AI hub

The AI Hub is a central, trusted access point where open-source and commercial generative AI models can be connected in a sovereign, secure, transparent and future-proof manner. The AI Hub is currently under development and initially focuses on institutions affiliated with SURF. With the AI Hub, institutions can decide for themselves which language models their users are allowed to access.

Directing the use of language models with the AI hub

The AI landscape is changing rapidly. It offers opportunities to transform the research and education domain. Within the context of Dutch education and research, countless applications can be envisaged. Every week, new AI models appear that outperform previous ones. Legislation is being developed, such as the AI Regulation, which must be enforced, while specific AI policies are often still lacking.

Users of artificial intelligence are eager to get started, but also face many challenges. Does my institution allow this specific AI model, and is my data safe? How can I use multiple models? Can I use models specific to my field and share them with others? How do we, as an institution, avoid becoming dependent on generative AI providers, and how is cost allocation handled? With the AI Hub, SURF and Npuls aim to address these challenges.

Securing workflows

The AI Hub is a central, trusted access point where users or applications can be connected to various AI providers, such as GPT and LLM models. Users can then set up and configure this connection point themselves, depending on their needs and application. The AI Hub runs in SURF’s own data centres and, after certification that still needs to take place, can be used with confidential data.

The need for configuration varies by organisational group: lecturers and students, for example, have different requirements and constraints than a research department. With the AI Hub, institutions determine for themselves how AI is used at institutional or group level. The AI Hub enables institutions to create their own applications and user interfaces, or to extend existing applications or interfaces with AI functionality. In this way, the AI Hub provides the infrastructure that secures various AI workflows.

The AI Hub itself does not have a web application or front-end application such as EduGenAI. It is primarily intended to provide responsible access to language models and to ensure financial settlement. It can be used by programs via a so-called API connection. The OpenAI standard is used for this API connection. The software for the AI Hub has been developed by SURF, and the source code will be made available as open source.

Public values as a starting point

With the development of the AI Hub, SURF and Npuls join forces with institutions. Together, we are building a future-proof AI platform that provides responsible access to language models.

The AI Hub provides access to commercial and open-source AI models via a central point. In this way, it helps to:

Prevent vendor lock-in. By offering the best open-source models locally, the influence of suppliers (cloud providers) is reduced.
Use artificial intelligence with sensitive data (after certification).
Put institutions in control, by allowing them to decide which models are available to users and which are not, within their institutional or departmental AI policies. Institutions can also offer self-developed language models to partner institutions.
Ensure independence from suppliers. The AI Hub provides access to a variety of models from different vendors, making it easy to switch both language models and suppliers.
Increase transparency. Users are informed about the characteristics of models and providers, such as energy consumption, enabling well-considered choices.
Improve accessibility. Users gain access to state-of-the-art open-source models for which compliance checks have been carried out, running on local and trusted SURF infrastructure.
Increase transparency. The code of the AI Hub will be released as open source, allowing our members and other interested parties to deploy the AI Hub within their own environments.

From idea to service

The intention is to offer the AI Hub to our members as a service. For this reason, the AI Hub is part of the so-called LCPM process (Life Cycle Portfolio Management). During this process, a product is developed from idea to service.

At present, the AI Hub is in phase 0.7: the pilot phase, in which, alongside tested use cases with members, a pricing model is also being developed. Around 85 AI use cases or applications have been registered that wish to connect via the AI Hub. Participating institutions receive an API key after submitting their use case and signing a pilot agreement. This provides them with access to AI language models currently running on SURF’s hardware.

AI Hub and EduGenAI: similarities and differences

Within EduGenAI, the Npuls Growth Fund programme uses the AI Hub as its technical foundation. The powerful AI models and the more secure infrastructure of the AI Hub provide a basis for EduGenAI’s accessible, education-focused user interface.

The Npuls Growth Fund programme uses the AI Hub as the technical foundation within the EduGenAI platform. The powerful AI models and the more secure infrastructure of the AI Hub form the basis for EduGenAI’s accessible, education-oriented user interface. Like the AI Hub, EduGenAI is currently in a pilot phase, but the focus of the two pilots differs.

EduGenAI offers an ‘all-in-one’ solution: a ready-made user interface specifically designed for education. With the AI Hub, institutions can create their own or third-party applications and user interfaces, or extend existing applications or interfaces with AI functionality. This can also be done outside the education domain. In both EduGenAI and the AI Hub, SURF, Npuls and the institutions are working towards future-proof solutions for education in relation to AI.

For more information see:

More information

Want to know more about the AI hub developments? Then send an email to ai@surf.nl.