OnshoreAI is an OpenAI-compatible inference API from Hostcomm. Every prompt is processed, and every response generated, by open-weight models running in a UK data centre in Harlow, Essex — under UK jurisdiction, outside the reach of US law, and at a fraction of frontier API prices.
What is UK sovereign AI? UK sovereign AI is inference where prompts, models and responses are processed and stored in the United Kingdom, by UK-incorporated providers, under UK law. Since Brexit the UK is a separate data-protection jurisdiction from the EU, so "EU data residency" does not meet the requirement: it is a cross-border transfer under UK GDPR. OnshoreAI meets it by architecture.
Since Brexit the UK is a separate data-protection jurisdiction. An "EU data residency" option does not keep your data in the United Kingdom — it's a cross-border transfer. Here's how the positions actually compare for a UK buyer.
| OnshoreAI | OpenAI | Anthropic (Claude) | Google (Gemini) | |
|---|---|---|---|---|
| Company HQ & legal jurisdiction | ✓ United Kingdom | ✕ United States | ✕ United States | ✕ United States |
| Inference runs in the UK — not the EU | ✓ Always — Harlow, Essex | ✕ EU only — not the UK | ✕ US / global only | ~ Older models only; newest EU |
| Data stored at rest in the UK — not the EU | ✓ Always — UK only | ~ UK option (Enterprise) | ✕ EU regions only — not the UK | ~ UK option (via Vertex AI) |
| Beyond reach of US CLOUD Act & FISA 702 | ✓ Yes — UK jurisdiction | ✕ No | ✕ No | ✕ No |
| Outside US export controls and a US vendor's kill switch | ✓ Open weights on UK infrastructure | ✕ Subject to US export controls | ✕ Subject to US export controls | ✕ Subject to US export controls |
| Model retirements: 30 days' notice, headers and a named replacement | ✓ Guaranteed on every model | ~ Vendor deprecates on its schedule | ~ Vendor deprecates on its schedule | ~ Vendor deprecates on its schedule |
| Works without enterprise tier or hyperscaler routing | ✓ Standard on every plan | ✕ New projects / Enterprise | ✕ Needs Bedrock / Vertex | ✕ Needs Google Cloud |
| One UK DPA with an ICO-registered processor | ✓ One UK DPA, published sub-processors | ~ SCCs, US affiliates | ~ SCCs / cloud DPA | ~ Google Cloud DPA |
Based on each provider's publicly available documentation. Last reviewed 12 September 2026. UK ≠ EU: since Brexit the United Kingdom is a separate data-protection jurisdiction, so an "EU data residency" option does not keep data in the UK and constitutes a UK–EU cross-border transfer. OpenAI, Anthropic and Google offer UK or EU residency to varying degrees — typically on enterprise tiers, eligible endpoints, specific models, or via hyperscaler routing (AWS Bedrock, Google Vertex AI); availability varies by model and region. All three are US-incorporated and remain subject to US law, including export controls that can restrict access for non-US customers without notice — a decision that sits with a foreign government, not with the provider and not with you.
"UK-hosted" isn't a slogan here. It removes specific compliance work and legal exposure that a US provider can't — no matter which region they store your data in.
You still sign a standard UK DPA — but because data never leaves the UK, there are no SCCs, no UK IDTA, no Transfer Risk Assessment and no reliance on the UK–US Data Bridge. The whole international-transfer question simply doesn't arise.
OnshoreAI is UK-incorporated, so it can't be compelled under the US CLOUD Act or FISA Section 702 to hand over your data — the exposure that applies to US-headquartered providers wherever they choose to store it.
AI processing is usually "high-risk", so a DPIA is often required. UK-only processing means it needn't assess third-country transfers or foreign government access — a shorter review, and a quicker procurement decision.
One UK processor with a published, all-UK sub-processor list — instead of a US AI vendor plus its (often US) sub-processors, each needing its own assessment, DPA and security review. For security questionnaires, ISO 27001 audits and RFPs, "processed in Harlow, Essex by a UK provider" is a one-line, defensible answer.
You pay for UK compute, not a frontier brand premium. Match the model to the job — everyday reasoning, code, long documents, research, embeddings, transcription — instead of paying flagship prices for everything. Published prices, no enterprise tier.
Designed to meet UK data-residency requirements by architecture for NHS, central and local government, housing, financial and legal services — and ties into the G-Cloud / Crown Commercial route.
