AI Integration for Business
We connect company data to the Claude or OpenAI API — through search over your documents (RAG) or AI agents that carry out steps — with answer-quality controls and an EU AI Act transparency layer. Data is hosted in the EU (Frankfurt) and processed under GDPR; no third-party model is trained on it. The same connection, memory and protective layers run in our product Innea.
What's included?
AI integration and on-prem LLM start at 180,000 CZK (on-premise deployment from 350,000 CZK because of hardware) and include connecting the model to company data, answer-quality controls and the transparency layer. Always included: defining what the model may decide on its own and what it may not — that is design work, not an afterthought.
Where does this run in production?
In Innea, Czech AI support between therapy sessions, the Claude API runs with automatic failover to another provider, alongside three-layer conversation memory. Its crisis layer is a textbook example of a bounded AI role: a risk signal passes detection, an assessment by a language model and a content safety filter, and a human in a queue decides the follow-up. The EU AI Act transparency layer is part of the design — the user knows they are talking to an AI from the start of the conversation.
Frequently asked questions
Will our data stay in the EU under GDPR?
Yes. The application and the database are hosted in the EU (Frankfurt) and processing follows GDPR. For the model itself we choose the provider and region by where your data may go — and if it may not leave the company at all, there is the local LLM.
Is the model trained on our data?
No. We use the providers' APIs in business mode, where data is not used for training — not free consumer accounts, where that may not hold. We go through the specific provider's terms before the start and record which model runs in which mode.
What if the AI answers wrongly — who is liable?
Liability is defined in the contract; the design limits it so that a model's mistake has no binding consequences on its own: a human or a rule in the code decides, not the model. The AI gets a precisely bounded role, and everything with binding consequences — a quote, a payment, a decision about a person — goes through human approval or a hard check in the code; that is how the crisis layer in Innea works. Where that line runs for you is the first thing we draw on the call.
Do we have to tell users they are talking to an AI?
Yes — for a chat or an assistant the EU AI Act requires it (Article 50). We don't bolt it on as a notice: the transparency layer in Innea is part of the design, and the user knows they are talking to an AI from the start of the conversation. We use the same pattern for you.