The use of AI at the University of Cologne
(Generative) artificial intelligence is integrated into all performance and operational areas at the University of Cologne. This includes, for example, AI research and the use of AI tools in research, teaching and learning (including skills development) and in a wide range of administrative and university management applications.
These different areas give rise to various use cases and corresponding (specialized) requirements. Various overarching offers, both current and planned, are described below.
Background – In-house and commercial offers, as well as different modes of provision
For various reasons, such as funding, compliance, security and data sovereignty, we are striving for a hybrid model that combines ‘in-house’ and commercial offers. In practical terms, this involves, for example, the planned provision of both open-source and commercial models.
We are also working toward different modes of provision, including: access to specific AI applications in the cloud, the provision of interfaces (APIs), e.g. for the use of our own software or in research, as well as the use of models on our own hardware, including local hardware.
Overview of available and planned AI offers
In order to meet the University of Cologne’s diverse and varied AI-application needs, we rely on a wide range of tools, with the aim of establishing a comprehensive AI offer that addresses as many use cases as possible. To this end, a pillar model has been developed to address as many needs as possible, based on target groups and through the efficient use of human, technical and financial resources. These pillars are described below.
Important: The use of AI at the University of Cologne is governed by the AI Guidelines. These guidelines were developed in collaboration with an external consultancy firm, an external law firm, the Legal Department and the Vice-Rectorate for Teaching and Studies, and has been ratified by the Rectorate and the staff councils. They are supplemented by the Guidelines FAQs and a whitelist for AI applications.
Pillar 1: LLMs on the high-performance computer / Open Source-KI.nrw
In collaboration with Ruhr University Bochum, as part of the Open Source-AI.nrw (OSKI.nrw) project and with funding from the Ministry of Culture and Science of the state of (MKW), various high-performance graphics cards from NVIDIA were procured for training and inference purposes and integrated into our new high-performance computer, ‘RAMSES’.
The aim is to provide widely accessible Large Language Models (LLMs) that have been enhanced with custom content (via Retrieval Augmented Generation), using a frontend developed in collaboration with other universities in North Rhine-Westphalia, thereby enabling the professional use of high-performance AI models on local hardware in research and teaching. To this end, the ITCC has explicitly tasked staff with developing and expanding the offer, utilizing the MKW’s financial resources and the technology available through RAMSES. A significant expansion involving resources, models and options is planned for 2026.
You can access the LLMs here: https://chat.kiconnect.nrw
Target group(s): Staff and students in research and teaching
Please note: In accordance with the DFG Guideline on the Use of Artificial Intelligence in the Review Process, the models made available via OSKI.nrw may be used, amongst other things, for DFG peer reviews.
Current status: available, with further expansion planned.
Pillar 2: Access to commercial models
Taking particular account of data protection and classification, the AI Guidelines also permit the use of certain commercial AI models. The framework agreements for the use of OCRE cloud services are already available, and the possibilities and options are currently being reviewed in collaboration with the participating companies. The cooperation agreement regarding KI:connect has been signed. Individual agreements must then be drawn up on the basis of these offers. The allocation of costs and relevant billing arrangements must also be clarified in this context.
Target group(s): Staff
Current status: Framework agreements and options under review
Expected availability: End of Q2/2026
Pillar 3: AI in the central university administration
As part of the hardware procurement outlined in Pillar 1, the ITCC also has the option of deploying specialized models for specific application scenarios within the university administration on dedicated hardware, and making them available via a separate frontend. The objective here is to achieve a level of security and data protection that allows us to work with internal documents as well.
Target group(s): Staff and functions within the central university administration
Current status: Prototype models have already been put into operation; further scenarios are possible from Q2/2026, and general integration into applications is planned for Q3/2026
Pillar 4: Beginner’s model for initial experiments
This pillar is designed to provide individuals interested in AI, as well as research groups and institutes, with a low-threshold entry point to developing AI-based applications and the opportunity to test these applications for themselves. The model will make use of existing resources and, as a result, will not be particularly high performing. Instead, it will focus on general availability and user-friendliness for beginners. To this end, technologies such as containerization and orchestration are used, with implementation taking place as part of cooperation projects across North Rhine-Westphalia and Germany.
Target group(s): AI developers, research groups
Current status: The model is currently being developed.
Expected availability: Q3/2026
Pillar 5: AI on personal devices
Thanks to the growing availability of smaller yet powerful AI models, it has become possible to run these models on personal (high-performance) devices, such as laptops. This approach ensures that no data leave the device, enabling an extremely high level of data protection, security and control.
For example, programmes such as LM Studio, Jan or GPT4All allow you to use models on your own device within a familiar interface. However, the performance of the models depends heavily on the size of the model and the available hardware, for example, whether the laptop has a graphics card. The ‘provision’ or service offered here is primarily limited to providing instructions or a brief guide on how to use AI models on your own devices.
Target group(s): Staff and students in research and teaching
Current status: A guide or handbook is currently being planned.