YoKI

YoKI is Heidelberg University's open-source AI platform—like “ChatGPT hosted on the servers of Heidelberg University”—that provides on‑premises access to multiple large language models (LLMs) operated exclusively on university servers to ensure data sovereignty and privacy. Accessible from the university network or via VPN, content never leaves campus and prompts are not retained or used for training. It is developed jointly by the Engineering Mathematics and Computing Lab (EMCL) and the University Computing Centre (URZ).

Under the hood, YoKI runs a curated set of open‑source models—Llama and DeepSeek for general text generation, Aya for multilingual communication, and Qwen with a focus on coding—so users can select the model best suited to their task. DeepSeek variants leverage distillation and pruning to reduce compute and energy while preserving quality; all models are operated on‑prem for compliance and customization.

Looking ahead, YoKI will keep pace with modern LLMs and expand capabilities: institution‑specific chatbots powered by Retrieval‑Augmented Generation (RAG), an improved UI with better chat features (e.g., history), and multimodal options to meet academic, administrative, and compliance needs.

• Project steering and roadmap planning to keep YoKI on track and define next steps
• General software architecture and technical direction across services and integrations
• Ensuring YoKI satisfies legal and compliance requirements (e.g., data protection, EU AI Act)
• User outreach to gather, synthesize, and prioritize requirements across faculties and units