Self-hosted · EU data

Self-Hosted AI & LLM Hosting

Run your own AI on a Swedish server

Host open-weight language models yourself with Ollama and Open WebUI — on a cloud VPS or a GPU dedicated server in Sweden. Your prompts, documents and fine-tunes stay on your own machine in the EU, never sent to OpenAI, Anthropic or any US API. Full data control for AI that touches sensitive information.

See recommended plans Talk to us →

Open models like Llama, Mistral, Qwen, Gemma and DeepSeek now run well on your own hardware. With Ollama to serve the model and Open WebUI for a ChatGPT-style interface, you get a private AI assistant your whole team can use — without sending a single token to a third-party API. On an Internetport server, that inference happens in Sweden, under EU law.

Why self-host your AI

Prompts never leave the EU

Sensitive documents and customer data stay on your server in Sweden — not in a US AI provider's logs.

No per-token billing

Run as many prompts as you like for a flat monthly price. No surprise API bills, no rate limits.

Any open model

Pull Llama, Mistral, Qwen, Gemma, DeepSeek or Phi with one command. Swap models whenever you want.

Open WebUI included

A familiar chat interface with users, RAG over your own documents, and an OpenAI-compatible API for your apps.

What to run it on

AI inference is memory-hungry. Small quantised models run on a CPU VPS; larger models and fast responses want a GPU.

VPS · small models
Good for3B–7B quantised models, personal use
RAM8 GB
vCPU4
InferenceCPU (slower)
€ 18 / month excl. VAT
Order
Recommended
VPS · bigger models
Good for7B–13B quantised, small team + Open WebUI
RAM16 GB
vCPU6
InferenceCPU (slower)
€ 32 / month excl. VAT
Order
GPU dedicated
Good forFast responses, larger models, many users
GPUDedicated accelerator
InferenceGPU (fast)
SetupConfigured to order
Request a quote
GPU servers

Not sure which fits your model and user count? Tell us your use case and we'll size it.

Getting your private AI running

1. Order your server

A VPS for small models, or a GPU dedicated server for speed and larger models.

2. Install Ollama

One command installs the model runtime. Pull a model with ollama pull.

3. Add Open WebUI

Run the Open WebUI container for a chat interface, user accounts and document RAG.

4. Secure it

Put it behind TLS on your domain and restrict access. Your private AI is live.

Frequently asked questions

Yes, with limits. Small and medium models (roughly 3B–13B parameters) in quantised form run on a CPU VPS — fine for personal use, drafting and testing, though responses are slower than a GPU. For fast responses, larger models or many concurrent users, a GPU dedicated server is the right choice. We're happy to advise based on your model and user count.

Any open-weight model supported by Ollama — including Llama, Mistral, Qwen, Gemma, DeepSeek and Phi — in a range of sizes and quantisations. You pull and swap models with a single command, so you can match the model to your hardware and task.

For many organisations, yes. When you self-host, prompts and responses never leave your own server in Sweden — there is no data transfer to a US AI provider and no third-party processing agreement to rely on. That makes AI far easier to use with personal data, confidential documents and regulated information.

Yes. Both Ollama and Open WebUI expose an OpenAI-compatible API, so your own apps and scripts can call your self-hosted model with the same client libraries they'd use for OpenAI — just pointed at your server. Migrating an existing integration is usually a one-line change.

Your own AI, in Sweden

Self-hosted LLMs on a Swedish VPS from € 18/mo excl. VAT, or a GPU dedicated server for speed. Your data never leaves the EU.

See plans Talk to us