Quick answer#
n8n AI Agent Builder is worth shortlisting when you want agents inside inspectable business workflows, not a promise of hands-off automation. Its new Agents preview adds conversational agents alongside the established AI Agent node. The September 21 release changes the comparison: n8n is no longer only a canvas for connecting isolated AI steps.[14]
Verified October 3, 2026. In our EUR browser session, Starter displayed €24 monthly or €20 per month billed annually; Pro displayed €60 monthly or €50 per month billed annually. Those are platform prices, not unlimited model usage.[1]
- Best for: technically comfortable operators who need app integrations, repeatable routing, inspectable executions, and selected reasoning steps.
- Avoid if: you expect a free, fully maintained production agent, unrestricted commercial redistribution, or guaranteed correctness without evaluation.
- Our recommendation: pilot a narrow workflow with read-only tools first. Buy annual access only after measuring execution volume, model charges, and recovery work.
This is a research-based buyer review, not a hands-on reliability benchmark. We inspected official pages, pricing controls, terms, release notes, a community report, and a live ad preview. We did not purchase a plan, run a production n8n agent, or test cancellation. This site promotes Hermes and managed hosting; that commercial interest does not make Hermes the right replacement for every workflow.
What exactly are you buying?#
The operator is n8n GmbH, identified in its self-serve terms as a Berlin company registered under HRB 212509 B.[10] “AI Agent Builder” describes capabilities in the n8n platform rather than a separately priced course or certification.[1][2]
Keep three things separate. The AI Agent node is an agent step in a workflow. The newer Agents preview gives an agent instructions, tools, skills, sub-agents, knowledge files, memory, and stored conversations. Assistant helps build and manage automations and has its own credit allowance.[1][14]
The changelog says Agents and workflows can call one another. An agent can call a workflow with that workflow's credentials; a workflow can call an agent through Message an Agent. Existing AI Agent node workflows remain supported.[14] This combination is the practical reason to consider n8n: keep business rules explicit where they should be explicit, and allow model judgment only where it adds value.
For the narrower architectural choice, our Hermes Agent versus n8n comparison owns that separate intent. This dossier covers the purchase, price, complaints, and operating obligations.
Pricing: currency, commitment and execution limits#
Our live browser displayed the following EUR base selections. Prices can depend on region, selected volume, and billing interval; check the final order. The text extractor separately returned dollar-denominated cards, including a different Business figure. We do not treat a dollar sign as interchangeable with euros.[1]
- Starter: €24 monthly, or €20 per month billed annually; 2,500 workflow executions, one shared project, and five concurrent executions.[1]
- Pro: €60 monthly, or €50 per month billed annually at the selected 10,000-execution tier; three shared projects. Higher selected volumes can change the price and capacity.[1]
- Business: €800 monthly, or €667 per month billed annually at 40,000 executions in the EUR browser; self-hosted, with a Cloud waitlist rather than an included hosted instance.[1]
- Enterprise: contact sales; deployment, execution allowance, support and contract need an individual quote.[1]
- Community Edition: a self-hosted version is available; your hosting, operations, model use and license suitability still need budgeting.[1][11]
The €20 annual-billing headline is a monthly equivalent, not the monthly-cancellable price. Listed Cloud prices exclude taxes under the self-serve terms.[10] We did not complete checkout or verify the tax-inclusive total for a particular country.
The product page advertises a 14-day Cloud trial without a credit card.[2] The pricing FAQ gives the trial 1,000 executions, five concurrent executions, a 180-second execution timeout, one project, 800 Assistant credits and 100 AI credits. Business trial requirements differ and include a card.[1] A trial is therefore not a production-capacity guarantee.
Hidden running costs: four different meters#
Execution billing does not make AI inference free. n8n describes a workflow execution as an entire run rather than each individual step.[1] For the new Agents preview, the changelog is more specific: one turn counts as one execution, and its workflow-tool and sub-agent calls do not count separately.[14] Do not extrapolate that preview rule to every unrelated workflow design.
Build your budget around these distinct categories:
- Platform executions: triggers, conversation turns, and your actual workflow architecture determine consumption. Estimate from representative traffic, not the number of boxes on the canvas.
