EU-sovereign AI infrastructure

Every model you need,
running inside the EU.

On-demand model APIs, private GPU hosting, and an intelligent gateway for your own network — one OpenAI-compatible interface, priced per token, governed under Dutch law.

0

Models online

0

Providers

<0ms

EU p50 latency

0.0%

Uptime SLA

OpenAIAnthropicGoogleMetaMistral AIDeepSeekQwenxAICohereAmazonNVIDIAMicrosoftOpenAIAnthropicGoogleMetaMistral AIDeepSeekQwenxAICohereAmazonNVIDIAMicrosoft

↳Anatomy of a request

01The request

Your application sends a prompt.

Any OpenAI-compatible client, pointed at one endpoint. No SDK to learn, no vendor-specific schema to adopt.

02The gateway

Amsterdam authenticates and routes it.

Keys, quotas, and policy are checked at the edge. The semantic cache answers repeat questions before they reach a GPU.

03The model

The right model answers — inside the EU.

Routing picks on quality, latency, and price. Weights run on hardware we operate, in data centres we can name.

04The receipt

Metered, logged, returned. Nothing crossed the border.

Every token is counted and priced. The trace, the prompt, and the response stay inside the same jurisdiction.

scroll
EU BOUNDARYYOUR APPOpenAI SDKHTTPS / JSONGATEWAYauth · route · cachepolicy · budgetAMSTERDAM, NLMODEL POOL — 49RECEIPTtokens 1,842 · €0.0031logged in eu-north-1

Platform

Catalog

Integration

Three steps, one afternoon.

01

Repoint the base URL

Any OpenAI SDK, LangChain chain, or vLLM client keeps working. One environment variable changes hands.

02

Pick your tenancy

Serverless tokens by default, or a dedicated private instance when residency or throughput demands it.

03

Watch cost, not invoices

Semantic caching, per-team budgets, and live tracing keep spend and latency flat as traffic grows.

Why CIPHER

Built for teams that answer to a regulator.

Data residency
Servers, backups, and logs stay in the Netherlands. Sub-processors are disclosed and contracted.
Published pricing
Every model carries a per-million-token rate. No negotiated tiers hiding the real cost.
No proprietary SDK
The wire protocol is OpenAI's. Leaving us costs one environment variable.
Operated, not resold
We run the inference stack ourselves — vLLM, SGLang, Triton — on hardware we control.

Customers

“Switching to CIPHER TECH cut our AI latency in half and gave our compliance team a single page to prove EU data residency. The per-token pricing is refreshingly transparent.”
CTO — European fintech scale-up
“We evaluated every major AI gateway on the market. CIPHER UX was the only one that let us route sensitive prompts to a local model while bursting to the cloud for everything else.”
Head of Engineering — Healthcare technology provider

Sectors we serve

Financial servicesLogisticsHealthcarePublic sectorLegalManufacturingEnergyRetailFinancial servicesLogisticsHealthcarePublic sectorLegalManufacturingEnergyRetail

Writing

Next

Bring one workload. We will tell you what it costs.

Send us a representative prompt and volume. You get a sizing estimate, a hosting recommendation, and an honest answer on whether we are the right fit.