n8n is the open-source automation platform that finally crossed the chasm in 2026 — Zapier-class workflow builder with native AI Agent support, at one-tenth the cost when self-hosted. The developer-friendly choice that's now also enterprise-viable.
The 30-second take
n8n is workflow automation. Visual builder, 500+ integrations, custom JavaScript/Python nodes when you need them, and — critically since 2024 — first-class AI Agent nodes. You can build a workflow where an HTTP webhook fires an AI Agent that calls 3 tools, summarizes the result, and posts to Slack — all in one visual canvas.
The differentiator vs Zapier or Make.com: n8n is open source. You can self-host on a $5/month VPS and run thousands of workflows for the price of dinner. At scale (50K+ executions/month) the economics get absurd in n8n's favor.
What n8n does well
Self-host or cloud. Docker compose to spin up your own instance in 10 minutes; managed cloud if you don't want ops. Same product either way.
AI Agent node is first-class. Built-in support for OpenAI, Anthropic, Google, local Ollama, plus tool-calling, memory (vector store), structured output, and a visual conversation builder. You can prototype an agent in 20 minutes and ship it the same day.
Custom code where needed. Drop into JavaScript or Python for any node. No vendor lock-in on transformation logic — your business logic stays in code you can port elsewhere.
Fair-code license. Source-available, free to use commercially as long as you're not building a competing automation SaaS. n8n calls this "sustainable use" — most teams' lawyers approve quickly.
Connector quality. 500+ integrations and counting. Coverage is narrower than Zapier (7K+) but the depth is comparable for the top 100 SaaS apps — fewer rough edges per integration.
Where n8n stumbles
Connector breadth is the gap. If your stack includes a niche app, Zapier probably has it and n8n probably doesn't. The HTTP Request node + custom code is the fallback but it's real work.
Self-hosting requires ops literacy. Docker, Postgres, environment variables, periodic upgrades, monitoring. A solo founder can handle it; a marketing team cannot. Use the cloud tier in that case.
Cloud tier pricing tightens fast. Free tier is generous; paid cloud starts at $20/month for 2,500 executions. At 100K+ executions/month you're paying $400+/month — at which point self-hosting becomes a no-brainer (a $20/month server handles 1M+ executions).
Visual builder gets messy at scale. A workflow with 30+ nodes becomes hard to navigate. The community is good at modularization patterns (sub-workflows, reusable triggers) but you'll outgrow the default canvas.
Pricing reality check
Self-hosted: $5-50/month VPS + your time for upgrades + monitoring. Effectively free at small scale.
n8n Cloud:
- Starter: $20/month — 2,500 workflow executions
- Pro: $50/month — 10,000 executions + more concurrency
- Enterprise: custom — SSO, audit logs, SLAs, higher concurrency
For a small business running ~5K executions/month: $20-50/mo on Cloud or $5/mo self-hosted.
For an agency running 100K+: self-host on a $40-80/month server (10-50× cheaper than Cloud).
How n8n compares
- n8n vs Zapier: Zapier for breadth + simplicity. n8n for cost at scale + custom code + open source. Most teams end up using both.
- n8n vs Make.com: Make has a slicker visual editor and better visual debugging. n8n is open source and cheaper. Make wins for non-technical builders; n8n wins for technical builders.
- n8n vs Lindy: Lindy is an AI-first conversational agent platform. n8n is workflow automation that happens to have AI Agent nodes. Different products for different problems: pick Lindy when the agent IS the product; pick n8n when the agent is one node in a larger workflow.
Bottom line
n8n is the right default for technical teams in 2026. Self-host on a small VPS, get Zapier-class capability at near-zero marginal cost, drop into custom code when needed, ship AI Agent workflows in hours. The learning curve is real but the ceiling is high. Non-technical teams should start with Make or Zapier; technical teams should start with n8n.
Try n8n → · How to use n8n with AI · Compare with alternatives