AI Solutions
AI solutions that run where they pay off most.
In the cloud, in a private environment or on your own servers: we build artificial intelligence around your data and your processes. From the first analysis to day-to-day operations.
Cloud or your own servers? It depends: on you.
ChatGPT in a browser tab is a start, but it isn’t yet a tool for your business. A real tool knows your data, talks to your systems and takes visible work off people’s hands. That is what we build: AI solutions in the cloud when you want to start fast and stay flexible, or on your own infrastructure when your data should never leave the building. Both are good routes. The only question is which one fits you. We work across Germany and Europe: we rack the hardware on your site, everything after that runs from anywhere. Whether your offices are in Hamburg, Munich, Vienna or Amsterdam makes no difference to the result.
On-premise AI: your models, your servers, your data.
If your data needs to stay in-house, by choice or by regulation, that is exactly what we build: AI systems on your premises or on dedicated servers in Germany and the EU. Open language models tuned to your hardware, with a clean interface for your team. You keep full data sovereignty, and the cost stays predictable instead of growing with every seat.
On-premise AI means the language model runs on servers the company owns or that are operated exclusively for it. Prompts, documents and answers never leave its own infrastructure. Open models that can be run locally make this possible. With a cloud service, the processing happens at the provider instead.
- Selection & operation of open AI models
- Full setup: hardware, software, user interface
- Connected to your documents and systems
- Maintenance, updates and support in production
Assistants that know your business.
A chatbot that has read your manuals, contracts, product data and processes and gives your team or your customers the right answer in seconds. We connect AI models to your company knowledge, with proper permissions and verifiable source citations.
- Internal knowledge assistants for your team
- Support bots with real answers, not canned phrases
- Source citations & permissions built in
Processes that take care of themselves.
Sorting and answering emails, reading documents, moving data between systems, writing reports: AI agents handle the routine while your team makes the decisions. We look for the processes that eat the most hours and automate them step by step.
- Document & email processing
- Automated data maintenance across ERP, CRM and shop
- Workflows with approval steps: you stay in control
Your systems. Just smarter.
You don’t need to buy anything new to use AI. We integrate AI features straight into your existing landscape: your online shop, ERP, CRM or your own application. Via APIs to leading models, or with local models when your data demands it.
- Product copy & translations in your shop
- Intelligent search and recommendations
- AI features inside your own software
Three situations we know well.
AI stays abstract until it takes real work off someone’s desk. So here are three starting points from companies we have built for. One of them may sound familiar.
Wholesale
40,000 items in the catalogue, and your sales desk answers the same twelve questions all over again every day.
An assistant that has read your item master data, price lists and supplier documents answers in seconds: what the alternative part is, when it ships, and at what quantity the tiered price kicks in. Every answer cites its source, so your people can back it up. The time that frees up goes into the cases where a human really matters.
Building supply & industry
The installation guideline sits as a PDF on a network drive. The person who needs it is standing on a construction site.
We connect technical documents, data sheets and approvals to an assistant your field team can query from a phone. Answers point back to the document and page rather than inventing something plausible. And because many manufacturers cannot let design or test data leave the building, the same system will happily run on a server inside your own network.
Professional services
Every tender reads much like the last one and still costs two days of preparation each time.
An AI agent reads the documents, pulls out deadlines, required certificates and knock-out criteria, and hands you a summary with the open questions listed. You still decide whether to bid, just on the first morning instead of the third. Where a wrong call would be expensive, we put an approval step in front of it.
Cloud, private or in-house: how to tell which one fits.
This question opens almost every AI project. There is no universal answer, but there are testable signals. Three routes we take when building AI solutions in practice: equal in standing, different in strengths.
Models from the cloud
You reach leading models through an API. The fastest route to a working result, and every new model generation is available to you the day it ships.
- You want to start in weeks rather than quarters
- The data is uncritical, or can be anonymised before processing
- Load varies a lot and you would rather pay for what you use
Private EU environment
An open model runs on a dedicated machine in a German or European data centre. Yours alone, with a data processing agreement in place.
