BYO API Key vs. Included AI Credits in n8n Hosting

When you start evaluating n8n hosts, the pricing page usually looks clean: a monthly fee, some compute specs, maybe a note about "unlimited executions." The AI bill isn't on that page. It shows up later — on your OpenAI invoice, or when a workflow silently fails because you hit a rate limit at 2 AM.

There are exactly two models for how AI billing works inside a hosted n8n instance. Understanding the difference before you pick a host saves a lot of retrofitting.

---

Model 1: Bring Your Own API Key (Every Major Host Today)

n8n Cloud, Elestio, Sliplane, and Hostinger all follow the same pattern: they give you the n8n instance, and you wire up AI nodes by pasting your own API key from OpenAI, Anthropic, or whichever provider you use.

This sounds fine in theory. In practice, it creates several compounding problems.

The surprise bill problem

n8n's AI nodes — the AI Agent node, the OpenAI node, the LangChain sub-graph nodes — call the provider API on every execution. If a workflow runs on a schedule every 15 minutes and chains two LLM calls, the token spend adds up fast. And it's unpredictable: a workflow that worked fine for weeks suddenly touches a larger document, or a loop misfires and calls the LLM far more times than intended. Your API spend that day can spike dramatically. Nothing in the n8n interface warns you in advance; you find out when the OpenAI dashboard sends an alert — or when the invoice closes.

The key rotation problem

API keys have a lifespan. Security-conscious teams rotate them periodically. Every rotation means:

1. Generate a new key in the provider dashboard.
2. Find every n8n credential that references the old key.
3. Update it. Test each workflow.
4. Revoke the old key.

With one workflow this is a short job. With many workflows across multiple credential sets, it becomes a real operational burden. And when you need to migrate your instance to a different host, you're re-entering all of them.

The rate-limit problem

OpenAI, Anthropic, and Google all enforce rate limits per API key — requests-per-minute and tokens-per-minute ceilings that vary by account tier. New API accounts start with low ceilings. A workflow with a loop that processes many items simultaneously will hit the RPM limit, get a 429 error, and either retry-storm or silently skip rows depending on how the error handler is set up.

Unlocking higher rate-limit tiers requires meeting cumulative spend thresholds on the provider's platform. You pay to unlock the right to pay more.

The model-lock problem

When you hard-code a specific model into a credential, switching to a different provider means creating a new credential set, updating every node that used the old one, and re-testing. Most teams end up locked on whatever model they started with because the switching cost feels higher than it's worth.

---

Model 2: Included AI Credits (What It Actually Looks Like)

The alternative is a host that bundles AI spend into the subscription price itself. Instead of connecting your own API key, the AI nodes in your n8n instance point to the host's LLM proxy — already authenticated, already paid for as part of your plan.

The practical differences:

No credential setup. When you drop an AI Agent node or an OpenAI node into your canvas, you select the host's built-in credential. It works immediately. No key, no provider account, no billing profile to set up.

No surprise bills. Your monthly cost is the subscription. The AI usage is covered within the plan's included credit allocation. You know what you're paying before the month starts.

Model switching without key juggling. If the host's proxy surfaces 350+ models, switching from GPT-4o to Claude Sonnet to Llama 3.3 is a dropdown change in the node — not a credential migration. You can run A/B prompt tests across models inside the same workflow using an IF node or a Switch node to route to different model names.

No rate-limit cliff from fresh accounts. The host aggregates usage across their infrastructure. You're not starting from zero with rock-bottom limits on a brand-new provider account.

---

Head-to-Head Comparison

Factor	BYO API Key	Included AI Credits
Upfront setup	Paste key, set up billing	None — works out of the box
Monthly cost predictability	Variable (usage-dependent)	Fixed with the plan
Surprise bills	Yes — loops, doc size spikes	No
Key rotation burden	Manual, across all credentials	N/A
Model switching	New credential per provider	Dropdown in the node
Rate limits	Per-account (start low)	Host-aggregated
New-user friction	Medium (need provider account)	Low
Control / BYOK option	Full	Depends on host

The BYO model is not wrong — it makes sense if you already have substantial API credits from a prior deal, if you need a specific model the host's proxy doesn't expose, or if you want direct provider relationships for compliance reasons. But for most teams running n8n for internal automation, the included model removes an entire layer of operational overhead.

---

How to Tell Which Model a Host Uses

The easiest signal: look at the "get started" docs. If the first few steps include "go to OpenAI, create an account, add a payment method, generate an API key," it's BYO. If the AI nodes are described as working without any credential setup, credits are included.

A secondary signal: check whether the host's pricing page mentions anything about AI or LLM usage. BYO hosts have nothing to say about it because it's not their problem. A host with included credits will call it out because it's a meaningful part of the value.

---

How It Works on AgentRoost

AgentRoost gives you your own n8n instance — your login, your workflows, your data, running at https://<your-id>.agentroost.app. You own it, same as self-hosting, without the DevOps.

The AI credits are included in the subscription. Here's the actual journey:

1. Sign up at agentroost.app — email, Google, Microsoft, or Discord.
2. Pick the n8n framework, give your instance a name.
3. Your private n8n editor opens. Build your first workflow.
4. Drop in an AI Agent node or an OpenAI node. Select the built-in AgentRoost credential. It works — no key, no provider account.
5. Need a different model? Switch the model name in the node. 350+ options, same credential.

Webhooks get a public HTTPS URL automatically. No Nginx config, no certificate renewal. The AI nodes are already wired.

Plans start at $19.99/mo all-in — compute, hosting, and AI credits in one line item. There's a 14-day money-back guarantee and you can cancel anytime.

If you're evaluating whether the included-credits model fits your workload before committing, the pricing page breaks down what each tier includes, and the n8n page walks through what's pre-configured out of the box.

Compare plans · See the n8n instance details