n8n connects apps, code and AI inside workflows teams can inspect

n8n is a workflow automation platform for connecting applications, transforming data and orchestrating AI alongside ordinary software steps. A workflow begins with an event or schedule and passes information through connected nodes until the intended job is complete. Its appeal is the combination of a visual representation with technical control: a team can use existing integrations where they fit and write code or call an API where they do not. That flexibility makes it useful for internal operations and custom integrations, provided someone owns the workflow as running software.

01 / ProductWhat n8n automates, and where AI fits

The product’s feature and pricing overview¹ describes integrations, code steps, custom API requests and workflow execution tools. AI is one possible component of a workflow, not a requirement for using the platform. A scheduled database synchronisation can remain entirely rule-based. A document-processing workflow might use a model to extract information, then use explicit validation and ordinary API calls to update the destination system.

That distinction is valuable because different parts of a job require different behaviour. Interpreting a messy supplier email may benefit from a model. Matching an invoice identifier against an existing record should use a reliable lookup. Deciding whether the automation already processed that invoice requires stored state. An n8n canvas can make these boundaries visible, but drawing the nodes does not automatically establish that the data flowing between them is correct.

n8n also offers a deployment choice. Its Cloud and self-hosting guide² separates hosting from edition: Cloud is managed by n8n, while self-hosted installations can use Community or paid editions. A free edition and a managed service are different propositions. Choose first who should operate the runtime, then which collaboration, governance and capacity features the team needs.

02 / AudienceWho should choose n8n?

n8n is particularly relevant to technical operations teams, developers and builders who regularly encounter workflows that outgrow a simple app-to-app connection. A common reason is data shape: one system supplies a nested object, another expects a different schema, and the process needs branching or custom validation between them. Being able to combine integration nodes with code can make that work easier to explain and maintain than a collection of disconnected scripts.

It is less attractive when nobody is prepared to diagnose failures or maintain credentials. A working prototype can become a business dependency surprisingly quickly. If the team mainly needs standard integrations and wants the platform to absorb more of the operational detail, our Zapier guide provides a useful comparison. Our Make profile focuses on visual scenarios and how data bundles affect execution and cost. Compare an actual workflow, because abstract integration counts do not establish that the exact operation you need exists.

For organisations considering self-hosting, the relevant skill is operating a service, not merely starting a container. Somebody must own updates, backup restoration, access and incident response. n8n’s hosting documentation³ outlines installation choices and edition behaviour. The practical question is whether controlling the deployment solves a real requirement that justifies that work.

03 / WorkflowA proposed supplier-document workflow

Consider an operations team receiving supplier invoices by email. This is an illustrative workflow design, not a benchmark or private n8n test. The desired result is a reviewed record in the accounting system, with the original document attached and a clear exception path. The trigger receives a message, saves a durable reference, extracts the relevant fields and checks whether the invoice has already been processed.

Use AI only for the part that benefits from interpretation. A model can propose supplier name, invoice reference and line-item data from the document. Explicit code or validation rules should then check required fields, arithmetic and expected formats. If a supplier identifier cannot be matched confidently, route the item for review instead of creating a new supplier automatically. The workflow becomes more dependable when each step has a narrow, inspectable output.

Where should human approval sit?

n8n’s human review for tools guide⁴ describes pausing an AI agent before selected tools execute and showing the proposed parameters to an approver. Approval can therefore apply to the action itself rather than merely to a final summary. For the invoice example, the reviewer should see the proposed record and source document before the write operation. A vague “continue?” message would provide little meaningful control.

The record-creation step should use a stable source identifier so the workflow can detect duplicates. If the accounting API returns an uncertain result, check whether the record exists before retrying. This is a design recommendation about the integration, not a guarantee supplied by the canvas. A retry is useful only when the operation’s behaviour under repetition is understood; otherwise it can turn a temporary network problem into duplicate business data.

What happens when a step fails?

The error-handling documentation⁵ explains assigning an error workflow that begins with an Error Trigger and receives execution context. For this process, use that mechanism to place an item in an actionable exception queue, including the original document reference and failed stage. An alert without enough context simply moves the investigation into somebody’s inbox. The owner should be able to resume or reconcile the specific item.

Test missing attachments, an already imported invoice, an invalid model response, an expired credential and a downstream timeout. Then check the accounting system, stored document and exception record together. A green workflow execution does not by itself prove that the intended accounting record is complete. Conversely, a failed notification after a successful import should not cause the import to be repeated blindly.

04 / PricingHow n8n pricing works

The pricing page¹, accessed on 15 September 2026, bases paid usage on complete workflow executions with unlimited steps. The published annual-billing entry tiers below are in EUR. The selected execution allowance matters, as do concurrency and collaboration features. External model or application services used by a workflow may have their own charges; the n8n subscription is not a universal bundle for every connected vendor.

An execution-based model can be attractive for a long workflow with many transformations. It should not be confused with unlimited runtime or infrastructure. A chat-triggered workflow may execute repeatedly during a conversation, while a scheduled batch may process many records within a run. Estimate the actual triggers and execution boundaries in the design. Compare this with task- or credit-based alternatives using the same workload and the same success criteria.

For self-hosting, include the infrastructure and operating cost as well as any licence. A small server bill can omit the largest cost: somebody responding when the workflow stops during an important business process. For Cloud, examine whether the selected plan provides the concurrency, execution history and administration needed to diagnose that process. Buying more usage and buying the ability to operate a team service are related but distinct decisions.

05 / DistinctionsSelf-hosting changes the scaling and licensing questions

The queue-mode guide⁶ describes separating trigger handling from execution workers, with Redis carrying queued work and a database storing execution information. That architecture can help a deployment handle more concurrent work. It also adds components to monitor and recover. Scaling the worker pool does not remove limits imposed by an external accounting API or model provider, so rate handling still belongs in the workflow design.

n8n is source-available rather than simply unrestricted open-source software. Its Sustainable Use License explanation⁷ describes permitted internal use and commercial restrictions, including hosted access and some embedded uses. Ordinary internal automation and offering a customer-facing automation product can fall into different arrangements. Read the current examples for the intended use and obtain clarification from n8n where the boundary is unclear; do not infer resale rights from access to the code.

The larger product lesson is that visibility and control have operational value only when someone uses them. A technically flexible workflow can be easier to adapt to a difficult integration, but it can also become difficult for a nontechnical colleague to maintain. Use clear node names, defined data contracts and an exception process that a second operator can understand.

06 / QuestionsWhat should an evaluation establish?

A useful pilot should answer whether the exact integrations work, whether the team can diagnose failure and whether the chosen hosting model is supportable. We have not tested a private n8n deployment for this profile. The public documentation establishes the available building blocks, while the proposed supplier workflow illustrates the responsibilities that remain with its designer. Measure completed business records and review effort, not simply executions per day.

Also decide what execution data should be retained. Workflow logs can contain the same personal or commercial information as the source systems. Keeping everything indefinitely may make debugging convenient but create unnecessary exposure and storage. Retain enough evidence for the task, restrict access appropriately and test restoration of the information needed to resume work after an interruption.

07 / DecisionChoose n8n when flexibility has a clear owner

n8n is a strong candidate for evaluation when a workflow needs custom logic and the team wants to see and control how its parts fit together. Begin with a real process, identify the authoritative record and design the exception path before scaling it. The right deployment is the one the organisation can maintain reliably, whether that is n8n Cloud or a self-hosted edition.