Don’t describe
your workflows.
Wire them.
n8n is a node-graph for your whole stack. Drag nodes, snap connections, drop into real code exactly when you want it, wire AI agents in as first-class citizens — and watch every run pulse across the canvas. Self-hosted or cloud. Your call, always.
Grab a node. Snap it in. Run it.
Everything in n8n is a node on an open canvas — triggers, HTTP calls, transforms, AI steps. You don’t fill in forms; you draw the system and the graph becomes the documentation your team never had to write. No dialogs. No YAML archaeology. This board is real — finish the workflow with your own hands.
- 1Drag all three nodes into their sockets
- 2Wire GitHub Trigger → IF merged?
- 3Wire IF merged? (true) → Slack
- 4Execute — watch a real run land
Every no-code tool has a ceiling. Ours is deleted.
When a built-in node can’t say what you mean, drop a Code node and write
real JavaScript or Python — right inside the workflow, with the previous
step’s items in scope. npm packages when you self-host. Expressions everywhere:
{{ $json.lead.score }}.
This Code node actually executes in your browser. The input is three raw leads a webhook just delivered. Edit the code — press ⌘⏎.
Agents you can watch think.
n8n’s AI Agent node wires any model — OpenAI, Anthropic, Google, Mistral, or fully offline via Ollama — to tools that are just other nodes on your canvas. Every reasoning step is a traceable run: prompt, tool call, observation, answer. Pick a task and run the loop. It pauses for your human approval before it acts — because that’s wiring too, not lock-in.
- 01Any model, swappable. OpenAI, Anthropic, Gemini, AWS Bedrock, or a local Ollama model — switching providers never rebuilds the workflow.
- 02Tools are literal nodes. The agent can call Postgres, Slack, your private HTTP endpoint — even another workflow — because tools hang off the same canvas.
- 03Memory & RAG as wiring. Chat memory, Postgres/Redis stores, vector retrieval from Pinecone, Qdrant or pgvector — clipped on, not coded up.
- 04Humans stay in command. Approval steps, guardrails and evaluations are first-class nodes. You inspect every prompt and every answer.
- 05Speaks MCP, both ways. Call MCP tools, or expose your own workflows as an MCP server for Claude, Cursor and friends.
500+ ways in. Zero ways out.
Pre-built nodes for the tools your team already runs — with auth and pagination handled. No node for your niche system? The HTTP Request node speaks any REST or GraphQL API and imports your cURL one-to-one. Try the shelf — type to filter.
…plus 460+ more nodes, thousands of community templates, and your own Code nodes when nothing off the shelf fits.
One container. Your iron. Zero meter.
n8n Community Edition is source-available fair-code — the full repository on GitHub under the Sustainable Use License, free to self-host with unlimited executions. Your workflows export as JSON and check into Git. Your credentials live in your vault. Air-gap it if you need to; the canvas feels identical.
Cloud — hosted in Frankfurt, SOC 2 compliant — is one click away when you’d rather not run the box. Move between them freely; the graph is portable.
Fair-code, honestly
Sustainable Use License. Source on GitHub, free for your own business use — you just can’t resell n8n as a service.
Executions unmetered
Self-hosted runs are free and unlimited. Scale out with queue mode on Redis when the graph gets hungry.
Your data stays yours
Run on-prem or air-gapped. Encrypted credentials, SSO/SAML & LDAP on Business and Enterprise tiers.
Serious people already trust this canvas.
Threat-intel SOAR automation on n8n — “it did everything we wanted, all in one tool.”
Building an AI-first company culture — “n8n was the big unlock.” — Ollie Scheers, CTO
technical teams shipping with n8n · source: n8n.io
Pick your port. Plug in.
The canvas is a free Community Edition, a €20 cloud plan, or this exact graph on your own hardware in five minutes. The first node is the only hard one.
$ docker run -it --rm -p 5678:5678 docker.n8n.io/n8nio/n8n