- Model usage: Agents can use a model for which you have credentials or Gateway credits. A tool-heavy turn may produce several model calls even when the platform counts one execution.[14]
- Builder credits: Starter includes 1,600 monthly Assistant credits; Pro includes 4,000 or 9,600 depending on tier. Unused monthly credits do not roll over, and extra credits can be purchased.[1]
- External services and operations: paid app accounts, search, databases, storage, hosting, backups and the human time required to investigate failures belong in the total. The terms explicitly separate third-party subscriptions from payments to n8n.[10]
Do not buy a chat subscription assuming it covers an API-backed automation. Ask which key or Gateway allowance each agent uses, and reconcile usage against provider billing. Our AI-agent cost guide is a starting point for that separate operating budget, not a quote for n8n.
Business overages deserve particular attention: the pricing FAQ says workflows continue, overages may apply, and quotes €4,000 for additional buckets of 300,000 executions.[1] Get the applicable currency and contract treatment confirmed before choosing Business because you expect uninterrupted overage handling.
Features that matter, and what remains preview#
The new Agents preview supports connected channels, schedules, reusable workflows as tools, published versions alongside editable drafts, and approval requirements on sensitive tools.[14] That is more capable than a simple prompt-to-workflow generator. It is also explicitly preview software: n8n warns that agents can make mistakes and behavior can change.[14]
Useful capabilities to evaluate include:
- A deterministic path for validation, routing and error handling, rather than asking a model to improvise every step.
- Human approval attached to tools that change records or send consequential messages.[14]
- Per-session tool-call records and workflow execution history for investigating what happened.[1][14]
- Self-hosting when infrastructure control is genuinely necessary, with an owner for patching and recovery.
- Model choice and the ability to keep existing workflows while adding agents.[14]
Assistant itself supersedes AI Workflow Builder: the changelog describes planning, building, testing and iterating, with approval for high-impact actions such as publishing.[14] Treat those as documented capabilities, not evidence that every generated workflow will be safe or correct.
If the right design uses both products, the n8n integration guide separates workflow handoffs from replacing your automation backbone.
License and self-hosting: free is not unrestricted#
n8n calls its model fair-code and source-available. Its Sustainable Use License permits internal business, non-commercial or personal use, while restricting redistribution and other commercial uses. The documentation says uses outside the license require a separate commercial agreement.[11]
That is not the same as an unrestricted open-source license. Before embedding n8n in a customer-facing product, reselling hosted access, or letting customers connect accounts and build workflows, get the actual use case assessed against the license. Do not infer permission from GitHub availability.
Self-hosting also does not eliminate paid feature boundaries. Business is self-hosted and still metered; its pricing FAQ says licensed instances contact the license server daily and combined instance usage counts toward the quota.[1] Infrastructure ownership and software entitlement are separate decisions.
A timely migration issue makes maintenance concrete. The 3.0 preparation page says the upcoming October 2026 transition will require Docker-based self-hosting and remove runnable npm/npx installations. It also removes version 1 of the AI Agent node and its older agent modes.[15] This was upcoming guidance when checked, not a claim that 3.0 had already shipped. Inventory templates, export backups, and rehearse migration before upgrading production.
Cancellation and refund terms#
The refund policy for monthly subscriptions says you can request that month's payment back within 14 days of the billing date, and the account must be cancelled to qualify.[9] The self-serve terms say annual payment is non-refundable and paid in advance; monthly cancellation retains access until the end of the month.[10]
Keep plan-specific rules separate. The Business pricing FAQ says cancellation takes effect at the end of the applicable monthly or annual term and does not offer pro-rata refunds.[1] Do not treat the monthly help-center policy as a universal Enterprise or Business money-back guarantee. Ask support which policy governs the exact order and preserve that answer before paying.
We did not test refund processing, response times, or account deletion. Mandatory consumer rights and individually agreed contract terms may affect your position; this is a purchasing checklist, not legal advice.
