Local AI

AI that keeps your data in-house.

Not every piece of information belongs in a cloud service. For customer data, contracts, personnel files or design plans we set up AI models that run on your own server or in an EU data centre of your choice. The data never leaves your control.

Start a project FAQ

When local AI makes sense

  • Sensitive dataHealth, personnel or financial data that should not go to a cloud provider.
  • Contractual requirementsCustomers or clients prohibit passing data on to third parties.
  • IndependenceNo dependence on price changes and terms of use of large providers.
  • Without internetThe AI should also work offline, for example in production.

For many tasks, cloud models from European or international providers are perfectly sufficient. We tell you honestly which option suits your data, and combine both where needed.

What we set up

Models for local operation

Openly available language, speech and image models that we tune to your tasks. They run entirely on your hardware.

Server and operation

Your own server on site or a rented server in an EU data centre, for example in Germany. We advise on hardware, set everything up and take care of updates.

Applications on top

Assistants, agents and searchable company knowledge that use the local model, with user management and logging.

Example: AEGIS, our agent system, runs entirely on premises, without cloud and without passing on data.

Documentation

A description of the processing, the technical and organisational measures and the models used, as a basis for data protection and the AI Act.

Models and providers

We are not tied to any provider. For local operation we use openly available models, combined with cloud services for tasks without sensitive data where needed.

Openly available models, run locally

  • Mistral
  • Meta (Llama)
  • Google (Gemma)
  • Alibaba (Qwen)
  • DeepSeek
  • Microsoft (Phi)
  • IBM (Granite)
  • OpenAI (Whisper, speech recognition)
  • Stability AI (images)
  • and more

Cloud services, combined where needed

  • Anthropic (Claude)
  • OpenAI (GPT)
  • Google (Gemini)
  • Mistral
  • ElevenLabs (speech and voice)
  • DeepL (translation)
  • Perplexity (research)
  • and more

To be honest: the limits

Today, local models are very good at texts, summaries and questions about your documents. For particularly demanding tasks, the largest cloud models are often still ahead. On top come costs for hardware or server rental. That is why we test beforehand with your real tasks whether the quality is sufficient.

How we proceed

  1. Initial conversation

    Free and without obligation. We clarify data, tasks and requirements.

  2. Test with your tasks

    We compare local models using real examples from your daily work.

  3. Setup

    Server, model and application, secured and documented.

  4. Operation

    Updates, model changes and support, so the solution stays up to date.

Costs

We set the exact scope in a clear quote after the initial conversation.

Local AI setup
from €1,500

Model, server setup and a first application, excluding hardware.

Hardware or server rental
as needed

Your own server or a rented server in the EU, shown separately.

Initial conversation
free

We tell you honestly whether local or cloud suits your data.

Matching products

A product that shows what local AI looks like in practice.

AEGIS

On-premises agent system: no cloud, planning, checking and execution kept separate.

Learn more →

FAQ on local AI

Is local AI automatically GDPR-compliant?
It makes it much easier, because no data goes to third parties. You still need to define purpose, legal basis, access rights and deletion periods. We provide the technical documentation for this; it does not replace legal advice.
What hardware do we need?
That depends on the model and the number of users. For a small team a single server with a powerful graphics card is often enough, alternatively a rented server in a data centre. We only make a recommendation after testing with your tasks.
Can we combine local and cloud AI?
Yes, and that is often the best solution: sensitive data stays with the local model, general tasks are handled by a cloud model. We define together which data may go where, also as a written AI policy if you like.
Who takes care of updates?
We do, if you wish. New models appear all the time; we test them and swap them in when they are better for your tasks.
What does local AI cost?
Setting up a local model with a first application starts at €1,500, excluding hardware. Your own server or server rental is added as needed and shown separately.

Further reading

From the blog (in German)

Terms explained

More AI services

AI solutions at a glance

Consulting, assistants, agents, company knowledge, infrastructure and training at a glance.

Learn more →

Company knowledge with AI

Answers from your own documents, with sources. Knowledge assistant from €1,000.

Learn more →

AI agents

Agents that handle multi-step tasks, with your approval. Individual agents from €2,000.

Learn more →