n8n Gateway Credits: Pay for AI Without an API Key on n8n Cloud (and When to Avoid It)

Building an AI agent on n8n Cloud often means juggling several provider accounts before you write your first prompt: an OpenAI or Anthropic key for the model, a Firecrawl key for scraping, another for web search. Each account comes with its own billing, its own saved card, its own key to secure. Since version 2.36 (released August 18, 2026), n8n Cloud offers an alternative: Gateway credits, a single prepaid balance that lets you use AI models and a handful of third-party services directly, with no provider account required. Here's what actually changes, and why it isn't always the right call.

How it works

On a node that supports it, a new option now sits next to the usual credential setup: "Use Gateway credits." Selecting it means the request no longer goes straight to OpenAI or Anthropic with your own key — it routes through a proxy operated by n8n, which authenticates with the provider on your behalf and deducts the request's cost from your instance's credit balance. No key to generate, no account to create, no card to register anywhere other than your n8n Cloud subscription.

The balance is shared at the instance level: everyone working on it draws from the same pool, visible directly in the editor next to the relevant node, as well as in a dedicated tab of the Cloud admin dashboard.

Providers and tools covered

On the language-model side, Gateway credits cover OpenAI, Anthropic, Google Gemini, Alibaba Cloud Qwen, MiniMax and Moonshot Kimi — enough to power an AI Agent without touching its own credentials. On the tool side, coverage extends to Brave Search, Firecrawl, Browserbase, LlamaParse and PDF.co — a good chunk of what typically gets wired into an agent for web research or document extraction.

Each Cloud plan includes a monthly credit allowance (in the low thousands depending on the tier, with a smaller amount during the free trial that expires when the trial ends). Beyond that, you either upgrade or wait for the monthly renewal.

Cloud-only, absent from self-hosted

This is the point that matters most for this blog's self-host audience: Gateway credits don't exist outside n8n Cloud. On a self-hosted instance, including the Community Edition, you keep creating your own credentials with your own API keys exactly as before — see our guides on connecting OpenAI or connecting Claude and GPT to n8n. If you chose self-hosting for cost control or infrastructure ownership — a trade-off covered in our n8n self-hosted vs Cloud comparison — this feature changes nothing for you.

The real trade-off: convenience, not savings

n8n states it aligns Gateway credit rates with providers' public list prices, with no particular discount. In other words, this isn't a way to pay less for AI — it's a way to avoid creating and securing extra provider accounts. That's a useful distinction to keep in mind before switching an existing workflow over, especially if you already track the real cost of your AI API calls closely or have set up a circuit breaker to cap an agent's spend.

When it makes sense

The scenario where Gateway credits genuinely help is fast onboarding: a training workshop, a client demo, a prototype you want to show in five minutes without asking a client to create an OpenAI account they don't have. It's also handy for an agency building a first agent for a client before that client has decided which AI provider they want to commit to long-term, on their own pricing terms.

When to keep your own API keys

For a production workflow, three reasons argue for keeping your own credentials over Gateway credits. First, cost control at scale: with your own key, you pick your provider, your exact model, and your own overage alerts, instead of depending on a shared monthly allowance. Second, access to models outside the gateway's scope: Gateway credits cover neither self-hosted models via Ollama nor multi-provider routing through OpenRouter, useful for automatic fallback if an AI provider goes down.

Third, there's a platform-dependency question worth asking plainly. Routing every AI call through a proxy managed by n8n, rather than through a direct relationship with the provider, adds a layer of infrastructure you don't control: pricing, availability or terms can change unilaterally, and switching automation tools later becomes a heavier migration. This is exactly the mechanism described in a landmark study by Opara-Martins, Sahandi and Tian, published in 2016 in the Journal of Cloud Computing: as more resources and integrations migrate to layers managed by a single provider, vendor lock-in risk increases, often without the teams involved anticipating it at the time of adoption (Opara-Martins, Sahandi & Tian, 2016, Journal of Cloud Computing). Using your own API keys, by contrast, keeps the contractual relationship directly between you and the model provider.

The compliance angle: where your data actually goes

For sensitive use cases — personal data processing, confidential documents in a RAG pipeline — who handles the request before it reaches the model deserves documentation, not a shrug. With your own keys, your GDPR record of processing activities describes a direct flow to the model provider. With Gateway credits, an extra technical intermediary — n8n's proxy — sits in that flow. That's not necessarily a problem, but it's one more link to add to your data-processing map, alongside the other ISO 27001 or SOC 2 compliance evidence you're already documenting.

Wrapping up

Gateway credits solve a real friction point — juggling multiple provider accounts — but only on n8n Cloud, with no pricing advantage, and at the cost of an extra intermediary in your data flow. For a quick test or a demo, it's a genuine time-saver. For a production workflow, especially on compliance-sensitive topics, keeping your own API keys remains the clearer choice — whether you run on Cloud or self-hosted. That's exactly how our AI Inbox Pack (€79) and RAG Assistant Pack (€119) are designed: your own API keys, your own Supabase database, an architecture you understand and control end to end.