n8n is a fair-code licensed workflow automation tool for connecting APIs and manipulating data through visual workflows made of nodes. A trigger starts an execution, later nodes can transform or validate its data, and a final node can perform an action in a verified destination system.
For example, a hypothetical event-intake workflow could receive a webhook, validate the event ID and payload version, normalize the fields, and write one logically unique record to a database. The important question is not whether n8n can connect two services in principle. It is whether the required operations, credentials, data controls, ownership, and failure handling work for your specific process.
n8n can run as n8n Cloud or as a self-hosted deployment. That choice changes who operates the automation layer, where data is hosted, how updates are managed, and how much technical responsibility your team accepts.
What is n8n?
n8n is a visual workflow automation and orchestration tool. Its editor represents workflow steps as nodes that pass data to other nodes for actions such as transformation, routing, API requests, and database operations. A typical workflow starts with a trigger, processes data through one or more steps, then performs an action or routes the execution to an exception path. The n8n documentation covers workflow construction, integrations, Cloud, self-hosting, and operational features.
n8n supports visual and low-code work, but visual construction does not eliminate the need to understand data formats, authentication, API behavior, permissions, rate limits, and failure handling. More complex workflows may use API configuration or custom transformations in the Code node. The Code node cookbook provides examples and recipes. It does not turn those examples into tested production implementations.
n8n is an automation and orchestration tool, not a general-purpose backend service by default. It can coordinate services and process events, but a long-running, stateful, business-critical process may belong in a dedicated application or backend service instead.
Treat n8n as an integration layer with operating responsibilities, not as a substitute for deciding where business state and authorization belong.
How to decide whether n8n fits your team
Start with operating ownership rather than a simple technical versus nontechnical label. Someone must own workflow changes, credentials, failed executions, data permissions, and incident response. For self-hosted n8n, someone must also operate the instance itself.
- n8n may fit when workflows require custom API orchestration, data transformation, private-network access, or control over deployment location, and a named team can maintain the integration layer.
- A managed platform may fit better when business teams need to own common SaaS automations and the organization does not want to operate another service.
- Application-native automation or a backend service may fit better when the process is narrow, deeply stateful, or difficult to govern as a visual workflow.
Before selecting a platform, verify each required application operation. Record the exact trigger and action, authentication method and scopes, pagination behavior, rate limits, and any HTTP API fallback. An app or node appearing in a directory does not prove that it supports the operation your workflow needs.
n8n can be appropriate for internal workflows, custom systems, and integrations that need more control than a purely managed platform provides. It can also become a maintenance burden when nobody owns patching, credentials, monitoring, backup recovery, and exception handling. This is a decision framework, not a claim that one platform is universally easier or more scalable.
What n8n workflows can do, and how to make them dependable
A useful workflow has an explicit input contract, a named system of record, deterministic validation, an idempotent write strategy, and a defined failure owner. The following design is hypothetical. Its field names, unique key, destination behavior, and exception rules are proposed implementation choices, not vendor-published schemas.
For one logical event record, a proposed input might include:
{
"tenant_id": "tenant_123",
"source_system": "example_source",
"source_event_id": "evt_456",
"source_event_time": "2026-11-21T10:15:00Z",
"payload_version": "v1",
"raw_payload_reference": "record://source/456"
}
The source and destination are deliberately unspecified until their current node coverage, API operations, authentication, rate limits, and idempotency behavior are verified. A documented n8n capability can support the sequence, but it does not confirm that a particular third-party connection is available.
The event key identifies the business event. The combination of workflow ID and execution ID identifies one n8n run. Keeping those grains separate prevents multiple runs from being collapsed into one event record and makes replays diagnosable. If the workflow also stores citations, use a citation-level key such as observation ID plus citation index, rather than prompt ID plus date. Aggregate summaries need their own scope, such as tenant, period, engine, model, and query-set version.
n8n provides execution history for inspecting runs and supports retrying failed executions. A retry is an operator action, not proof that a write is safe to repeat. Preserve the original request key, response status, and vendor request ID where the destination supplies one, subject to privacy controls.
Using AI inside an n8n workflow
Use deterministic rules for exact checks: required fields, duplicate detection, permissions, allowed status transitions, numeric thresholds, and whether a write is authorized. AI can assist with bounded tasks such as extracting entities from unstructured text, classifying ambiguous language, summarizing records, or ranking candidates for review.
If AI output feeds another system, require structured output and validate it before any write. A proposed classification result might contain a label, numeric confidence, extracted entities, policy flags, source record ID, model version, prompt version, and transformation version. The label should be checked against an allowlist, confidence should be numeric and within the expected range, and policy flags should route the record to review when required.
Model confidence is a classification signal, not authorization. Keep the raw input, structured output, validation result, model metadata, and reviewer decision separate so an operator can explain why a field changed.
