n8n

What n8n is

n8n is a workflow automation platform combining a visual node editor with code and connections to applications and AI models. It can be used as a hosted service or deployed on an organisation’s own infrastructure. Its name abbreviates “nodemation”, combining nodes and automation.

This approach represents the process while allowing transformations where a standard connection is insufficient. That flexibility requires understanding the data in transit and how outcomes will be verified. Choosing hosted or self-managed deployment is an operational decision, distinct from the logic being automated.

Reference documentation: n8n source repository and project background · Hosting n8n and self-hosted editions.

Origins and development

Jan Oberhauser founded n8n in Berlin in 2019. The project retrospective connects his interest in automation with earlier visual-effects work and dates development to 2018. The first version appeared on GitHub in June 2019, followed by a wider public launch in October.

In his second-anniversary account, Oberhauser explained that he wanted to reduce the repeated work of creating, deploying and maintaining small scripts. Community involvement began through issues, proposals and repository contributions. This history helps explain the combination of a visual interface and technical extensibility.

The project developed from task automation towards orchestration of applications and models. Launch, company establishment and commercial expansion should be distinguished when comparing timelines; published dates may describe different milestones.

Reference documentation: n8n corporate milestones · n8n fifth-anniversary history · Jan Oberhauser: n8n second anniversary.

Funding, expansion and partnerships

Corporate milestones include a $1.5 million seed round in 2020, a $12 million Series A in 2021 and a $60 million Series B in March 2025. These figures concern historical announcements, not revenue or customer outcomes.

In October 2025, n8n announced a $180 million Series C led by Accel, including investors such as NVentures and T.Capital, at a $2.5 billion valuation. In May 2026 it announced SAP’s strategic investment at a $5.2 billion valuation and an agreement to embed n8n in Joule Studio.

In June 2026 Deutsche Telekom announced its certified-partner activity providing licences and services to businesses. A commercial partnership extends channels and implementation options; it does not establish that a particular integration is available in every account or configured in an individual organisation.

Reference documentation: n8n corporate milestones · n8n Series C, October 2025 · SAP strategic investment, May 2026 · Deutsche Telekom: certified n8n partnership.

Source availability and licensing model

n8n distributes its code under a fair-code model. On 17 March 2022 it replaced Apache 2.0 with Commons Clause with the Sustainable Use License. Source access does not mean an unrestricted open-source licence; conditions govern internal use, distribution and provision to third parties.

Current help distinguishes self-managed Community edition from Cloud and commercial licences. Hosting n8n as a service, white-labelling it or allowing external users to build their automations may require a commercial licence. Consult the contractual documentation for the particular use case.

Compare tools through three separate questions: whether code can be inspected, who operates the instance and which uses the licence permits. A free edition may support an internal process without authorising a proposed commercial model built around it.

Reference documentation: New Sustainable Use License, March 2022 · Sustainable Use License text · Which license is needed for a use case?.

Nodes, data and record correspondence

Nodes exchange arrays of items containing JSON-structured information and, where relevant, binary data. Understanding that structure supports transformation and record relationships. A node’s output needs to preserve the required correspondence with its inputs.

Editorial example: an action record includes a site, date and several pieces of evidence. Separating each evidence item may multiply output items, but should not multiply the count of actions. Preserve both action and evidence identifiers.

Define accepted values before transformation: a blank field, zero and “no data” do not necessarily mean the same thing. Standardising formats without agreeing meaning can produce a technically valid but misleading dataset. Tests should examine content and correspondence as well as row counts.

Reference documentation: n8n data structure.

JavaScript, Python and extensibility

The project uses TypeScript and Node.js. The Code node adds custom logic. Current documentation distinguishes JavaScript and native Python through task runners; Pyodide was removed in n8n 2. Available libraries depend on deployment and environment restrictions.

Code is useful where a precise transformation is awkward to express in a module. Specify inputs, outputs and edge cases before adding it. A small function with comprehensible examples is usually easier to maintain than a long sequence of hard-to-trace adjustments.

For example, when normalising site codes, preserve the original and produce a separate normalised field. If no match exists, return an identifiable exception. A cleaning rule should not invent an association that later appears confirmed.

Reference documentation: n8n source repository and project background · The Code node: JavaScript and native Python.

Cloud and self-hosting

n8n Cloud provides hosted operation. On an organisation’s infrastructure, a basic installation runs as Community edition; selected capabilities are enabled through Business or Enterprise licensing. Deployment guides cover installation, databases, users, configuration and maintenance.

The choice should identify responsibility for availability, updates, backups and restoration. An organisation-controlled server may support infrastructure choices, but connections to external models and applications still transmit information beyond that server.

An operational trial includes restoring a backup, replacing credentials and testing an update in a trial environment. Demonstrating one successful workflow run does not yet establish continuity. Team capability matters as much as product flexibility.

Reference documentation: Hosting n8n and self-hosted editions · n8n plans and execution-based pricing.

Queues, capacity and continuity

Queue mode distributes executions among worker processes and uses Redis for coordination. Documentation links this architecture with a shared database and specific requirements for binary files. Scaling involves several components, rather than simply adding a server.

Estimate duration, concurrency and data size. A workflow waiting for external responses may behave differently from one transforming a local list. Recording queue time and processing time helps locate bottlenecks.

Define what happens as volume grows: which tasks may wait, which should be explicitly rejected and who receives the exception. In OSH, an auxiliary document-preparation queue should remain separate from established channels for urgent situations.

Reference documentation: Enable queue mode.

Models, agents and human review

n8n combines model providers, tools, memory and LangChain-based components, alongside MCP connections. A workflow can reserve some operations for rules and use AI elsewhere. Documentation includes human review before an agent executes specified tools.

