n8n Review (2026): Honest, Evidence-Based Verdict

What n8n is, in one paragraph

n8n (pronounced "n-eight-n," short for "nodemation") is a workflow automation platform built around a visual canvas of connected nodes, but with three things most rivals don't offer together: you can write real code (JavaScript or Python) inside any workflow the moment the no-code path runs out; you can self-host the free Community Edition on your own server; and it ships serious AI-agent tooling, not a bolted-on "AI step." It has 500+ native integrations, plus any API through the HTTP Request node, and is fair-code (source-available). n8n says the project has over 200,000 GitHub stars (208k). That signals how deeply the developer and automation community has adopted it.

Our verdict and how the score reads

That headline number isn't a gut call. It's a weighted composite of eight axes (AI, value, capability, ease, integrations, user sentiment, docs, and support), each scored from the vendor's own documentation and pricing, independent review platforms, and the recurring themes across real user reviews. We weight value and capability most heavily because they decide whether a tool earns its place, and we publish every axis and its rationale above rather than hand you a single opaque star rating. Where a figure isn't independently verifiable we say so, and we never average a single platform's reviews and call it our verdict. Full method: how we test.

n8n's score is the honest shape of the tool: it tops the category on capability, ties for the highest value score and scores high on AI, where Latenode scores higher. It pays for that with a lower ease-of-use score than the beginner-first tools. That's the trade in one line: power and ownership in exchange for a steeper curve.

The AI layer — agents, RAG and model choice

If you're evaluating n8n in 2026, the AI story is the headline. It isn't a single "ask AI" action bolted onto an automation tool. It's a genuine framework for building agents. n8n's AI Agent node (LangChain-based) lets you assemble an agent that reasons, uses your other nodes as tools, and loops until it accomplishes a goal, all on the visual canvas, then refined in code where you need it^[1]. Around it sits everything a serious AI workflow needs: direct vector-store tools for Pinecone, Qdrant and Supabase, embedding nodes, memory, and output parsers.

That means retrieval-augmented generation (RAG), the pattern behind most useful business AI, is a first-class citizen. A typical n8n RAG pipeline watches a document source (say a Drive folder), splits and embeds the documents, writes the vectors to a store like Qdrant, and then, at query time, the AI Agent node loads that store as a retriever tool and answers questions with citations^[1]. Real teams are shipping exactly this:

• Internal knowledge assistants: an employee asks a question in Slack; the agent runs a similarity search over your wikis and policies in a vector store and returns an answer with citations, not a hallucination.
• Compliance and contract review: ingest policies into a RAG pipeline, then check a proposed process or contract against them; used in finance, healthcare and legal.
• E-commerce price monitoring: a scheduled agent scrapes competitors, uses an LLM to normalize product names, extracts pricing and stock, and recommends adjustments.
• Ticket triage: an agent reads a Zendesk ticket, classifies it, then replies, escalates, or opens a Linear issue depending on what it finds.

And the provenance question that's a caveat for most tools is a strength here. You can bring your own model and key on every plan (OpenAI, Anthropic, open-weight, even a locally hosted model via Ollama when self-hosting), and on Cloud Starter and Pro, Gateway credits (since n8n 2.36, Sep 2026) run models from OpenAI, Anthropic, Gemini and others without your own keys^[2]. Self-hosted with a local model, your data stays on your infrastructure. With a cloud model provider, your prompts go to that provider. n8n has also added standalone n8n Agents and the n8n Assistant, which builds and fixes workflows for you. Both are in Preview.

Capability and the code escape hatch

Underneath the AI, the fundamentals are built for control. The canvas handles branching, merging, loops, error handling and retries. And, the part that matters most for a technical team, a Code node running JavaScript or Python sits one click away throughout^[3]. When a no-code-only tool hits "not supported," n8n lets you write your way through: transform data any way you like, call an API no native node covers, or build a custom node (custom nodes run on self-hosted n8n only). Self-hosted. That ceiling is very high. On Cloud it is lower: the Code node can import only the crypto and moment modules, and native Python can import no libraries at all.

Value, and the two paths that define your cost

n8n is exceptional on value, but you have to pick the right path. The self-hosted Community Edition is free and has no execution quota, but it lacks SSO, projects, sharing, Git and environments, and n8n recommends self-hosting only for expert users. You run it on your own server (Docker is the common route, ideally with a Postgres database for production) and pay that server's cost. n8n publishes no figure for it. Its guides size a small server, and ask for at least 4 GB RAM and 2 vCPUs if you add the n8n Assistant.

n8n Cloud is managed hosting billed per execution: one production run of a published workflow, regardless of how many steps or nodes it contains. n8n's US pricing page lists Cloud Starter at $20 a month billed annually or $24 a month billed monthly (2,500 executions) and Cloud Pro at $50 a month billed annually or $60 a month billed monthly (10,000 executions), after a 14-day free trial with no card, Pro features and 1,000 executions^[4]. Business ($800 a month billed annually or $960 a month billed monthly for 40,000 executions) is a self-hosted licence, and SSO on Cloud is Enterprise-only.

