Frequently asked questions (FAQ)

What is SURF AI?

SURF AI consists of 2 components: the AI hub and eduGenAI. The AI hub links applications and various language models (LLMs) together. The eduGenAI chat application enables lecturers, students, staff and researchers to use language models on the AI hub. Contrary to what the name might suggest, eduGenAI can also be used for operational purposes or research. The AI hub and eduGenAI are interconnected and ensure that users can utilise (Gen)AI in the safest and most responsible way possible. See also: eduGenAI in 1 minute.

Who is SURF AI for, and how do I gain access?

SURF AI is available to all SURF members in vocational education (mbo), higher professional education (hbo) and university education (wo): for use in teaching, research and operational management. The institutions themselves decide how to deploy SURF AI within their organisation and who has access (students, lecturers, staff, researchers) to the eduGenAI chat application or their own application that utilises the AI hub directly. SURF will provide details of the registration procedure.

What does the SURF AI roadmap look like?

Until the end of December 2026, eduGenAI and the AI hub will remain in the pilot phase. From January 2027, SURF AI will be available as an (initial) sector-wide facility for all SURF members in vocational education (mbo), higher professional education (hbo) and university education (wo). It is not yet a fully-fledged ‘service’ like other facilities within SURF. SURF AI will be further developed in the coming period, collaboratively and for the benefit of all.

From January 2027, two versions of eduGenAI will be available. One features open-weight language models, accessible via the AI-hub. This is included in the sector service package. The second version includes, in addition to the models from the AI-hub, commercial models that can be accessed for an additional usage fee. With the first version, we can guarantee that data does not leave the platform.

From September 2027, every institution will have its own environment within eduGenAI. With their own environment, institutions can take control themselves, enabling eduGenAI to align more closely with their corporate identity, for example. They can also determine which of the available language models may be accessed and used within their institution, and to whom they grant access.

What makes SURF AI ‘safe and responsible’?

SURF AI is a secure and reliable environment for experimenting with generative AI. Multiple LLMs (language models) can be used in one place. By focusing on open source and co-creation, the education sector becomes less dependent on commercial suppliers. Data does not leave the SURF AI infrastructure (when using only open-source language models on the AI hub). This infrastructure runs in a top-tier Dutch data centre (tier 3, with 100 per cent energy from renewable sources) managed by SURF.

How much does SURF AI cost?

SURF AI is included in the sector services package (in Dutch), which comprises services that have been collectively selected within a particular sector to be purchased at a single rate. Each institution will be allocated a basic capacity, which we refer to as ‘capacity units’. Each institution can decide for itself how to use this allocation. Further information on the exact calculation and conditions of use will follow. If an institution chooses to use eduGenAI, including commercial models, these additional costs (‘tokens’) for the use of commercial models will be charged to the institution. 

How do you approach future aspirations and changes?

The AI sector is constantly evolving, and so is SURF AI. That is why we are developing a governance model: institutions can help determine the roadmap (what we do when, and what our priorities are) for further development. From 2027 onwards, this will be addressed in collaboration with SURF’s members. The aim is also to work together in a concrete way through co-creation across all areas, from the code itself right through to policy.

How does the chat feature work?

In eduGenAI, users can utilise various language models, all in one place. Would you like to see what that looks like? Instructional videos have been produced for the pilot environment. Watch the demo on the website of eduGenAI.

How is eduGenAI technically structured? 

eduGenAI is built on a mix of open-source components. The chat interface is based on LibreChat, which means we benefit directly from its extensive and growing functionality and the active community behind it. In addition to LibreChat, the platform uses a range of other open-source components that together form a scalable and secure ecosystem. 

How is determined which language models are available on the AI hub?

SURF has established criteria that language models must meet in order to be used on the AI hub. For example, consideration is given to the licence of the language models, whether they are General Purpose AI models, and the obligations under the AI Act. Ultimately, institutions themselves decide which of the models available on the AI hub they will activate for a particular application. The institution must therefore assess for itself whether the AI model meets the compliance requirements of the application and its institutional policy. 

European AI legislation has less stringent rules for applications in the field of research and development. It is expected that, from 1 January 2027, the AI Hub will also be able to offer more powerful AI models for research applications, such as strong open-weight models from China.

Is eduGenAI only for education?

No, eduGenAI is not specifically intended for education. It can also be used for operational or research purposes.

How did SURF AI come about?

eduGenAI began as a project within Npuls, bringing together knowledge and developments from educational institutions and SURF. The University of Amsterdam’s AI chatbot made it possible to quickly launch a nationwide pilot with eduGenAI. Since then, more than 60 institutions have collaborated on the development of eduGenAI. 

The AI hub originated at SURF as an idea to provide researchers with access to high-performance AI computing power in a responsible and scalable manner. It has since grown into a central access point for AI models. 

Both the AI hub and eduGenAI have been further developed and scaled up by SURF and Npuls in recent years.

Is eduGenAI the same as the UvA AI Chat?

No, eduGenAI and UvA AI Chat are not the same. During the pilot, ‘eduGenAI 1’ was based on the UvA AI Chat platform, and we worked closely with the UvA. The technical basis of the current eduGenAI is a further development of the pilot environment ‘eduGenAI 2’ and differs from that of UvA AI Chat. The UvA AI Chat and direct collaboration with the UvA made it possible to launch a nationwide pilot with eduGenAI quickly.