Every model below runs in the UK — language models in our Harlow data centre, transcription on our own servers in London. Prices are per million tokens (per minute for audio), ex VAT, on prepaid credit with no minimum. The same prices apply on every plan.
| Model | Model id | Best for | Context | Input / 1M | Output / 1M |
|---|---|---|---|---|---|
| Onshore Reasoning 262K | on-reasoning-262k | Everyday work — the best price-to-quality balance. Reach for it first. | 262K | £0.35 | £1.05 |
| Onshore Frontier 1M | on-frontier-1m | The hardest problems: deep analysis, long documents, agentic tasks. | 1M | £1.95 | £4.10 |
| Onshore Code 256K | on-code-256k | Code generation, review and refactoring. | 256K | £0.95 | £3.75 |
| Onshore Deep Research | on-research | Cited, long-form research answers. | 1M | £1.67 | £3.33 |
| Onshore Embed | on-embed | Vectors for semantic search and RAG. | 32K | £0.30 | — |
| Onshore Transcribe Pro | on-transcribe-dg | Speech-to-text. Runs on Hostcomm servers in London; audio never leaves the UK. | — | £0.01 per minute of audio | |
| Onshore Auto | on-auto · on-auto-cost · on-auto-quality | Let the router pick the cheapest capable model per request. | varies | Billed at the selected model's rate | |
Prices correct at 12 September 2026 and published live at app.onshoreai.hostcomm.co.uk. Monthly spending limits can be set per organisation and per API key.
Open-weight models running on UK infrastructure don't just tick the sovereignty box — they take your workloads outside the reach of a US vendor's roadmap and a foreign government's orders.
Your prompts and outputs are used to generate your response and nothing else. No model is trained on them — by us or by any sub-processor. This is a contractual term in the DPA, not a policy statement.
No model is withdrawn without at least 30 days' notice: Deprecation and Sunset headers on every API response, an email to every affected account, and a named replacement. You will never be silently rerouted to a different model at a different price.
Open weights and documented architectures can be inspected and written into your DPIA — no black-box third-country processor, and a clear answer when procurement asks exactly what runs where.
Why it matters. US-headquartered providers are subject to US export controls that can restrict access for non-US customers without notice. When that happens the decision sits with a foreign government — not with the provider, and not with you. Open-weight models running in the UK can't be pulled out from under you that way.
OnshoreAI uses an OpenAI-compatible API, so if you've called any modern LLM you already know how this works. Most teams are getting UK-hosted responses back the same afternoon they sign up.
Sign up in the browser — no procurement call needed to start building. A UK-based team is on hand if you'd rather be walked through it.
~2 minutesCreate a key in the dashboard and pick a model — a general-purpose one, or a specialist for code, research, embeddings or transcription. Set a monthly spending limit on the key if you want one.
~1 minuteBecause the API is OpenAI-compatible, you change the base URL and the key. Nothing else. Most SDKs and frameworks need no other changes.
~5 minutesYour first call is processed in Harlow, Essex and the response never leaves the UK. Pick the model you've validated and go.
Go livePython · openai SDK
from openai import OpenAI
client = OpenAI(
base_url="https://app.onshoreai.hostcomm.co.uk/v1",
api_key="ok_live_...",
)
reply = client.chat.completions.create(
model="on-reasoning-262k",
messages=[{"role": "user",
"content": "Summarise this clause in plain English."}],
)
print(reply.choices[0].message.content)
curl
curl https://app.onshoreai.hostcomm.co.uk/v1/chat/completions \
-H "Authorization: Bearer ok_live_..." \
-H "Content-Type: application/json" \
-d '{
"model": "on-reasoning-262k",
"messages": [
{"role": "user", "content": "Hello from the UK"}
]
}'
Using the Anthropic SDK? Point it at https://app.onshoreai.hostcomm.co.uk/v1/messages.
OnshoreAI is operated by Hostcomm — a UK contact-centre and telephony provider trading since 2004, already serving BT, HMRC, Scottish Power and 500+ UK organisations. Every credential below is held, not claimed, and independently verifiable.
Spin up a key and call a UK-hosted open-source model in minutes — or book a walkthrough of the architecture, the data-residency guarantees, and a live demo.
Page last reviewed 12 September 2026.
Covering data residency, UK GDPR, sub-processors, open-source vs proprietary models, model retirements, pricing, and how OnshoreAI keeps your AI workloads on UK soil.