Is n8n legit? Complaints and evidence limits#
The identified operator, detailed terms, published source code licensing, and active documentation support treating n8n as a real software supplier.[10][11][14] They do not establish that your automation will work reliably or earn money.
A concrete community report is more useful than a generic score. In the thread “How to access tokenUsage from a sub-node Model within an AI Agent?”, the author reports that DeepSeek token usage was visible in the model sub-node but absent from the AI Agent output despite Return Intermediate Steps. The reported setup is n8n 2.16.1 on Docker and Ubuntu.[8]
That is a version-specific user report, not our reproduced defect and not proof the current Agents preview has the same behavior. Its purchasing lesson is nevertheless sound: verify that your chosen configuration exposes enough usage data to bill clients or audit spend. Never assume the final agent answer is an accounting record.
The product page's customer testimonials are selected by the vendor.[2] We did not establish a representative complaint rate, independently validate time-saving claims, or run a comparative benchmark.
Advertising: what we actually verified#
The live Google Ads Transparency preview for n8n GmbH showed “Get Started for Free - AI Workflow Automation - n8n.io” and a last-shown date of October 2, 2026. Creative ID CR13300963693610139649 belongs to advertiser AR00517148705644609537.[16]
The live format was Text, despite an older local advertising export classifying it as Video/YouTube. We prioritize the current preview and do not claim verified YouTube placement. “Get Started for Free” should be read alongside the limited Cloud trial and self-hosted licensing options, not as unlimited free managed production.[1][11][16] Ad presence is not evidence of product quality, budget, or conversion performance.
Pros, cons and the strongest alternatives#
Pros: an inspectable workflow foundation, reusable tools, documented approval gates, self-hosting options, and a newer conversational-agent layer.[1][14] The clearest fit is an operator who needs both predictable integrations and limited AI judgment.
Cons: several cost meters, preview changes, license restrictions, paid governance boundaries, and operational work if self-hosted.[1][11][14] The platform can reduce integration work without removing engineering responsibility.
Consider Zapier Agents when a hosted app-connected agent is the more natural purchase, but compare its activities against n8n executions rather than comparing headline allowances. For regulated enterprise governance and a procurement-led platform evaluation, our Kore.ai Artemis review covers a substantially different product category.
Hermes is worth considering for a configurable personal-agent workflow rather than a visual automation backbone. If you choose that path, use the installation guide for self-management or the managed cloud option when infrastructure work is the obstacle. Neither is a drop-in replacement for n8n's canvas, shared workflow operations, or enterprise entitlements.
Decision checklist before subscribing#
- Pick one actual business outcome and keep the first version read-only.
- Separate deterministic workflow steps from decisions that need a model.
- Confirm the exact plan, currency, interval, execution quota and timeout.
- Identify whether runtime inference uses your keys or Gateway credits, separately from Assistant credits.
- Prove that usage records reconcile with the bill and that failures do not duplicate consequential actions.
- Test denied approvals, expired credentials, provider errors, and safe recovery before publishing.
- Confirm license fit and the owner of backups, patching, migrations and incident response.
- Read the applicable refund terms before trading monthly flexibility for an annual headline price.
Bottom line: choose n8n because you need its workflow architecture and can operate it responsibly. The new Agents preview makes it more interesting, not exempt from testing, cost control or maintenance.
Sources#
[1] https://n8n.io/pricing — n8n pricing and plan FAQ [2] https://n8n.io/ai-agents — n8n AI agents official product [8] https://community.n8n.io/t/how-to-access-tokenusage-from-a-sub-node-model-within-an-ai-agent/294577 — n8n community token-usage report [9] https://support.n8n.io/article/refund-policy — n8n monthly refund policy [10] https://n8n.io/legal/self-serve-terms — n8n self-serve terms [11] https://docs.n8n.io/privacy-and-security/sustainable-use-license — n8n Community license [14] https://docs.n8n.io/changelog — n8n changelog: Agents preview [15] https://docs.n8n.io/changelog/v30-breaking-changes — n8n upcoming 3.0 migration guidance [16] https://adstransparency.google.com/advertiser/AR00517148705644609537/creative/CR13300963693610139649 — Google ad preview: n8n GmbH