AI Integration

AI is not an end in itself – it has to solve real problems. We integrate suitable models where they demonstrably add value: in document processing, with customer enquiries, in internal knowledge bases, in text creation.

You decide whether the AI runs locally on your server – full data sovereignty, no cloud – or whether we connect external providers such as Anthropic Claude, OpenAI or Mistral, cleanly encapsulated and replaceable at any time.

• Local AI models on your own server – full data sovereignty (e.g. with Ollama)
• Connection to Claude, OpenAI, Mistral and other LLM APIs
• RAG systems: your internal documents as a knowledge base
• AI agents for recurring tasks (research, classification, creation)
• Chatbots on your website or via Telegram/Slack
• Training and onboarding for your team
• Security, logging, auditing – no black-box installation

Data sovereignty: local or cloud – you decide

For sensitive or personal data we rely on local models that run directly on your server (Ollama, llama.cpp). Nothing leaves the building, full control, no ongoing cloud costs. Where it is uncritical, we deliberately connect the most powerful international models – Anthropic Claude, OpenAI, Mistral – cleanly encapsulated and replaceable at any time.

We advise you openly on which route makes sense for your case, instead of selling a standard solution. Often the combination is best: sensitive material stays local, the rest uses the strength of the major providers.

Off-the-shelf agent or a system of your own? Usually a system of your own

Agent frameworks such as OpenClaw or Hermes are quick to set up and useful for open-ended tasks: researching, sorting, drafting. But they come at a price. An agent thinks afresh on every run. For a process that runs identically every day, that means waiting time, ongoing model costs and a result that may turn out differently on the hundredth run than on the first.

So we build the system first and bring AI in only where understanding is genuinely required. A purpose-built program carries out the same step in milliseconds, identically every time, fully logged and with no cost per run. Our four-stage dunning process, the digital guard log with its automatic end-of-day report and the synchronisation between accounting and project management all work exactly like that: no agent, no model call, just code that does what it is supposed to.

Where an agent system is the right answer, we set it up, run it on your server and secure it. We recommend it only when the task genuinely calls for it. That is the difference between a solution and a trend.

AI, AGI and superintelligence: where we actually stand today

Three abbreviations come up more and more in conversations about AI. Narrow AI solves exactly one task. General AI, AGI, would handle any intellectual task at human level. Superintelligence, SI for short, would sit above that and surpass humans in almost every area. AGI and SI do not exist today. What does exist are language models and agents that complete several steps on their own.

For your decision today that means it is worth building your own technology so that it stays connectable: interchangeable model integration, clean interfaces, clear approvals. Then the next change of model is a configuration change and not a new project. How we place the term SI and what can actually be prepared is set out on our superintelligence page.

From chatbot to AI agent – from our own experience

We do not just talk about AI, we use it ourselves. The larger part of our own automation, however, runs without agents: purpose-built programs that do their job in milliseconds, identically every time and fully logged. Where a task is genuinely open-ended, an agent system is added. For that we run OpenClaw on our own servers, and we set up other frameworks such as Hermes just as readily; our internal knowledge base answers questions straight from our own documents.

These are exactly the building blocks we bring into your company: RAG systems that make your documents searchable, chatbots on your website or in messengers, agents for tedious routine. Including team training, logging and auditing – no black box that no one understands in the end.

AI is a tool, not an end in itself. We build it in exactly where it takes work off your hands – and nowhere it would only be expensive and complicated.