A useful approval shows the proposed operation and its input data. Authorising a call does not certify the agent’s entire reasoning. Restrict fields it may change and distinguish reading, preparation and execution.

As an editorial example, an agent could suggest tags for a materials library. Review should check the reference, language and content before saving classifications. Where evidence is insufficient, “requires review” may be the correct result, rather than forcing a category to complete the workflow.

Reference documentation: Integrate AI: models, tools, memory and MCP · Human-in-the-loop for tools.

Agent evaluation and output quality

Evaluation guidance covers deterministic checks, model-based assessment, feedback and human review. These techniques observe different dimensions. Correct formatting does not establish accuracy, and a favourable assessment from another model is not professional validation.

For a classification trial, prepare known examples across languages, incomplete texts and ambiguous cases. Compare outputs with a written criterion and record corrections. Also assess when the system recognises that it cannot resolve a case.

When changing the model, instructions or tools, repeat relevant cases. Preserving dates, versions and results distinguishes real improvement from impressions based on a few responses. Acceptance of an OSH workflow should consider information quality and the review workload left with the team.

Reference documentation: Evaluating AI agents.

Plans and total operating cost

The consulted commercial offering primarily measures monthly workflow executions, with unlimited steps subject to plan conditions. Offerings differ in concurrency, projects, history, environments and governance. Community should not be assumed to include every enterprise feature.

Total cost includes hosting, administration, backups and connected services. For AI solutions, add model consumption and exception-review work. Running many nodes in one workflow does not make all consumed resources free.

A useful comparison runs the same case under equivalent conditions: input count, size, acceptable latency and recovery. Measuring cost per complete, traceable outcome avoids choosing on an isolated subscription fee that does not represent actual operation.

Reference documentation: n8n plans and execution-based pricing.

Security, updates and historical advisories

n8n publishes information about its security programme and SOC audits, with an available SOC 3 report and SOC 2 access for enterprise customers. Hosted-service controls do not replace the configuration required for a self-managed installation.

In February 2026 it published a critical fix for an expression vulnerability that could allow command execution by authenticated users with editing permissions. The advisory identifies affected and patched versions. This historical incident reinforces the importance of updates; it does not automatically describe a current version’s condition.

Review current advisories, restrict editing permissions and maintain a tested update and restoration procedure. Workflows can access credentials and external systems, making platform administration part of the security of the whole integration.

Reference documentation: n8n security programme · Expression sandbox advisory, February 2026 · Current n8n security advisories.

Community and reception

JavaScript Rising Stars ranked n8n first for annual GitHub-star growth in 2025, with over 112,000 added. This indicates developer and community interest, rather than paid-customer numbers or proven business outcomes.

The community contributes examples, issues and extensions. Before reusing someone else’s solution, check its origin, dependencies and permissions. Template popularity does not establish that its connections or rules suit an OSH environment.

Professional assessment should combine documentation, internal trials and maintenance capability. A flexible tool may contribute substantially to a technically capable team while requiring more effort than expected from another. This explains why the same platform can fit two organisations differently.

Reference documentation: JavaScript Rising Stars 2025 · n8n source repository and project background.

How it can support OSH tasks

These cases are Sabentis editorial proposals, not preconfigured integrations or preventive capabilities attributed to n8n:

• Data preparation: transform site codes and dates before bringing authorised records into an analysis tool.
• Reference checks: verify that a material request contains an identifiable version and route exceptions for review.
• Reconciliation: compare identifiers between source and destination to detect missing records, without interpreting a transfer failure as an absence of preventive activity.

The workflow should separate data transport, validation and professional decisions. It should not automate an individual risk conclusion from incomplete fields or copy health information for convenience.

Hypothetical example: preparing a monitoring report

An organisation wants to collect action references monthly from an authorised source. The process does not alter source case files.

1. Define an input contract: action identifier, site code, date, status and extraction date. Document what each status means.
2. Check data types and accepted codes. Set incomplete records aside for review instead of inventing values.
3. Transform valid records into the destination format, retaining the original identifier and reporting cut-off date.
4. Test a failed connection and a repeated record. Recovery should identify what was written before the error and prevent a second copy.
5. Reconcile received records against accepted and rejected ones. Check a sample against the source before using the report.

An acceptable output includes both the prepared dataset and an exception list. A report with no visible errors may still be incomplete unless expected totals and extraction age are checked.

Errors, hosting and operational ownership

n8n can associate an error workflow with another workflow; it starts with Error Trigger and runs when an execution fails. The documentation also covers reviewing previous executions and deliberately stopping a workflow with Stop And Error under a defined condition.

This helps expose failures, but someone must own their resolution. Document execution-data retention, restoration, updates and credential custody. For self-hosting, include server capacity and backups. A workflow that successfully sends a request to an external service still needs to check the response and the persisted result.

Reference documentation: Handle errors gracefully.

When to choose n8n

Consider it where technical capability exists to maintain integrations and specific deployment or transformation choices are needed. Criteria should include service continuity, documentation and replacement of the person maintaining it.

Make and Zapier provide alternative approaches for comparison. A useful comparison uses the same input volume, exceptions and recovery test. The most flexible tool is not always the one requiring least maintenance.

Frequently asked questions about n8n

Does hosting it on my server guarantee privacy?

Not by itself. Review permissions, backups, execution logs and external destinations. Server location is only one part of data handling.

Does an error workflow automatically fix the problem?

Not necessarily. It defines a response to failure; resolving an expired credential or incorrect input may require intervention.

How can missing records be prevented?

Use stable identifiers, source-to-destination reconciliation, exception management and tested recovery. Counting executions is not the same as counting complete records.