Canvas workflows that start from a webhook or schedule
n8n is a workflow automation platform written in TypeScript. You lay nodes out on a canvas: a trigger (webhook, schedule), then calls to other services, branching logic or AI models. n8n GmbH, the company behind it, runs a hosted cloud and ships a Docker image for self-hosting. The README claims more than 1,500 integrations and 9,000 workflow templates.
Where ToolJet builds internal interfaces and Dify builds LLM applications, n8n starts from service-to-service workflows and runs AI agents inside them. The comparison of workflow automation and AI agent tools sets n8n's license and editions against the other tools'.
Code nodes, AI agents and a built-in MCP server
- Code where the canvas runs out: the Code node runs JavaScript or Python. Self-hosted, it can import external npm modules; on n8n Cloud, only
cryptoandmomentare available. - AI agents as workflow steps: AI nodes built on LangChain, swappable model providers (OpenAI, Anthropic, Google, open models), an MCP client node, and optional human approval before an agent calls a tool.
- A built-in MCP server: an MCP client can create a workflow from a description, edit existing ones (since 2.13) and run the ones it has access to.
- Queue mode is free: the main instance hands executions to Redis and workers pick them up. Only multi-main mode needs a license.
Not open source by its own docs, with paid SSO
The license is not open source, and n8n says so. The Sustainable Use License, written by n8n, replaced Apache 2.0 with the Commons Clause on March 17, 2022. It limits use to your own internal business purposes or to personal, non-commercial use, and allows redistribution only free of charge and for non-commercial purposes. Because the OSI rules out restrictions on use, n8n's own documentation declines the open source label. Activepieces covers the same ground under MIT, outside its ee directories.
The license FAQ draws the commercial line. You can build automations for clients on your own instance as long as they cannot create or edit them, charge for setup, maintenance, training or consulting, and run n8n behind your product if users only trigger automations you built. You cannot host n8n so clients build their own workflows, let your product's users configure workflows (through your UI, the API, MCP or an AI agent), fork the code to launch your own automation product, or white-label it. Embedding the n8n interface in a product requires an OEM agreement.
Files marked .ee fall under a separate license. Any file with .ee. in its name, or inside a .ee directory, is covered by the n8n Enterprise License: you may modify and test it without a subscription, production use needs an Enterprise license, and n8n owns any changes you make to it.
The free Community edition leaves out team features. The editions comparison page lists what it lacks: workflow and credential sharing, projects, SSO (SAML, LDAP), Git version control, environments, external secrets and log streaming. Free registration (Registered Community) adds folders and debugging in the editor. The Business plan is self-hosted only and costs €667 a month billed annually for 40,000 executions, with SSO, LDAP and Git. Log streaming, external secrets and multi-main mode are Enterprise features, priced on request.
n8n pitches itself against Zapier on what gets counted. Its comparison page bills per full workflow execution, however many steps it has, where Zapier counts tasks. The Starter cloud plan, at €20 a month billed annually, includes 2,500 executions.
Scaling out means Postgres and Redis. SQLite is the default database and Postgres the only other one supported, since 2.0 dropped MySQL and MariaDB. Queue mode adds Redis and worker processes, and the docs do not support that distributed setup on SQLite. In queue mode, binary data cannot be stored on the filesystem: the docs point to S3 external storage, which the Community edition does not include.
A new minor version almost every week. 2.0.0 shipped on December 8, 2025 and 2.43.0 on October 6, 2026. The Docker image has two channels: beta, the newest release, which the docs warn may be unstable, and stable for production (2.43.1 and 2.42.4 as of October 7, 2026). The 1.x line still gets patches: 1.123.83 came out on September 30, 2026.