n8n (pronounced "nodemation") has become one of the most starred automation tools on GitHub, with 206,000+ stars. It solves a problem that Zapier and Make struggle with: how do you build automation that handles real-world complexity without writing a separate microservice every time you hit an edge case?
This guide covers what n8n is, how it works, concrete workflow examples you can replicate, and when you should choose n8n over the alternatives.
What is n8n?
n8n is a fair-code workflow automation platform. "Fair-code" means the source is available on GitHub and free to self-host, but commercial restrictions apply to hosting it as a service for others. For teams using it internally, it is effectively free with no execution limits.
A workflow in n8n is a directed graph of nodes. Each node either triggers the workflow (a Webhook, a Schedule, a new row in a database) or takes an action (calls an API, transforms data, sends a message). Data flows between nodes as JSON arrays, and every execution is logged and replayable from the UI.
What makes n8n different from Zapier or Make is that anywhere in that graph, you can drop a Code node and write JavaScript or Python. There is no wall where no-code ends and you have to go elsewhere. The visual canvas and the code editor coexist, and switching between them is a single click.
Core concepts: nodes, triggers, and workflows
Trigger nodes start a workflow. Common triggers include an incoming webhook, a cron schedule, a new message in Slack, a new row in a PostgreSQL table, a new email in Gmail, or a manual test run.
Action nodes do something with the data. They read from or write to external systems: send a Slack message, create a HubSpot contact, execute a SQL query, call a REST API, or transform data with JavaScript.
Workflows chain nodes together. Outputs of one node become inputs of the next. Branching, merging, looping, and error handling are built into the canvas. You can test a single node with pinned data without triggering the entire workflow.
Credentials are stored separately from workflows, encrypted at rest, and referenced by name. Sharing a workflow does not expose the underlying API keys or OAuth tokens.
Concrete workflow examples
GitHub issue to Slack alert
A Webhook node listens for GitHub push events. A Set node extracts the repository name and author. An IF node checks whether the commit message contains "BREAKING". If true, a Slack node posts to #engineering-alerts. If false, the workflow ends silently. Build time: under ten minutes, zero code required.
Nightly lead enrichment
A Schedule trigger fires at 02:00 each night. A PostgreSQL node pulls leads created in the past 24 hours that have no enrichment data. An HTTP node calls the Clearbit Enrichment API for each lead. A Code node restructures the response into the CRM format. A HubSpot node updates each contact record. A Slack node posts a summary to #sales-ops.
AI security incident enrichment
A webhook from a SIEM fires when a medium or high severity alert is created. An HTTP node fetches the IP addresses from the alert payload. A loop iterates over each IP: an HTTP node queries VirusTotal, an AI Agent node (using Claude or GPT-4) summarises the threat context in plain English, and an HTTP node patches the incident ticket in Jira with the enriched description. Average enrichment time: under 30 seconds per incident.
Employee onboarding sequence
HR submits a form when a new hire is confirmed. n8n creates the user account in Okta, provisions a GitHub organization invite, creates a Notion onboarding page from a template, schedules a welcome email for their first Monday, and posts a welcome message in the team Slack channel, all from a single workflow triggered by the form submission.
AI agents in n8n
n8n AI agent nodes let you build workflows where a language model makes decisions, not just transforms data. An agent node receives a goal, a list of available tools (nodes it can call), and optional memory. It reasons through the goal, calls tools in sequence, and returns a structured result.
Unlike prompt chains, agent workflows are reactive: the model decides which tools to call and in what order based on what it learns at each step. n8n makes every step of that reasoning visible on the canvas, including the prompt sent, the response received, and what tool was called next.
n8n supports OpenAI, Anthropic Claude, Mistral, Ollama (for local/offline models), and any model accessible via a compatible API. Switching models is a dropdown change; the workflow structure does not change.
Human-in-the-loop nodes pause execution until a person approves or rejects the next action. This is particularly useful in security workflows where an AI may flag a suspicious user account: the workflow pauses, sends the context to a Slack channel, and waits for an analyst to confirm before proceeding with a disable action.
MCP support
n8n added Model Context Protocol (MCP) support in 2025. This allows n8n workflows to connect to any MCP-compatible tool server, including database connectors, file systems, and custom internal tools, using the same protocol that Claude, Cursor, and other AI clients use. Teams that have already built MCP servers for their internal systems can reuse them directly in n8n agent workflows without rebuilding integrations.
Self-hosting versus n8n Cloud
The self-hosted community edition is free, has no execution limits, and runs on any Docker host. You manage infrastructure, updates, backups, and availability. Suitable for teams with existing DevOps capacity.
n8n Cloud is fully managed. No server to operate, automatic updates, guaranteed uptime SLAs. Priced per workflow execution and active workflow count. Suitable for teams that want automation without infrastructure overhead.
The enterprise self-hosted tier adds SSO, RBAC, audit logs, SIEM integration, Git-based deployment, isolated environments, and dedicated support.
The decision is straightforward: if your organization has a no-external-data policy, self-hosting is the only viable option. If your team does not want to manage one more service, n8n Cloud removes the operational burden.
n8n vs Zapier vs Make
Zapier wins on pure integration count (7,000+ versus n8n's 500+) and accessibility for non-technical users. Make sits in the middle. n8n wins on data control, custom logic, AI agent depth, and cost at volume. The tradeoff is operational complexity if you self-host.
Teams replacing Zapier with n8n typically cite three reasons: customer data cannot leave their infrastructure, per-execution costs at scale are significant, and they keep running into workflows where Zapier cannot handle the required logic without adding a middleware service.
Who should use n8n
n8n is a strong fit for technical teams: developers, DevOps engineers, IT operators, and security analysts who need automation that handles complex branching, custom data transformation, or privacy constraints. It is also well suited to organizations running high-volume pipelines where per-execution SaaS costs add up, and to teams building AI agent workflows that need full visibility into model decisions.
n8n is not the best choice for non-technical users who need simple app-to-app automation without custom logic. Zapier's user experience is more accessible for that use case. It is also not ideal for teams with no capacity to maintain a self-hosted service who have low automation volume, where n8n Cloud becomes expensive relative to Zapier's free tier.
Getting started
The fastest way to try n8n is with Docker. A single command spins up a local instance:
Run docker pull docker.n8n.io/n8nio/n8n then docker run -it -p 5678:5678 -v n8n_data:/home/node/.n8n docker.n8n.io/n8nio/n8n. Open localhost:5678, create your account, and build your first workflow. The canvas is self-explanatory: drag a trigger node, connect an action node, click Execute to test.
For production self-hosting, the n8n documentation recommends running behind a reverse proxy with a PostgreSQL database instead of the default SQLite, and enabling a public domain for webhook URLs. The community forum has guides covering most common deployment patterns including Docker Compose, Kubernetes, and Coolify.
Summary
n8n occupies a specific and valuable position in the automation landscape: visual workflow orchestration with no limits on custom logic, self-hostable with enterprise controls, and increasingly capable as an AI agent runtime. For IT and engineering teams that have hit the limits of Zapier or Make, whether on data privacy, code flexibility, or cost, n8n is worth evaluating. The self-hosted community edition costs nothing to test, and a working Docker deployment takes under five minutes.