AI Gateway

Two lines of code

The gateway is OpenAI-compatible. If your code, framework or off-the-shelf tool can talk to OpenAI, it can talk to api.node.uk: swap the base URL, swap the key.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.node.uk/api/v1/models/qwen3.6-27b/v1",  # line 1
    api_key=NODE_GATEWAY_TOKEN,                                   # line 2
)

reply = client.chat.completions.create(
    model="qwen3.6-27b",
    messages=[{"role": "user", "content": "Where does this prompt get processed?"}],
)

Each model is served under its own path, GET https://api.node.uk/api/v1/models returns the live catalogue, and streaming works as you would expect. Credentials are issued from your portal and backed by your tenant's Keycloak SSO realm.

One endpoint, many routes, one price list

Behind the endpoint we aggregate a number of external providers and route each request based on cost and availability. You never deal with provider accounts, keys or invoices - we hold the routes, you hold one credential and one bill.

The pricing policy is deliberately simple:

• Every model is listed once, at our price for the cheapest route that can serve it \u2014 quoted in USD per million tokens (per minute for audio) so you can compare us directly with any other provider, with the GBP equivalent in brackets. Billing is in GBP.
• The published price is what you pay. If we have to fail over to a more expensive route in an emergency, we absorb the difference - failover never charges you extra.
• Prices update daily against current route costs and the live USD→GBP exchange rate. The catalogue was last generated on 2026-07-08.

The models split into two residency classes, and we label every model with its class:

• UK-hosted - running on GPUs we own in our UK datacentre. Requests are processed entirely on Node infrastructure, under UK jurisdiction. No US CLOUD Act exposure, no third-party AI provider in the data path.
• Partner-routed - routed to a vetted partner provider and processed on that partner's infrastructure. These requests leave our infrastructure, and the UK-residency guarantee does not apply to them.

Popular models

A snapshot of what people actually run through the gateway, with our current prices. Audio is priced per minute; everything else per 1M tokens (input / output).

Browse all 305 models with prices → or read which model for which task → if you are not sure where to start.

Metered onto one bill

There are no subscriptions, seats or minimum commitments on the gateway. You pay per million tokens, metered per request: every call is logged with its token counts, attributed to the credential that made it, and rated onto the same monthly invoice as your apps. Usage is visible as it happens, and your £15 welcome credit works on the gateway from day one.

Because every person, service and agent gets its own Keycloak-issued credential, your invoice tells you who spent what - useful the moment you have more than one developer, and essential the moment you have autonomous agents making calls.

Built on the platform, not beside it

The gateway is a first-class part of the Node Platform: APISIX at the front, Keycloak for identity, per-request usage events feeding the billing pipeline. We run our own company's AI traffic through it - every summary, doc draft and coding agent - which is exactly the treatment your traffic gets. If you want private models dedicated to your tenant rather than the shared catalogue, that is a managed AI deployment and we should talk.

---

Frequently asked questions

Where is my data processed when I use the gateway?

It depends on the model you call, and we label this per model. UK-hosted models run on GPUs we own in our UK datacentre: prompts and completions are processed there and never leave our infrastructure. All other models are routed to a vetted partner provider, so prompts and completions for those requests are processed on the partner's infrastructure and leave ours. The UK-residency guarantee applies only to the UK-hosted models, and every model page states which side of that line it sits on.

How does routing and failover work?

We aggregate a number of external providers behind one endpoint and route each model to the cheapest route that can serve it reliably. The price we publish is the price you pay. If a route fails and we have to fail over to a more expensive alternative, we absorb the difference - failover never charges you extra.

How are prices set, and how often do they change?

Each model's price is the cheapest available route cost plus our margin. We quote in USD so you can compare us line-by-line with other providers, and show the GBP equivalent in brackets — invoices are in GBP. Prices are regenerated every day against current provider rates and the live USD to GBP exchange rate, and the catalogue shows the date it was generated. What you see on the model page is what lands on your invoice, per million tokens (or per minute of audio).

Is the API really OpenAI-compatible?

Yes. The gateway speaks the OpenAI API - chat completions, embeddings and audio transcription, including streaming - so existing SDKs and tools work by changing the base URL and API credentials. Each model is served under its own path, and GET /api/v1/models returns the live catalogue.

How does billing work?

Every request through the gateway is metered per request and per token, attributed to the credential that made it, and lands as a line item in pounds on the same monthly invoice as your apps. There are no subscriptions, seats or minimum commitments, and your £15 welcome credit applies to gateway usage too.

Can I run embeddings and audio transcription?

Yes. The catalogue includes text embedding models for search and RAG, and speech-to-text models priced per minute of audio, alongside chat, code, reasoning, vision and safety models. Embeddings and audio are currently partner-routed; our UK-hosted set does not yet include an embedding model.

Are there rate limits?

Yes, and we would rather tell you than have you discover them. Each credential has a per-client rate limit (a default of 100 requests per minute) enforced at the gateway. If you need more, talk to us and we will set a budget that fits your workload.

How do credentials work?

Access is issued through your tenant's single sign-on realm on our Keycloak identity platform - the same identity system as the rest of your workspace. Each person, service or agent gets its own credential, so usage in your bill is attributable per actor, and revoking a credential shuts off its access immediately.

Talk to us about UK-resident AI.

Drop us a line and an engineer will help you pick the right models, set rate limits that fit your workload, and get your team or your agents calling the gateway.

Name*

Company Name*

Phone Number*

Email Address*

Message*

Get more detail