← All services
Service · Private AI

Private AI, on your own servers

Open models like Qwen and DeepSeek on servers you control: in your office, or on a dedicated server in an EU data centre. Your documents and prompts never go to an AI provider. We size it, set it up, connect it to your tools and keep it running.

Talk to us about private AI →See our own lab
Where it runs
Your office or an EU data centre
Models
Qwen · DeepSeek · Liquid AI
Sent to AI providers
Nothing
Maintenance
By us or your IT team

What you get

An AI system that belongs to you, not a subscription to someone else's.

  • A chat assistant for your team that answers from your own documents.
  • Search across contracts, plans, manuals and recordings, including scans and audio.
  • Help with writing, summarising, translating and code, without pasting anything into a public AI tool.
  • Connections to the tools you already use, such as your file storage or document archive.

Typical setups

We size the hardware to your team and use case.

In your office

GPU workstation

One machine with one to four GPUs on your premises. Good for a team that wants an assistant and document search, with the data never leaving the building.

In an EU data centre

Dedicated GPU server

A server rented only for you, for example at Hetzner in Finland. No hardware in your office, still no AI provider in between.

Scale

Small GPU cluster

Several GPU servers managed with Kubernetes, on-premise or in the data centre. For many users, larger models and automated workflows.

Prices depend on hardware and scope. We send you a rough estimate once we know your use case.

How it works

From the first call to a system your team uses every day.

  1. Sizinguse cases, users, data, budget
  2. Setuphardware, models, security
  3. Your datadocuments and tools connected
  4. Runupdates, monitoring, new models

Why us

We only recommend what we've run ourselves.

  • Our own AI workstation with four RTX 3090 GPUs, where we test new open models as they come out. See the lab →
  • Private document AI already in production on Vetra Cloud, with the model running inside the customer's cluster. See the project →
  • Years of keeping systems online: Ethereum validators since 2020 with 99.9% uptime, and a multi-tenant Kubernetes platform. See Vetra Cloud →
  • Qwen
  • DeepSeek
  • Liquid AI
  • llama.cpp
  • Docling
  • Kubernetes

Questions

What clients usually ask first.

Does our data go to an AI provider?

No. The models, your documents and every prompt stay on your own server, whether that's in your office or a dedicated server in an EU data centre. Nothing is sent to OpenAI, Google or any other AI provider.

Which models can it run?

Open-weight models such as Qwen, DeepSeek or Liquid AI's LFM family, chosen for your use case and your hardware. We switch or add models as better ones come out.

What does it cost?

That depends on the hardware and what you want to do with it. Tell us your use case and we'll send you a rough estimate for hardware, setup and maintenance.

Who looks after it once it's running?

We can, with updates, monitoring and model upgrades. Or we hand it over to your IT team with full documentation.

Does this help with data protection?

Because nothing is sent to an outside AI provider, there's no data transfer to document for that part. We help you describe the setup for your data protection officer.

Want your own private AI?

Talk to us →