WordPress + Private AI: Chatbot plugin for the Safe Swiss Cloud website

Author Eberhard Lauth on 21 April 2026 Eberhard Lauth's blog

An AI chatbot on a company website can significantly improve the user experience. It can guide visitors, answer frequently asked questions, and is available around the clock.

The Safe Swiss Cloud website runs on WordPress. The official WordPress repository offers many plugins that make it straightforward to integrate chatbots. However, these are all built for the major providers such as OpenAI (ChatGPT), Anthropic (Claude), or Google (Gemini). Their APIs can be integrated with these plugins in just a few clicks — but this raises an uncomfortable question: where does the chat data actually go?

For Safe Swiss Cloud, this is not an abstract data protection question but a strategic one: 100% data sovereignty is part of the company’s DNA. An AI chatbot on one’s own website that answers visitor enquiries using US-based AI infrastructure is therefore not a viable solution.

The Issues with “Simple” WordPress Chatbot Solutions

From a data protection perspective, most ready-made chatbot solutions are a black box. Your visitors’ enquiries — questions about pricing, products, and potentially personal matters — are forwarded to servers in the United States, processed there, and possibly used to train future models. For a company serving Swiss and European customers, and subject to the corresponding obligations under the Swiss DSG and the GDPR, this is untenable.

The Solution: Private AI as the Backbone

Safe Swiss Cloud operates its own AI service on Swiss infrastructure with Private AI. Models run on servers in Switzerland, under Swiss data protection law, with no disclosure to third parties. What was missing was a WordPress plugin that cleanly integrates the Private AI API into the existing website — and not merely making it work in some fashion, but enabling it to draw on Safe Swiss Cloud’s own knowledge for its conversations.

What the Chatbot Can Do

The plugin developed for this purpose therefore does more than answer enquiries generically. The chatbot understands the company’s context:

It knows the website’s content. The plugin automatically reads in all pages, posts, and FAQs (as well as any additional custom post types that may be added in future) and builds a searchable knowledge base from them. When a visitor asks a question, the chatbot first searches for relevant content on the website — and passes this to the AI as context. The result is answers that genuinely reflect the company’s reality, not generic AI phrases.

It knows more than just the public content. In addition to the automatically indexed website content, a private knowledge base can be maintained in the WordPress backend. This allows entries to be added that have never appeared on the website — internal FAQs, product details, pricing models, support answers to frequently asked questions. The chatbot draws on this content as context. This provides additional editorial control over what the chatbot knows, without that information needing to be publicly accessible.

It understands multiple languages. The website is multilingual, and so are its visitors. The chatbot automatically detects the visitor’s language and prioritises matching content — answering a German question in German, an English question in English, drawing on the linguistically appropriate page content in each case.

It stays with users throughout their visit. When a visitor navigates from one page to the next, the chat history is retained. There is no need to start over each time.

It fits the design. The colour, position, and label of the chat button can be configured in the WordPress backend. The widget texts — greeting, title, disclaimer — are translatable via common WordPress multilingual plugins (WPML, Polylang).

Why This Approach

Three considerations underpin this architecture:

Data sovereignty is not optional — it is part of the product. For Safe Swiss Cloud, a chatbot that processes data outside Switzerland would be a contradiction of its own business model. This is resolved by using the Private AI API for both chat and embedding.

The chatbot should know the company. An AI that performs well in general but knows nothing about the specific company is of limited use for visitor enquiries. Integrating the website content as context with an additional knowledge layer for RAG is the difference between a chatbot that genuinely helps visitors and one that fobs them off with generic answers.

Choose the right LLM model. The plugin is built in such a way that the LLM can be swapped out at any time without touching the website.

The Result

Visitors to the website can now ask questions in natural language and receive answers based on the actual content of the website — in their own language, in a widget that matches the design. All enquiries are processed via Private AI on Swiss servers. No data transfer to the United States, no disclosure to third parties, no compromises on data protection. The chatbot has been live on the website since March 2026 and is continuously being developed further.