Skip to content
PUBLIKUMmedia
Local AI

Local AI that never leaves your house

Many companies would like to use AI and are not allowed to: contracts, personnel files, calculations and design data do not belong on the servers of a provider overseas. Local language models solve this. They run on your own hardware or on a server in Germany, and the data does not leave the machine it is processed on.

Contact: Juri Hoffmann, certified AI manager (MMAI®)Updated:

How you can measure us
  • 01Operation on own infrastructure, for example at IHK-Bildungszentrum Bonn/Rhein-Sieg
  • 02Certified AI manager (MMAI®) on the team
  • 03Data processing agreement under Art. 28 GDPR as standard
  • 04An honest assessment of where a local model is enough and where it is not
What we do about it

In detail

Models on your own network

Open language models on your hardware or on a dedicated server in a German data centre. No data leakage, no dependence on a single provider.

Knowledge from your documents

Questions to manuals, contracts and archives, answered with document and section as the source. Existing access rights stay in place.

Agents with a local core

The same AI agents as with a cloud model, just without the cloud: checking invoices, classifying enquiries, writing drafts.

Operation and evidence

Setup, updates and monitoring, plus documentation that lets you answer questions on data protection and the EU AI Act.

Operation

An AI server for your company, run by us

A local model needs more than a computer: choosing the model, setup, access rights, updates, monitoring. That is not a side project for an IT team that is already stretched. We take it on. You do not need to understand hardware. You decide which tasks the AI handles, and we make sure it runs.

01

Setup

One-off

We set up the AI server on your premises or in a German data centre and connect it to the systems it needs.

02

Operation

Ongoing

After that we maintain it: updates, servicing, monitoring and a contact person when something gets stuck.

03

Data protection

In the contract

We act as a processor under Art. 28 GDPR and provide the documentation you need to answer to data protection officers and supervisory authorities.

Where we are

What a local model can do and what it cannot

Open language models have caught up considerably. For summarising, classifying, extracting data from documents and writing drafts they are enough in most cases. For very long documents or intricate reasoning the large cloud models are still ahead. That is why we test on the actual task before anyone buys hardware. And we combine where it fits: local for everything confidential, cloud only for what is public anyway.

2020
founded
4
people
35+
websites live
★
NeuroWeb Award
From practice

Relevant cases

All work
Frequently asked questions

Local AI

What does local AI mean?

The language model does not run at a provider on the internet but on a machine you control: in your own premises or on your own server in Germany. Inputs and results do not leave that machine.

Is local AI GDPR-compliant?

It makes it considerably easier. If no data goes to third parties, transfers to third countries and the question of what a provider does with your inputs fall away. The other GDPR duties remain, such as purpose limitation and deletion periods. Binding legal advice comes from a law firm.

Are local models as good as ChatGPT?

For many business tasks yes, for some no. Summarising, classifying, extracting data and writing drafts work well locally. For very long texts or difficult reasoning the large cloud models are stronger. We test this on your task before you invest.

What hardware is needed?

That depends on the model and on how many people use it at the same time. For individual workflows a computer with a powerful graphics card is often enough, for a whole team rather a server. We decide after a test, not before.

Is it possible without my own hardware?

Yes. Then the model runs on a dedicated server in a German data centre that only you use. This is the usual starting point, because it shows whether your own hardware is worth it.

Where do you start?

With a task that involves confidential data and costs a lot of manual work today. We implement it as a pilot with a local model and measure what it delivers. Only then do expansion and hardware come up.

Does that sound like
your project?

A first conversation takes half an hour and costs nothing. If you would rather have the existing site checked first, the website audit is the smaller way in.