AI gateway pricing and architecture

We built api.node.uk to give UK developers a straightforward route to a UK-hosted AI model gateway. Global providers charge for compute in ways that obscure true cost at scale. Routing data overseas also introduces unnecessary compliance friction. Our gateway solves both problems through transparent pricing and a clearly mapped infrastructure layout.

How the gateway works

The gateway sits in our UK data centre. It routes your application requests to the appropriate model backend. You integrate api.node.uk into your stack using standard REST endpoints. We handle the queueing, load balancing, and key management. You send prompts and receive completions. The gateway tracks usage per application key. You can attribute costs directly to internal tools, customer portals, or automation workflows.

Chris Evans and Matt Collier lead the team that built and maintains the platform. We are a specialist firm, not a faceless vendor. We are happy to explain exactly how the routing logic works. We will also review your implementation for security.

Infrastructure and data residency

We host and operate the gateway infrastructure on our own hardware in the UK. The control plane, billing logs, and request routing remain under UK jurisdiction at all times. When you deploy an application, your keys and usage metrics never leave our estate.

Model execution depends on the specific model you select. We distinguish clearly between our own UK-hosted models and partner-routed options:

  • UK-hosted models: We run these on our own servers in the UK. Prompts, completions, and intermediate tensors stay within our facility. They carry full UK data residency guarantees.
  • Partner-routed models: Some larger or specialised models are served by upstream partners outside the UK. We route these through our gateway for consistent billing and key management. The inference work happens on their infrastructure. We label these explicitly in our model catalogue. You can make informed choices about data sovereignty.

We never imply the whole gateway is UK-resident. You will always see exactly where your data sits for each model call.

Metered AI API pricing UK

Predictable billing matters for production workloads. Global providers often charge by token count across shared organisational accounts. This makes it difficult to isolate the cost of a specific feature or department.

The gateway is prepaid and metered per token. You top up a balance, and each call draws it down at the published input and output rate for the model you called. There is no per-application monthly fee for gateway usage, no minimum spend and no contract. Live per-model prices are in the model catalogue, updated daily.

Optional auto-top-up refills your balance when it runs low, so a busy week does not interrupt service. Leave it switched off and the balance simply runs down.

What we do not do. There is no per-key monthly budget, and we do not queue or refuse individual requests to keep you under one. Your balance is the limit. If you want a hard ceiling on what a key can spend, ask us: we set request-rate limits per key and will agree ones that match your budget. We would rather set that with you up front than have you discover it during an incident.

Reasoning models generate thinking tokens before their answer. Those are billed at the output price even when they never appear in the reply, and a max_tokens that cuts the think block short can return HTTP 200 with an empty answer that is still charged. Give the think block room to close.

This structure gives you two advantages:

  1. Cost visibility: Usage is metered per model and per day and shown in your portal, so you can attribute spend to a tool or a team rather than estimate it.
  2. Compliance control: You can assign separate keys to GDPR-sensitive workflows and standard tasks. You keep regulated data isolated while paying a single invoice.

Open source AI inference hosting

We support a growing selection of open source models. This includes Llama, Mistral, and Qwen variants. Open source AI inference hosting means you are not locked into a proprietary runtime. You own your prompts, your fine-tuning data, and your deployment configuration. We provide the gateway as a stable entry point to these models. You retain the freedom to migrate your stack elsewhere if your needs change.

We update our model catalogue daily. Live figures and availability status are date-stamped. They render from our internal catalogue feed. If a model reaches end of life or undergoes a major architectural shift, we document the change alongside the deprecation notice.

Getting started

You need a transparent, UK-based entry point for your AI workloads. Review the full model list and live pricing on our gateway dashboard. Self-serve deployment is available for standard configurations. We offer managed setup for complex routing or custom compliance requirements.

Chris Evans and Matt Collier are directly available for technical questions. We do not hide behind ticketing queues or sales teams. Open a channel. Send your architecture questions. We will give you a straight answer about what the gateway can and cannot do for your stack.