, more powerfulEasiest to startMiddle ground

The counting unit is the part that decides the bill. n8n charges per execution, so a fifty-step workflow that runs once costs one execution. Zapier charges per task, so the same workflow would cost up to fifty tasks. That makes n8n dramatically cheaper for long, complicated workflows — and roughly equal for short ones.

There is also a difference in who the tool is for. Zapier is the easiest to start: pick two apps, pick a trigger and an action, and you are done in five minutes. Make sits in the middle, with a visual canvas like n8n but a simpler model. n8n is the most powerful and the hardest to learn, because it assumes you are comfortable with concepts like webhooks, JSON data and application programming interfaces.

The Company Behind n8n

n8n was founded in Berlin by Jan Oberhauser, and the public repository was created on 22 June 2019. It grew the slow way first — developers recommending it to each other — and then very fast: stars rose from about 160,000 at the 2.0 release in December 2025 to over 208,000 by October 2026, and forum members grew from about 115,000 to over 183,000 in the same period, according to the company's own figures. In October 2025 n8n raised a Series C round led by Accel — 180 million dollars at a 2.5 billion dollar valuation — and the company says users grew six times and revenue ten times in the year before that round. In May 2026 SAP invested at a reported 5.2 billion dollar valuation (reported, not confirmed by n8n) and began embedding n8n in its Joule Studio agent builder.

Who Should Use n8n — and Who Should Not

n8n is a strong fit for a technical team that automates its own work and wants its data on its own server; for a small business with one technical person who can own the system; for a developer building AI agents that need to touch real tools and data; and for anyone who has outgrown simple two-app connectors and needs branching, code steps and error handling.

It is a poor fit for a team with no technical person at all, because someone has to install it, update it and fix it when it breaks — and on the Cloud plans that someone is n8n, but the workflows themselves still need a person who understands data and logic. It is a poor fit for anyone who wants to resell automation as a service, because the licence forbids it.

Readers who liked the self-hosted AI theme of this post should also read the ComfyUI post published on Tech Know on 16 November 2026. ComfyUI is the node-based studio for AI images, video and audio, and it carries its own licence condition — GPL-3.0, which is open source but copyleft — The Ollama post from 13 November 2026 completes the set, explaining how to run the models themselves on your own computer.

Frequently Asked Questions

Is n8n open source?
No. n8n is fair-code, not open source. Its source code is public under the Sustainable Use License v1.0, which allows internal business use, personal use and free non-commercial distribution, but does not allow you to host n8n and sell access to it.

Is n8n really free?
The self-hosted Community Edition is free, with unlimited executions and users, under the Sustainable Use License. n8n Cloud starts at 20 a month for 2,500 executions, and there is no free Cloud plan, only a trial.

What is the latest version of n8n?
As of 16 November 2026, the newest release is n8n@2.47.7, released on 13 November 2026. Note that earlier the same week the 2.47 line was still the beta track and the stable release was 2.46.7, dated 9 November, so the line was promoted only around 12-13 November. Version 3.0 is a pre-release, not the latest stable version, and its stable release is expected in early December 2026. Support for the 1.x line ends on 31 December 2026.

Can I sell a service built on n8n?
Not under the free licence. You cannot host n8n and sell access to it, or let your customers build workflows inside your product. You need a commercial agreement with n8n for that. Consulting work for clients, charged by time, does not need one, according to n8n.

How many apps does n8n connect to?
More than 400 built-in integrations, according to n8n's own list. The project description claims more than 1,500, counting apps reachable through generic connections.

Is n8n better than Zapier or Make?
It depends on your workflows. n8n is cheaper for long, multi-step workflows because it charges per execution, not per step, and it is the only one of the three you can self-host for free. Zapier is easier to start, and Make is cheaper for short workflows at low volume.

Can n8n run AI agents?
Yes. The AI Agent node, with a chat model, memory and tool nodes, lets a workflow decide its own steps, and the Agents builder added in August 2026 adds version history and publication. Watch the model costs: every agent step is another call to the model, and the Cloud plans limit AI assistant credits.

Is it safe to self-host n8n?
Yes, if you keep it updated and do not expose it carelessly. Put it behind a strong password, keep it behind a reverse proxy, update it weekly, and apply security updates the same day. Cloud instances are patched automatically; self-hosted ones are not.

The Bottom Line

n8n is one of the most capable automation tools available in late 2026: a visual canvas anyone can learn, an AI agent builder that turns workflows into helpers that think, more than 400 integrations, a free self-hosted edition, and a community of hundreds of thousands of users shipping templates and answering questions.

But the tool comes with a condition that must be understood before the first workflow, not after the hundredth: it is fair-code, not open source, and the Sustainable Use License draws a firm line around selling access to it. Respect that line, pick Cloud if you want someone else to operate it and self-hosting if you want control of your data, count your executions before you pick a tier, and keep the software updated.

By Umar Farooq, Tech Know.

Post a Comment

0 Comments