Annual billing saves about 17%, and since August 2025 every plan includes unlimited users, workflows and steps. Executions are the main Cloud meter, with conditions. Over the limit, Cloud executions fail immediately (not queued) until the 1st of the month or an upgrade, and extra executions exist only on Pro-2 billed monthly, at $2 per 1,000. AI adds a second meter: Gateway credits on Cloud Starter and Pro, or your own provider's bill. n8n Assistant credits are a separate allowance. These are the US figures: n8n bills US addresses in USD and its pricing page shows prices for your browsing location, so readers elsewhere may see another currency; our pricing guide has the full table.

The honest guidance: if you have the technical capacity to run a server well, self-hosting can cost less and keeps your data in-house. If you don't, Cloud buys back the operations and is a sensible trade. The free edition only pays off if you'll actually maintain it. And here is the strategic escape hatch most tools can't offer: if Cloud executions ever get expensive at your volume, self-hosting the free Community Edition removes the execution meter. The paid self-hosted Business licence is metered too, at 40,000 executions a month.

Is it really free? The fair-code question

Buying committees ask this first, so here's the plain answer. n8n is fair-code, released under the Sustainable Use License: source-available, and n8n says it isn't open source under the OSI definition. For your own business it is free: you can self-host it, use it internally, modify it and build on it at no licence cost. You can also build and run automations for clients and charge for setup, maintenance and consulting, as long as those clients can't open the editor or change the workflow logic.

The limits matter for agencies and SaaS builders. The licence forbids hosting n8n as a service, letting clients or end users build or edit workflows through the editor, a custom UI, an API, MCP or an AI agent, forking n8n into your own automation product, and white-labelling it. If your intent is run our own automation, there's no catch. The licence covers self-hosted n8n only. n8n Cloud has its own, narrower terms, so check those before doing client work on Cloud.

Scaling — the part most reviews skip

n8n runs in two modes, and knowing the difference saves you a nasty surprise at scale. Regular mode (the default) runs executions in the main process and, on self-hosted n8n, sets no concurrency limit unless you configure one. When you need more throughput or hit spiky load, you move to queue mode, where the main instance creates executions and separate worker processes run them, coordinated by Redis. For scale, n8n says a single instance can handle up to 220 workflow executions per second, and more with additional instances. Queue mode is free on self-hosted. On Cloud it is an Enterprise option, enabled by contacting n8n.

The catch is that queue mode is an infrastructure project, not a toggle. It wants a Redis instance and a database (Postgres), and filesystem binary data isn't supported in queue mode; external S3 or Azure storage needs a Business or Enterprise licence. Multi-main mode, the high-availability setup, is self-hosted Enterprise only and not available on Cloud. This is the clearest example of n8n's core trade: the power is there, but at scale you (or your infrastructure) do the work.

User sentiment — read across the sources

The number most review sites stop at, user sentiment, is for us one axis out of eight. Here is how the review platforms rate n8n today:

The larger signal for a tool like this is its community as fair-code software: n8n says the project has over 200,000 GitHub stars (208k) and a community of more than 200,000 developers^[4]. In our assessment the split by user type is predictable. Developers and power users get the most from the depth, the code escape hatch and the AI agents. Newer users, especially on Cloud, should expect a learning curve and a debug experience that assumes you understand the data structure flowing between nodes. Read together: a deeply capable tool that rewards technical fluency.

Where n8n is weak (the honest part)

A steeper learning curve. n8n is developer-oriented. The node-and-data model is powerful but assumes more technical fluency than beginner-first tools. In our assessment it's where newer users will spend the most time, so budget real ramp-up time, especially on Cloud.

Debugging assumes data-structure fluency. A downstream node often shows empty output when the data shape upstream isn't what you expected. The execution view is excellent once you can read it. Reading it is a skill, though. Inspect each node's output rather than guessing when something breaks.

Self-hosting is real operations. The free edition is free of licence cost, not of responsibility: you own updates, backups and security. One specific trap: n8n encrypts stored credentials with a key generated on first run, so if you rebuild or restore without saving that key, every credential is unrecoverable. Save it on day one.

Scaling and enterprise HA take work. As covered above, queue mode is a migration with real infrastructure, and multi-main high availability is self-hosted Enterprise only. n8n is superb for a technical team. It is not a zero-effort enterprise appliance.

A narrower native catalog. 500+ native integrations cover the mainstream stack but trail the largest players. The HTTP Request node and code reach almost any API. It just may take a build step.

Who should use it — and who shouldn't

n8n fits developers and technical teams who want control and code, privacy- and compliance-sensitive teams who need to self-host so data stays in-house, and AI builders assembling agentic, RAG-powered workflows with a model they choose. If you have no technical capacity to set up and maintain it, or you want the gentlest onboarding and the widest zero-setup catalog, a hosted no-code leader may serve you better. We'd say so plainly.

The bottom line

n8n is a top control-and-AI pick in automation. It does the fundamentals on a visual canvas, lets you drop to code the instant you need it, ships serious AI-agent and RAG tooling with a model you choose, and, rare in this space, its free Community Edition runs on your own infrastructure with no execution quota, though without SSO, projects, sharing, Git or environments. The catches are the learning curve, the operations that come with self-hosting, and the infrastructure work that scaling demands. So the one thing to weigh before committing is whether you have the technical capacity to use that power. If you do, few tools give you more. Fewer still let you own it outright. New to it? Start with our tutorial; to weigh cost, see the pricing guide; building with AI, read the AI playbook.