EU AI Inference API
An OpenAI-compatible endpoint, hosted in Sweden
A managed inference endpoint running open-weight models — Llama, Qwen, Mistral, Gemma, DeepSeek — on our own GPUs in Sweden. Point your existing OpenAI client at our base URL, keep your code, and your prompts stay in the EU under EU law. It's your endpoint running open models, not a relay to a US provider — so there's no third-party account to suspend and no per-token surprise bill.
Teams building on AI want two things that rarely come together: the drop-in ergonomics of the OpenAI API, and control over where their data goes. This endpoint gives you both. We run open-weight models behind an OpenAI-compatible API on bare-metal GPUs in our Swedish data centre, so switching an existing integration is usually a one-line change — and every prompt, document and completion is processed inside the EU, never sent to a US AI provider.
Why a Swedish inference endpoint
Drop-in compatible
Keep your OpenAI client. Swap the base URL and API key — your calls, streaming and tool-use work unchanged.
EU data residency
Inference runs on our GPUs in Sweden, under EU law and GDPR. Prompts are not sent to, or stored by, a US provider.
Your endpoint, no bans
A managed service we run for you on open models — there's no third-party consumer account to get flagged or suspended.
Flat, predictable pricing
Dedicated capacity at a monthly price. No per-token metering, no bill shock when a job runs hot.
One line to switch
Already calling OpenAI? Point the same client at your Swedish endpoint.
# Same OpenAI SDK — only the base_url and key change
from openai import OpenAI
client = OpenAI(
base_url="https://ai.internetport.com/v1", # your EU endpoint
api_key="ipt-••••••••••••••••",
)
resp = client.chat.completions.create(
model="llama-3.3-70b",
messages=[{"role": "user", "content": "Hej!"}],
)
print(resp.choices[0].message.content)Works with the official OpenAI libraries, LangChain, LlamaIndex and anything that speaks the OpenAI API.
What powers it
Bare-metal GPUs
Served from our own GPU servers in Sweden with vLLM — real accelerators, not a reseller layer.
Open-weight models
Llama, Qwen, Mistral, Gemma and DeepSeek in several sizes. Ask for a specific model and we'll load it.
Chat, embeddings, tools
Chat completions, streaming, function/tool calling and embeddings — the endpoints your stack expects.
Shared or dedicated
Start on shared capacity, move to a dedicated GPU when you need guaranteed throughput and isolation.
Indicative tiers
Priced on capacity, not tokens. Tell us your model and throughput and we'll size it.
Starter | |
|---|---|
| Good for | Prototyping, low volume |
| Models | Up to ~8B, shared |
| Capacity | Shared GPU |
| Data | EU (Sweden) |
Team | |
|---|---|
| Good for | Production apps, steady traffic |
| Models | Up to ~70B |
| Capacity | Reserved GPU share |
| Data | EU (Sweden) |
Dedicated GPU | |
|---|---|
| Good for | High throughput, isolation, SLAs |
| Models | Your choice, reserved VRAM |
| Capacity | Whole GPU server |
| Data | EU (Sweden) |
Prices indicative — tell us your use case for a firm quote.
Frequently asked questions
Your AI API, in the EU
An OpenAI-compatible endpoint on open models, served from Sweden. Keep your code, keep your data in the EU.
Request access Prefer to self-host?