For an AI-assisted classification workflow, the destination should receive only approved fields. Low-confidence, regulated, policy-sensitive, irreversible, or externally visible actions should go to a human review queue. A designated business reviewer resolves ambiguity, while the integration owner handles schema, permission, and destination-write failures. See AI agent implementation for relevant implementation support.
n8n Cloud, self-hosting, and the ownership trade-off
n8n documents Cloud and self-hosted deployment, including Docker and npm options. Cloud means n8n manages the hosted service. Your team still owns workflow design, connected-service permissions, credential choices, data mapping, and the response to failures in your process.
n8n operates the hosted service
Your team still manages workflow logic, credentials, connected-system permissions, data mapping, and exception handling. Consider it when you want managed n8n infrastructure without operating another service.
Your team operates the instance
You gain control over deployment location and infrastructure, while taking responsibility for installation, supported-version maintenance, network exposure, authentication, secrets, monitoring, backups, recovery testing, and incident response.
The current n8n pricing page states that Cloud data is stored in the EU on servers located in Frankfurt, Germany. That describes n8n’s stated hosting arrangement, not the storage location of every connected third-party service. Self-hosted data is stored wherever the customer hosts the instance.
Before self-hosting, assign an instance owner, review public webhook exposure, define a patch process, approve credential and community-node policies, name a backup and restore owner, and identify an incident contact. n8n’s security audit capability can identify selected risks involving credentials, webhooks, nodes, expressions, and instance settings. It does not replace patching, network controls, access management, backup testing, or a full security review.
n8n pricing, executions, and licensing
As of November 21, 2026, n8n’s pricing page bases Cloud usage on monthly workflow executions. An execution is a complete workflow run, not an individual node step. The same page lists Starter and Pro as Cloud plans, Business as self-hosted only, and Enterprise as available on Cloud or self-hosted. Prices, plan availability, support, and included features can change, so check the live page before budgeting.
The current page lists selected Business features such as SSO and LDAP, multiple environments, Git-based version control, and queue-mode scaling. It lists forum support for Business and dedicated support with an SLA for Enterprise. These features should not be assumed to be included in the free Community edition.
The self-hosted Community edition is free, but it is not a zero-cost operating model. Infrastructure, engineering time, monitoring, backups, security work, recovery testing, and support arrangements affect total cost. Compare complete workflow runs with the labor and infrastructure required to operate the instance.
Community uses n8n’s Sustainable Use License rather than a conventional permissive open-source license. n8n’s license guidance permits internal workflow use while describing restrictions for hosted services, white-label deployments, customer-facing workflow builders, and products that allow external users to configure workflows. If an agency, customer, or external user will access or configure a workflow product, review the guidance for that specific model instead of assuming a general permission.
n8n vs. Zapier: compare the workflow you need
n8n supports Cloud and self-hosted deployment. Zapier positions itself as a managed cloud platform. n8n currently prices Cloud plans by workflow executions, while Zapier’s current pricing is task-based and includes capabilities across its broader platform products. These units are different, so a meaningful comparison requires the same workflow, run volume, data path, and plan assumptions.
- Deployment and ownership: compare n8n Cloud, n8n self-hosting, and Zapier’s managed model based on who operates the service and who handles failures.
- Integration coverage: verify the exact trigger, action, authentication method, scopes, pagination, rate limits, and HTTP API fallback for every required application.
- Customization: assess whether custom transformations, private-network access, or API orchestration justify the additional operating responsibility.
- Team governance: check environments, access controls, support, workflow ownership, and review processes rather than assuming a directory count represents administrative fit.
n8n’s integration directory includes nodes and other building blocks, while Zapier’s directory advertises apps. Zapier’s own pages also use different rounded app counts. Those figures are dynamic and not directly comparable. Use the directories to identify candidates, then confirm the operation in current vendor documentation.
Zapier’s vendor comparison explains its positioning, but its evaluative claims are not an independent benchmark. If a managed Zapier workflow is under consideration, Zapier automation consulting is relevant implementation support.
A practical selection rule
Choose n8n when your team values workflow or infrastructure control, needs custom API or private-network work, and has an accountable owner for deployment and integrations. Choose a managed platform when multiple business teams need to build common SaaS automations and managed operations matter more than control of the automation environment.
For a narrow task, application-native automation may be simpler. For a long, stateful, business-critical process, consider a conventional backend service. The correct choice depends on the system of record, exact operations, permissions, run volume, exception path, and available operating capacity.
Test a small proof of concept with representative data, expected run volume, permissions, duplicate events, concurrent executions, retries, and failure cases. Approve a broader rollout only after the exact operations work and a named owner accepts the ongoing responsibility.
- Confirm the required trigger, action, authentication, scopes, rate limits, and any HTTP API fallback.
- Name the system of record and the owner of workflow changes, credentials, failures, and incident response.
- Test concurrent events with a database uniqueness constraint, transactional upsert, or destination-supported idempotency key.
- Validate AI output against an allowlist and route uncertain or policy-sensitive results to human review.
- Check data permissions, Sustainable Use License conditions, plan-dependent features, execution or task charges, and operating effort.
