Framing
Mapping of your data, sensitivity level, regulatory constraints, use cases.
AIxH, a digital marketing and AI agency in Luxembourg, designs, hosts and runs your sovereign AI chatbot, in Luxembourg and across Europe: we frame the project, choose the architecture, install the model, connect your content and put the chatbot live. You have no AI engineer to recruit, no supplier to choose between and no data-location clause to clear with your legal team. Your documents. Your conversations. Your index. Everything stays within a European perimeter.
Your sovereign chatbot
Live in under two months.
Everything is included: design, European hosting, integration of your content and AI Act compliance.
Book a call with our teamSovereignty
This point deserves an explanation, because it often decides the fate of the project. A chatbot relies on two separate elements: the place where it is installed, and the language model it queries to phrase its answers. The two are not necessarily in the same place.
Your website and your chatbot can be hosted in Europe. But if the model they query is supplied by an American company, every prompt sent to that supplier contains extracts of your documents, the conversation history and so on.
The CLOUD Act is an American law adopted in 2018. It allows the United States authorities to require an American company to hand over the data it holds, including data stored outside the United States. The deciding factor is therefore the law the supplier answers to, whatever the location of the server.
This is why a legal department will block a project that was otherwise judged compliant.
AIxH designs the architecture around this constraint, so that the question no longer arises at validation stage. This requirement is part of our AI audit & integration service.
Scope
Mapping of your data, sensitivity level, regulatory constraints, use cases.
Selection and comparison of LLMs (Large Language Models), choice of hosting platform, sizing of computing capacity. You are not left to decide alone on a market that changes constantly.
Ingestion of your documents, chunking, vectorisation, search index, guardrails, system prompts, interface in your brand colours, launch on your website, your intranet or your applications.
AI Act transparency notice, record of processing activities, retention periods, access rights, documentation that will stand up to a regulator.
Re-indexing whenever your content changes, continuous evaluation of answers, tested version upgrades.
Benefits
Support
Access to information becomes immediate: procedure, contract, product sheet. The employee asks the question and gets a sourced answer, with the reference document attached, without having to go through several tools one after the other.
Customer relations
Recurring requests are absorbed by the chatbot, which lets your advisers spend their time on the cases where they add real value.
Sales
The chatbot qualifies contacts upstream: it asks the relevant questions, records the useful information and hands your team a prospect whose need is already defined.
Controlled answers
AIxH does more than connect an LLM to your website and hope it behaves.
RAG architecture
Before phrasing an answer, the system searches for the exact passage in your validated content, then writes only from that source.
Every answer cites the document it comes from and its version, so verification is immediate.
When a question falls outside the corpus, the system says it does not have the information.
Any update to a procedure is taken into account the same day, as soon as the content concerned is re-indexed.
Embedded in your website, your intranet or your applications.
The sources integrated by the AIxH chatbot
Compliance
Article 50 of the AI Act is respected: a transparency notice informs users that they are talking to an automated system.
The record of processing activities, retention periods and log purging are defined and documented.
Access rights management ensures that each user can only query the content open to them.
Conversations and the sources consulted are fully traceable.
The technical documentation is structured in a form that can be produced to a regulator.
Worth reading: Labelling AI-generated content becomes mandatory in Europe on 2 August →
Method
This first step defines the data perimeter, the regulatory constraints, the level of sovereignty required and the priority use cases.
A restricted corpus is assembled, the model is installed and the answers are tested on your real questions, so you can judge before committing to the next stage.
The full corpus is integrated, the interface and guardrails are deployed, the legal notices are published and your teams are trained.
Re-indexing, continuous evaluation, corpus enrichment and tested version upgrades are handled over time.
From €450 per month, all inclusive, and under two months between the first meeting and go-live. Design, European hosting, integration of your content, AI Act compliance and ongoing operation: everything is in the subscription. You validate the prototype before committing to the next stage.
Worth reading: AI chatbots must resolve 80% of customer requests without a human by 2029 →
FAQ
A chatbot relies on two separate elements: the place where it is installed and the language model it queries. If that model is supplied by an American company, the CLOUD Act, an American law adopted in 2018, allows the United States authorities to require that company to hand over the data it holds, including data stored outside the United States. The deciding factor is the law the supplier answers to, whatever the location of the server. AIxH designs the architecture around this constraint.
Everything is included: design, European hosting, integration of your content, AI Act compliance and ongoing operation (re-indexing whenever your content changes, continuous evaluation of answers, tested version upgrades). You validate the prototype before committing to the next stage.
Through a RAG architecture: before phrasing an answer, the system searches for the exact passage in your validated content, then writes only from that source. Every answer cites the document it comes from and its version. When a question falls outside the corpus, the system says it does not have the information.
Under two months between the first meeting and go-live. The project follows four steps: framing audit, prototype tested on your real questions, go-live with training for your teams, then ongoing operation.
Client reviews
★★★★★ 5.0 / 5 on Google (43 reviews) · See our Google profile →
Reviews were left on Google in French and are translated here.
“We entrusted AIxH with the design of a chatbot for our website as well as the management of our digital campaigns, and we were not disappointed. The project was carried out professionally and the work delivered was of excellent quality. A special mention for Grégory, with whom the collaboration was pleasant and efficient. We highly recommend them!”
“Beyond the professionalism and commitment of the team, it is a real pleasure to work with AIxH on any type of project. The team always adapts to find the most appropriate solutions to our requests. We are particularly grateful to have been able to build a long-term, trust-based collaboration with AIxH.”
Contact
Together we will carry out a framing audit. It answers three questions: which data must stay in-house, what level of sovereignty is required, and what operating volume the system needs.
Book a call with our team