Private AI that runs on your own hardware.

Your team already wants to use AI. The question is where your data goes.

When someone pastes a contract, a client’s accounts or a drawing into a cloud chatbot, that text goes to the provider’s servers. For a lot of firms that’s exactly the material that isn’t supposed to leave: client files under professional secrecy, patient information, unreleased designs, supplier contracts.

Running the model on your own hardware changes that. The questions, the documents and the answers stay on a machine you own, and after installation it can run fully offline, with no internet connection at all. There are no per-question fees to a model provider, so cost depends on the hardware and support you choose, not on how much your team asks.

One assistant that knows your documents, on a server that’s yours.

• A server you own. Sized to your number of users and documents before you buy anything. The exact hardware depends on how many people use it and which model you run.

• An open-weight model you choose. Models whose weights you can download and run yourself. In the lab demo an open-weight model of about 30 billion parameters ran on a single small desktop AI box (an NVIDIA DGX Spark). You aren’t locked to one provider, and you can switch models later.

• A document assistant. It searches your own files and answers from them, so the answers come from your material rather than the internet. Answers drawn from your files name the source document, section and page, and a click opens that document in the library. Document assistant

• A workspace your team will recognise. A chat app and a document library your team can use in Swedish or English, with projects to organise work and custom assistants for specific tasks.

• An admin console you control. Your admin decides which models run and sees system status at a glance. An audit log, kept on your own server, records actions such as sign-ins, document uploads, admin changes and the tools the assistants use. It doesn’t record the questions themselves: chats are stored on your server, where authorised admins can search them. A built-in check shows whether the system is network-isolated. The admin console is in English.

• Access set per person. Your admin decides who can open each document library. People who aren’t members don’t see it, and the assistant won’t answer them from it. Sign-in supports single sign-on over SAML and two-factor authentication.

• Rollout and ongoing care. Offered in tiers based on how many people will use it. We’ll go through what fits on a call.

Start with a pilot, not a purchase order.

1. 01

   15-minute call. What documents, how many people, what they need to ask. I’ll tell you honestly if local AI is the right fit or if something simpler would do.

2. 02

   Demo. You see the assistant answer questions from a set of made-up company files, the same demo I run in my lab.

3. 03

   Pilot on your documents. A limited set of your own files, on your hardware, for four weeks. You decide what goes in.

4. 04

   Decide. Keep it and size the real setup, or stop. You keep what we learned either way.

Built for firms that hold other people’s information.

• Law firms: contracts, precedents and internal guidance that fall under professional secrecy.

• Accounting and audit firms: client books, procedures and internal know-how.

• Manufacturers: drawings, specifications and supplier documents you’d rather not upload anywhere.

• Clinics and care providers: routines, guidelines and admin documents. Not for clinical decisions.

• Suppliers to the public sector: when your contracts limit where data may be processed.

What running AI locally solves, and what it doesn’t.

• It keeps the data in your building. No third-party AI provider sees your prompts or documents.

• It doesn’t make you GDPR-compliant on its own. You still need a lawful basis, retention rules and clear rules for what goes in. I’ll set up the technical side; your legal adviser owns the legal side.

• The EU AI Act depends on what you use it for, not where it runs. An internal document assistant is a different case from AI used in hiring or credit decisions.

• It can be wrong. Like any AI model, it can misread a document or answer confidently and badly. People should check answers that matter, and the setup should make that easy.

• It isn’t free. You pay for hardware, setup and optionally support. You stop paying per question.

One engineer, start to finish.

Austin Pontén is a Gothenburg-based security and infrastructure engineer who builds private AI systems on open-weight models, so companies can use AI without their data ever leaving their own hardware.

Pontén Solutions is just me, so the person on the first call is the one who sizes the hardware, sets it up and answers when something needs fixing. Before this, I built AI automations for clients, including a claims-intake agent that’s now in production. Recent work

Find out in 15 minutes if private AI fits your firm.

Tell me what your team needs to ask its documents. I’ll tell you what it would take, which hardware class it needs, and whether a pilot makes sense. I read every message myself.