Crystal — On-Premise AI
An AI assistant on your servers, not on someone else's cloud
Try it now in the portal
Crystal is a BoarDesign chatbot based on the local Gemma 4 model (4B parameters). It runs on our VPS server via Ollama: no calls to OpenAI, Anthropic, or Google. Conversations aren't stored on the server (they remain only in your browser), and attached documents are automatically deleted from the server within a few hours. It's installed on a server with standard CPUs, with no need for GPUs or other advanced hardware. This makes Crystal ideal for: conversational website front-end, FAQs, lead pre-qualification, request sorting, contact data collection, and high-level descriptions of company services. Within this scope, a 4B model (4 billion parameters) is perfectly competitive, runs quickly on the VPS, costs zero per call and maintains data sovereignty. Try it out in the Web Services portal.
Deployed at your premises
The same architecture can be replicated within your company: in-house server or dedicated VPS (Virtual Private Server), open-source models (Gemma, Llama, Mistral or other) chosen based on your requirements, data that never leaves the company network. NIS2/GDPR compliance, RAG knowledge base for your documents, custom integrations with your systems. All turnkey.