- The data must not leave the EU, but owning hardware would be overkill
- You need a predictable monthly cost instead of per-request billing
- Compliance asks you to name the place where processing happens
On your own premises
The model runs on your hardware inside your network. Prompts, documents and answers never leave the building.
- You work with design data, HR records or client files
- Many people will use it, and past a certain headcount owning the hardware pays off
- A certification or a major customer dictates where your data may live
In practice most companies mix: the sandbox runs in the cloud, the system holding contract data stays in-house. We cost out both routes before you commit, including what switching later would take.
Straight answer
When an AI project with us makes no sense.
- When it isn’t yet clear which task is supposed to get better. “Something with AI” does not turn out well, however healthy the budget.
- When the data the AI is meant to answer from is scattered, stale or contradictory. Then tidying up is the first job, not the model.
- When you need a promise that a wrong answer will never appear. Nobody can give you that. Good systems make mistakes verifiable; they do not rule them out.
What we are genuinely good at: picking one task where the benefit is measurable, getting it running on your real data, and then telling you honestly whether scaling it up is worth it.
From first conversation to running system.
- 1
Analysis
We look at your processes and your data, then tell you honestly where AI pays off and where it doesn’t.
- 2
Proof of concept
A small working prototype with your real data. Results in weeks, not months.
- 3
Delivery
The prototype becomes a system: integrated, secured, documented.
- 4
Operations
We monitor, maintain and keep improving, with guaranteed response times.
What companies ask us before they start.
Is this GDPR-compliant?
Yes. With on-premise systems your data never leaves the building. For cloud setups we use servers in Germany and the EU with clear data processing agreements.
Do we need our own hardware?
Not necessarily. We assess whether dedicated hardware pays off or a private cloud VM is the better choice, and deliver either one ready to run.
What does an AI project cost?
A proof of concept is usually a manageable fixed-price project. After that you know what a full rollout delivers and what it costs, long before you invest at scale.
How long does it take?
That depends less on the technology than on your data: if the documents are digital and reasonably organised, it moves quickly. If we have to open them up first, it takes longer. After the first conversation we tell you which order of magnitude to plan for.
Which models do you use?
The ones that fit the task: leading commercial models via API, or open models on your infrastructure. We are not tied to any vendor.
What is RAG, and do we need it?
RAG stands for retrieval-augmented generation. Before answering, the language model is handed the relevant passages from your own documents and answers from those, with a citation. You need it whenever the AI should know your company rather than just general knowledge. Fine-tuning a model for that purpose is, in most cases, neither necessary nor economical.
Will our data be used for training?
No. We do not train models on your data, and the cloud APIs we use contractually exclude customer data from training. With an in-house setup the question does not arise at all: the data stays inside your network.
Does this pay off for a company with 30 staff?
Often yes, especially where a small team has to cover a lot of ground. Headcount is not the deciding factor; a task that recurs often enough is. Ten enquiries a day at ten minutes each is a strong candidate.
What happens when the AI gets something wrong?
It can happen, and anyone who tells you otherwise has not shipped one yet. That is why our assistants cite their sources: every answer points to the passage it came from, and your team can verify it in two clicks. Where the outcome has consequences, an approval step sits in front of it. A person still decides.
The four paths in detail
- AI phone assistantAnswers calls when nobody is free and hands the request over so someone can pick up the work.
- Process automationRecurring steps between ERP, shop and inbox, built along the systems you already run.
- Company knowledgeAn assistant that searches your documents and names the source behind every answer.
- Corporate LLMYour own language model in house or in a private cloud, when your data calls for it.
Related topics
- Cloud & ServersThe infrastructure a private or in-house AI runs on: set up, monitored, with a committed response time.
- AI in e-commerceProduct copy, search and recommendations right where your revenue happens.
- Custom SoftwareFor when the AI feature belongs inside your own application rather than in a tool beside it.
Where would AI move the needle for you? Let’s find out.
30 minutes, free and without obligation: we look at your process and tell you honestly whether AI will help there. If it will not, we say that too.
- Free and without obligation
- 30 minutes, by video call or phone
- An honest assessment, not a sales pitch