Institutional Access to AI Services
Description
In this section, you will find explanations on how to access the AI services in three main steps:
- selecting the subject matter for the contract
- selecting the deployment
- limiting access
Note that you can always discuss with us customization and alternative options.
Contract Selection
- All AI Services are integrated to Academic Cloud. If you institution is not yet using it, you need an Academic Cloud Basis contract.
- Access to the Open-Weight-Models hosted by GWDG in Göttingen using the Academic Cloud service Chat AI.
- Orthogonal to this offer are Chat AI external models which are hosted by third parties.
- Based on your choice, we have a different service descriptions and requirements that must be met.
- Please write a request to verkauf@gwdg.de or support@gwdg.de and include the following information:
- Full address and contact information
- List of Academic Cloud services (e.g. ‘Basis’, ‘Chat AI’, ‘Chat AI externe Modelle’)
- The Size of your institutions (e.g. 27.000 students, 5.000 employees) and the number of intended users for each service option
- Any additions that we should consider in our offer
Contracts for Open Weight models hosted by GWDG
The service option Chat AI includes the AI services Chat AI, Arcana/RAG, CoCo AI, Image AI, Voice AI and Protein AI.
Contracts for optional access to external models
The service option Chat AI external Models gives access to models hosted by third parties. You get access to these models through our Chat AI webinterface in Academic Cloud if you also order the Chat AI service. There is also an option to use this service only via API (see below). For an overview of models please see Available Models.
Entry point from DFN
If you wish to access our AI services via our partnership with the DFN, you must also sign the corresponding contracts between you, the GWDG and the DFN. Please contact verkauf@gwdg.de.
Entry point from KI:Connect
The default approach for KI:Connect is to use open weight models via API endpoint. Legally there is NO need to make an AVV with us but it useful in case someone accidentally puts in GDPR personal data. Institutions that use access to external models via API as part of our partnership with KI:Connect require an additional API key. Please contact kisski-support@gwdg.de for this.
Deployment
Generally, it is your choice how to deploy our services:
- Deployment 1) Using the existing Web interface for end users (such as https://chat-ai.academiccloud.de/)
- Deployment 2) Integrating AI APIs into services you deploy locally (such as your own chat interface)
Deployment 1: Using existing web frontends
To use the AI services, it is generally possible to use our web interface. An Academic ID is required.
Deployment 2: Integrating AI APIs into YOUR services
With an Academic ID, you can also request an API key via the KISSKI website (click on “Book” on the page). You can use the API key in another webfrontend, plugin or your own code.
Limiting Access
Restricting Models
We provide a large assortment of Large Language Models (see Available Models. Available models are regularly upgraded as newer, more capable ones are released. Not all of these may be approved for use by your institution. In order to restrict the use to certain models, the authorized person of your institution must inform us in writing of the selection of models.
Restricting Model Access for specific groups
You may specify different user-groups within your institution by assiging functional affiliations (e.g. employee@university.de, student@university, researcher@university) and instructing us which models are available for each user group. In order to gain access to the IdM objects (accounts, groups, shared mailboxes and resources) of your institute or institution in the GWDG identity management system, it is necessary for us to set up special authorizations for the IdM portal for you. Access to the special authorizations forms can be found here.
Limiting external model access by quotas
With your order confirmation you can impose a finanical ceiling for the usage of the external models. As a result, total consumption can be reduced to the corresponding monetary value of this limit.