ARSoftware builds custom AI software for enterprises: AI agents, LLM integrations and internal AI tools – GDPR-compliant by design, with EU hosting options and transparent data flows. Engineered in Bavaria, Germany.
ARSoftware builds three kinds of AI systems – on their own or combined, depending on where the effort sits.
Agents take over work that is done by hand today: gathering information, processing documents, writing results back into your systems. We define which decisions an agent makes on its own and which need a human sign-off. Every step is logged so it stays clear what happened.
We put language models where they earn their place: inside existing applications, internal portals or interfaces. That includes context design, connecting your own data sources, and tests that check whether the output stays usable. Model access is encapsulated, so switching provider does not mean rebuilding.
Tools for your team: search across your own documents, drafts for recurring text, analyses that would otherwise live in spreadsheets. We start with one clearly scoped use case a team uses daily, rather than a platform for everything. What proves itself, we extend.
In AI projects, data protection decides early what is technically an option at all. So we settle these points during design, not at acceptance.
For each use case we establish what data a model actually needs to see. Personal fields are removed, pseudonymised, or never transmitted in the first place.
Models can run through EU data centres, and open-weight models inside your own infrastructure. Which option fits depends on your data classification and on how good the output has to be.
The data processing agreement, sub-processors and technical and organisational measures are agreed before the project starts – including which providers are in the processing chain at all.
You get documentation of which data goes to which model, where it is processed and how long it is retained. That makes maintaining your record of processing activities easier.
You remain the controller under the GDPR. We provide the technical implementation and the documentation for it – this does not replace legal advice.
Every AI project at ARSoftware runs in four steps: discovery, design, build, operate. After each one you decide whether to continue.
We look at the process you want to improve: who works on it, what data exists, where the effort accumulates. The result is an assessment of whether AI holds up here, including when the answer is no.
Architecture, data flows, model choice and hosting option are settled and documented before the first line of code. You know up front which data goes where and how we will measure the result.
We work in short feedback loops and show running increments early. Testing covers not only the functionality but whether the model output is good enough for the use case.
You get the code, the documentation and the access. On request we stay on for operations, monitoring and further development – for instance when new model versions arrive.
We settle the data protection questions during design, before any code is written: what data the model needs, which fields are removed or pseudonymised, where the processing happens and how long the data is retained. On top of that come EU hosting options, a data processing agreement with technical and organisational measures, and documentation of the data and model flows. You remain the controller under the GDPR – we provide the technical implementation and the evidence for it, not legal advice.
Yes. AI models can run entirely in the EU – either with providers that operate EU data centres or as open-weight models inside your own infrastructure. The trade-off is output quality against operating effort: large commercial models are often more capable today, while open-weight models give you full control over the processing. Which option is sufficient for your use case is decided during design and fixed before development starts.
We build on large language models (LLMs) – either commercial model APIs or open-weight models that can be hosted in the EU. The applications themselves are built in the TypeScript and Node ecosystem, and in Swift for Apple platforms. Model choice is made per project, and model access is encapsulated so a later switch stays possible.
An AI software project at ARSoftware runs in four steps: discovery, design, build, operate. Discovery checks whether AI is the right tool for the process at all; design settles architecture, data flows, model choice and data protection before development begins. We then build in short feedback loops and hand over code and documentation, with ongoing support on request. After each step you decide whether to continue.
At our size that is a fair question, and we do not dodge it. One commitment holds in every project: you receive the full source code, the documentation and the access credentials, so another team can take over without starting from scratch. We build on widely used technologies and encapsulate model access, so nothing stays tied to ARSoftware as a supplier. Everything beyond that – availability, response times in operation, and what applies if we are unavailable for a longer period – depends on how critical the system is for you, and is therefore agreed contractually per project. We raise the question ourselves before operations begin.
The cost depends on scope, which is why we do not publish flat rates. The main cost drivers are how tightly the use case is scoped, the number and nature of the systems to be integrated, the data protection and hosting requirements, and ongoing operation including model usage costs. After discovery you get an effort estimate for a clearly scoped first step, before committing to anything larger.
Tell us briefly what you are working on. You get an honest assessment of whether AI is the right fit, and what a sensible first step looks like.
Or email us directly: contact@arsoftware.tech
