An AI assistant that stays within your walls.

Why your own AI environment

The real estate and facilities department of a Dutch independent public body wanted to make generative AI available to its employees without becoming dependent on commercial cloud providers. Working documents, contracts and internal knowledge must not leave its own infrastructure. At the same time, the organisation saw teams around it using ChatGPT for work where that is not sufficiently secure. There was a need for a fully fledged AI assistant that works just as easily as ChatGPT but runs entirely on its own infrastructure. No data leaking out, no vendor lock-in and full control over which models are used and for what.

What we built

Together with SevenLab, Facilitech built an AI platform based on Open WebUI, an open-source alternative to the ChatGPT interface that you host entirely yourself. From a single environment, employees can switch between multiple open models. We run four side by side. OpenAI's gpt-oss-120b for heavy reasoning tasks, Alibaba's Qwen 3.5 for multimodal tasks and work in which the AI takes multiple steps on its own, Z.ai's GLM-5.2 for long documents thanks to its context window of 1 million tokens, and Google's Gemma 4 for fast question and answer. That's how we keep control over quality, speed and cost. A small model for light tasks, a larger model for heavy reasoning.

The assistant can load and query documents via RAG. Manuals, contracts and internal documentation serve as sources for the answer. Large collections go through retrieval and smaller documents go fully into context. On this foundation, we built variants for each role. One for project documentation, one for investment analyses and one for maintenance scenarios. Each variant has its own system prompt, its own knowledge base and its own settings. Using the Model Context Protocol, we connected the assistant to internal information systems so employees can query their own data directly. Think of a question that starts with internal project documentation and ends with a check in an information system. We deliberately disabled internet search so everything stays within the organisation's own environment.

What it delivers

Employees now work with an AI assistant that feels like ChatGPT but runs entirely within the organisation's own infrastructure. No more documents sent to external parties, and the organisation knows exactly where its data is. By running multiple open models side by side, the organisation stays flexible and keeps control over cost and performance. A small model where possible and a large model where necessary. The role-based variants let teams start right away with an AI tailored to their work. Facilitech supports the organisation in expanding knowledge bases, adding new models and connecting more internal systems. That way, the environment grows with what the work demands.