n8n Review: Cloud Costs, Self-Hosting and Buyer Guide

n8n belongs on the shortlist when a business needs workflows that combine application integrations, custom requests, data transformations and optional AI steps. Its attraction is control over the process, including a self-hosted option. Its main buying risk is assuming that a visual workflow eliminates engineering and operational responsibilities.

This is a desk-researched buyer evaluation using official n8n pricing, hosting documentation and licensing guidance, checked for November 2026. We have not load-tested the product or measured time savings for this article. The pilot recommendations are proposed evaluation criteria, not reported test results.

Our recommendation: begin with one internal, reversible workflow and a named owner. Prefer managed hosting unless someone is explicitly accountable for infrastructure. Choose n8n because it fits the workflow and maintenance model, not because self-hosting sounds free.

The official feature matrix lists integrations, HTTP and GraphQL requests, JavaScript/Python code steps, webhooks, error workflows and automatic retries. AI capabilities include an AI Agent node, MCP client and server functionality, evaluations and human approval for tool calls. Some newer features are explicitly marked preview; do not base a critical commitment on their permanence.

A sensible business workflow might receive an inquiry, validate required fields, look up a customer, classify the request and prepare a response for approval. Use deterministic rules for mandatory fields and routing constraints. Reserve AI for genuinely ambiguous work, such as summarizing free-text context.

The distinction matters financially and operationally. A rules-based step has a predictable contract. An AI step needs representative examples, output validation and a failure path. n8n can coordinate those pieces, but it does not make a model's interpretation automatically correct.

As of November 2026, the dollar-denominated version of n8n's pricing page retrieved for this review lists these annual-billing prices. Search results can show euro-denominated figures; verify currency, tax and billing terms in your own checkout rather than treating these as converted local quotes.

n8n prices these plans by workflow executions rather than individual steps, and lists unlimited users, workflows and integrations. That does not mean unlimited throughput. Starter lists five concurrent executions; Pro advertises up to 50, with tier-dependent details in the comparison table.

Illustratively, 100 runs per day over a 30-day month equals 3,000 runs, above Starter's listed allowance. That is workload arithmetic, not a vendor performance estimate. Test how your actual schedules, child workflows, retries and failure handling appear in usage reporting before committing annually. Model-provider charges and third-party service fees can remain separate costs; unlimited workflow steps is not unlimited AI inference.

The hosting documentation says a self-hosted installation without a license key runs as the free Community edition. Adding a Business or Enterprise license enables the corresponding paid edition. Docker Compose is among the documented deployment options.

Self-hosting shifts work onto your team: patching, access controls, backup restoration, database maintenance, monitoring and incident response. Include that labor in the comparison with Cloud. Storing workflow credentials yourself also creates an obligation to protect them. Self-hosting does not keep data local when a workflow sends it to an external AI or application API.

Licensing deserves its own review. n8n's Sustainable Use License guidance describes the software as source-available, not OSI open source. It permits internal business use within its terms, but selling hosted access or embedding customer-facing functionality may require a separate agreement. The guidance permits commercial consulting and support; that is not blanket permission to resell the platform. Ask n8n about your specific distribution model.

Pick a workflow with a known input, a measurable output and a safe fallback. For example, prepare an internal CRM update without sending customer messages automatically. Use sanitized data first, then a limited production sample after security approval.

1. Establish the baseline: record manual handling time and common exceptions.
2. Exercise failures: expired credentials, missing fields, duplicate events, API rate limits and malformed model output.
3. Prevent duplicate writes: use a stable event identifier and check whether the destination already processed it.
4. Test recovery: make sure retrying a failed run does not repeat completed payments, messages or record creation.
5. Assign operations: define who receives alerts, how runs are inspected and when the workflow is disabled.

Measure successful outcomes, not just successful executions. A run that finishes while filing a lead under the wrong account is still a business failure. Include time spent resolving exceptions in the savings calculation and keep human approval for consequential outbound actions.

n8n is a strong candidate for technical operations teams that want branching workflows and custom integrations without implementing every connection from scratch. Cloud can make sense when those teams want to focus on workflow behavior rather than infrastructure. Self-hosting can fit organizations with an existing operational owner and a concrete deployment requirement.

Skip it if your workflow is already covered by a reliable native integration, nobody can debug API failures, or the business expects unattended automation with no exception handling. Do not select Community solely to obtain paid governance features: compare the edition matrix before deciding where to host.

For a broader shortlist, read our automation alternatives guide and AI automation planning guide. The final purchase test is straightforward: does the workflow remain understandable and recoverable when its original builder is unavailable? If not, fix that before expanding automation.