Inference that stays in the UK, under UK law, by UK providers. Prompts, models and responses are processed and stored in the United Kingdom by UK-incorporated companies. Since Brexit the UK is a separate data-protection jurisdiction from the EU, so "EU data residency" is a cross-border transfer under UK GDPR, not sovereignty.
OnshoreAI is a UK-sovereign, OpenAI-compatible inference API operated by Hostcomm. It runs open-weight large language models in a UK data centre in Harlow, Essex, so every prompt and response stays under UK jurisdiction. It is built for UK organisations that need capable AI without sending data to the United States or the EU. Hostcomm has provided UK contact-centre and telephony services since 2004 and is PCI DSS Level 1 assessed and ICO-registered.
The United Kingdom specifically — not the EU. Language-model inference runs in a UK data centre in Harlow, Essex; transcription runs on Hostcomm's own servers in London. Since Brexit the UK is a separate data-protection jurisdiction, so the "EU data residency" offered by other providers is not the same as keeping data in the UK — it involves a UK–EU cross-border transfer. With OnshoreAI, prompts and responses are processed and stored on UK soil and never leave it.
Not on its own. Moving personal data from the UK to the EU is an international transfer under UK GDPR. It is permitted because the EU holds UK adequacy, but it still has to be documented, and it does nothing about exposure to the provider's home jurisdiction if that provider is US-headquartered. UK-only processing by a UK provider removes both the transfer and the exposure.
Yes — by architecture. Data is processed and stored only in the UK, Hostcomm is registered with the ICO as a data controller and processor, and a Data Processing Agreement (DPA) is available pre-contract. Because OnshoreAI and its inference sub-processor are UK-incorporated, your data is not exposed to US extraterritorial legislation such as the CLOUD Act or FISA Section 702.
All of them are published. Language-model inference runs on a UK-incorporated partner's data centre in Harlow, Essex. Transcription runs on Hostcomm's own servers in London under a Deepgram on-premise licence — audio is never sent to Deepgram. The API layer, database, authentication, payments and email providers are listed with their function and data location on the sub-processor list. Changes are notified 30 days ahead.
Cost, control, and independence. Open-weight models on UK infrastructure cost a fraction of flagship proprietary APIs, let you match a specialist model to each task, and are not subject to US export controls or a US vendor's decisions. No provider in the chain trains on your prompts. For most production workloads, an open model on UK soil is cheaper and lower-risk than a US frontier API.
You get notice, headers and a named replacement. No OnshoreAI model is retired without at least 30 days' warning: Deprecation and Sunset headers on every API response, an email to every account that has used the model, and a recommended replacement. Requests are never silently rerouted to a different model at a different price — something no US frontier API commits to.
No. Your inputs are used to generate your response and nothing else. Neither Hostcomm nor any sub-processor trains models on customer prompts or outputs. This is a contractual term in the DPA.
Published per model, from £0.35 per million input tokens. See the pricing table above. The everyday model, Onshore Reasoning 262K, is £0.35 in and £1.05 out per million tokens; the flagship Onshore Frontier 1M is £1.95 and £4.10. Billing is prepaid credit with no minimum commitment, and you can set monthly spending limits per organisation and per API key.
Yes. Chat completions, embeddings and models endpoints follow the OpenAI format, and there is an Anthropic-compatible messages endpoint. In most SDKs you change the base URL to https://app.onshoreai.hostcomm.co.uk/v1 and the key, and nothing else.
A range of open-weight models, matched to the task. Onshore Reasoning 262K for everyday work, Onshore Frontier 1M for the hardest problems, Onshore Code 256K for software, Onshore Deep Research for cited long-form research, Onshore Embed for vector search, and Onshore Transcribe Pro for speech-to-text. Every model carries a published lifecycle status.
OnshoreAI is operated by Hostcomm, a UK contact-centre and telephony provider trading since 2004 and already serving BT, HMRC, Scottish Power, HomeServe and 500+ UK organisations. Hostcomm is a PCI DSS Level 1 service provider (five years running), a Crown Commercial Service supplier listed on G-Cloud 14, registered with the ICO, and rated Excellent on Trustpilot. The sovereignty claims are backed by certifications held for years, not asserted on a trust page.
Sign up, get an API key, and call the endpoint — most teams are running in minutes. The API follows OpenAI conventions, so existing integrations need minimal changes, and a UK-based team can walk you through setup, model selection, and your data-residency requirements. Use "Start using OnshoreAI now" to create an account, or book a short call with the team.
Still have questions? Our UK-based team is happy to go deeper.
Talk to the team