n8n Review 2026: The Open-Source Automation Platform Worth the Setup Time

By Jonathan Hildebrandt, Co-Founder at Pondero Tested: March to April 2026

n8n's real product is not the canvas editor or the integration count. It is the deletion of per-execution pricing. Every hosted automation tool, Zapier and Make included, charges per task or operation, which means your automation bill grows with your success. Self-hosted n8n severs that link: the cost of the ten-thousandth workflow run is the same as the first, which is to say nothing. After running an n8n self-host on a $6/month Hetzner VPS since January 2026, 23 production workflows, three on the AI Agent node, total spend $24 over four months for unlimited execution volume, that economic claim is the entire review. The 4.3 rating is that thesis minus the setup tax.

The pricing mechanism, stated precisely

The reason this matters is structural, not a discount. Hosted automation tools price the execution because the execution is what consumes their compute. Their cost scales with your volume, so your price must too. n8n's source-available model moves the compute onto your box, which decouples your automation cost from your automation volume. Past a few thousand monthly executions, that is not a saving, it is a different cost curve entirely.

Compare the line items as of the April 2026 pricing:

Platform	Price	Execution limit	Arbitrary code
n8n Cloud Pro	~$50/mo	10,000/mo	Yes (JS/Python)
n8n self-hosted	~$6/mo VPS	Unlimited	Yes
Zapier Professional	$69/mo	2,000 tasks	No (Paths only)
Make Pro	$16/mo	10,000 ops/mo	No

n8n Cloud is now euro-denominated (Starter around 20 euro/month, Pro around 50 euro/month on annual billing as of the May 2026 pricing page); the dollar figures here are the test-window approximations and the relative ranking has not moved. The point the table makes is blunt: for any technical team past trivial volume, self-hosted n8n beats every competitor on price-to-power, and it is not close.

The license clause the pricing page does not lead with

This is the load-bearing detail, and most reviews skip it. n8n is not MIT-licensed open source. It ships under the Sustainable Use License (a source-available fair-code license). The practical boundary: internal business use, automating your own company's work, is free and unrestricted on self-hosted. What is not permitted is hosting n8n as a service to third parties or embedding it as the core of a product you sell. For the overwhelming majority of buyers automating their own operations, this changes nothing. If your plan is to resell n8n's automation to your customers, the free self-hosted path is closed and you need a commercial agreement. Decide which side of that line you are on before you build, not after.

Code nodes: the logic ceiling Zapier cannot raise

The Code node runs real JavaScript or Python against your workflow data, and in self-hosted mode it can call external libraries. This is the second structural advantage, and it has a clean test. I built a lead enrichment workflow: pull a contact from a CRM, call the Clearbit API, score the lead with a custom function, route by score into different sequences. Two code nodes, four integrations, one workflow. The same logic in Zapier needs Paths Plus, custom webhooks, and realistically a middleware service to hold the scoring function, because Zapier has no place to put arbitrary code. That is not a polish gap. It is an expressiveness ceiling: a step-based tool with no code primitive cannot represent logic that does not decompose into its predefined steps.

The AI Agent node, and its real failure mode

n8n shipped a native AI Agent node in 2024. It drives OpenAI, Anthropic, Google, Groq, and local models through Ollama, and it plans multi-step tasks against tools you wire in, either other n8n nodes or raw HTTP. I built an inbox triage agent: read Gmail, classify intent with Claude, extract entities, route to Slack channels with a drafted reply, escalate to a human below 80% confidence. Build time was four hours. On a platform without a real agent primitive, that is a custom Python service plus its hosting.

The limitation is specific and worth planning for. The AI Agent node does not absorb LLM failure for you. Malformed model output can halt the run mid-execution. A dependable LLM workflow in n8n is therefore an engineering task, not a drag-and-drop one: build the retry path and the fallback branch before anything touches a customer. The node gives you the agent loop; it does not give you the reliability layer.

What you own when you self-host

The canvas is a freeform graph, not Zapier's linear step list. That is more powerful for branching and merging data paths and disorienting on day one if your hands expect a linear builder. The data inspector showing live per-node output during a test run is the single best debugging affordance in the category. Credential management is centralized, which is correct for security and buried in the UI, which is not.

The cost you accept for the pricing curve is operational ownership. Self-hosted reliability is exactly your infrastructure's reliability. My small VPS ran eight weeks without an outage, but business-critical workflows need the money you saved spent back on redundant hosting, database backups, and monitoring. The saving is real and so is the responsibility. n8n Cloud, by contrast, held zero outages across the same eight weeks with thorough execution logs and email, webhook, and error-workflow alerting. If you cannot staff the ops, Cloud is the candid choice even though it reintroduces the execution cap.

What the spread of 23 workflows actually showed

The candid shape of n8n is the distance between its easiest and hardest builds. Gmail to Slack took under 20 minutes. Invoice processing with PDF extraction into Airtable and Xero needed code nodes and ate a full day to build and debug. Trivial things stay trivial; hard things stay hard, and n8n does not pretend otherwise the way no-code marketing does. Between those poles sat lead routing, content repurposing, support ticket classification, competitive monitoring, and a GitHub-to-Linear sync, all in production on the $6 VPS.

The verdict

n8n is the best automation platform in 2026 for any team that can run a Docker container, and the reason is one mechanism: self-hosting deletes per-execution pricing, which is the cost that quietly kills Zapier and Make power users as they succeed. The code node and the AI Agent node mean you also do not trade power for that price. For developers and ops-minded teams, nothing else this year competes on price-to-power.

The recommendation flips on two clean conditions. First, if no one on the team is comfortable with Docker, JSON, and basic API work, the setup tax is real and Make at $16/month is the better on-ramp. Second, if your intended use is reselling n8n's automation to third parties, the free self-hosted path is closed by the Sustainable Use License and the economics change entirely. Outside those two cases, this is the platform.

Start from n8n and stand up the self-hosted Docker image first, since the pricing argument only holds on the self-hosted path.