AI & Strategy
Agent-Native: Making Your Website Readable for AI Agents
ChatGPT, Claude and Perplexity read your site live. How to serve the same content as Markdown without forking the human version.
AI Integration
We integrate language models into your software and processes – with your data, under your control.
dectria integrates large language models (LLMs) into existing software and business processes: RAG systems that answer based on your own documents, automated document analysis and data extraction, and AI agents that access your systems through interfaces. We process data in the EU or fully on premises.
We run AI in production ourselves: NetCero, an ESG platform in which dectria holds a stake, uses Azure OpenAI with data processing in the EU to analyze documents and tag content automatically. Via MCP, Claude or your own agents can connect securely to NetCero.
Every project starts with a concrete use case and a prototype on real data. Only when quality and benefit are measurable do we take the solution to production – with evaluation, monitoring, clear data flows and a deletion concept.
Your contact Michael Jauk Contact for AI & data 
What is AI integration – and when is it worth it?
AI integration means embedding a language model in existing software so that it works with your company's data and processes, instead of standing next to them as an isolated chat tool. The most common building blocks are RAG (the model first searches your documents and answers on that basis), extraction (structured data from unstructured documents) and agents (the model calls functions of your systems through defined interfaces).
It is worth it when a lot of knowledge sits in text, recurring tasks require language understanding, and people can review the results. It is not worth it when a task can be solved reliably with fixed logic, when the necessary data is missing or inaccessible, or when every decision must be fully explainable. Then classic software is cheaper and easier to trace.
Cloud model in the EU or local model?
| Cloud model in the EU | Local model | |
|---|---|---|
| Data processing | in the provider's EU data center, e.g. Azure OpenAI with EU data zone | on your own hardware; data never leaves the network |
| Model quality | current large models | smaller open models, depending on the hardware |
| Operation | by the provider | your own maintenance, updates and monitoring |
| Cost model | usage-based | hardware and operation, independent of usage |
| Suited for | broad tasks, quick start | strictly confidential data, offline operation |
Related topics
Capabilities
What We Build with AI Integration
Azure OpenAI & Claude Integration Ollama On-Premise Deployments RAG Systems (Retrieval-Augmented Generation) Document Analysis & Extraction Vercel AI SDK & AI Gateway Prompt Engineering & Evaluation Embedding-Based Vector Search AI Agents & MCP Model Selection & Benchmarking Privacy-Compliant AI Architectures
Use Cases
Typical Use Cases
Document Analysis & Data Extraction
AI reads contracts, reports and forms and turns their content into structured data – as NetCero does when analyzing ESG documents.
Knowledge Assistants with RAG
Assistants that access your company data and answer with sources – for research, support and knowledge management.
AI Agents & Process Automation
Models that trigger actions in your systems through defined interfaces such as MCP – for example classification, routing and report generation.
FAQ
AI Integration FAQ
Which AI models does dectria use?
Azure OpenAI in the EU data zone or a European region, Anthropic Claude, and open models via Ollama for on-premises deployments. We connect models from different providers through the Vercel AI SDK and, where useful, the Vercel AI Gateway: it routes requests between providers, handles outages and does not store prompts by default. If data must stay strictly in the EU, we connect Azure OpenAI directly.
How is data protection ensured in AI projects?
We process data in the EU or on premises, sign a data processing agreement with the model provider, do not use your data for model training and define a deletion concept. Whether an Azure deployment stays in the EU depends on the deployment type – we choose it to match your requirements. It is best to align the legal assessment with your data protection advisor.
What is a RAG system and when do you need one?
RAG (retrieval-augmented generation) connects a language model with your own data: the model first searches your knowledge base for relevant passages and answers on that basis. This reduces made-up answers and makes answers verifiable through sources – useful for support, knowledge management and compliance questions.
How does an AI project work?
We start with a concrete use case and a prototype on real data. Then we measure quality against test cases with expected results. Only when quality and benefit are proven do we take the solution to production – with monitoring, clear data flows and a deletion concept.
How are AI results checked?
With test sets that contain expected answers, ongoing spot checks by subject-matter experts, and human approval for critical steps. That keeps it clear where the model is reliable and where it is not.
Every project starts with a conversation.