Additional information for specific target groups
RAI is particularly helpful with text-based tasks, such as writing, reorganizing, summarizing, translating, or developing ideas. RAI is accessible only on the university network or via VPN.
Depending on the model, RAI is operated either by GWDG in Göttingen or via Azure in another EU country. User input is not used to train the AI. RAI may also be required in courses, where appropriate. There is also a group chat feature available.
Sample prompts to get you started
RAI can assist you in your daily work – whether as a voice assistant, a translation tool, or for quickly transcribing long documents. The following sample prompts illustrate specific use cases for different target audiences. Important: Always critically review AI-generated results to ensure their accuracy. AI models can generate content that sounds plausible but is actually incorrect (so-called “hallucinations”). Before entering any data, be sure to check its confidentiality rating (C1–C4).
“Please edit the following text for style: Improve the grammar, sentence structure, and clarity without altering the content, my argument, or the tone. Mark the parts you have changed. [Insert your own text]"
Key benefit: The content remains yours – RAI just helps with the finishing touches. This is comparable to proofreading and is one of the least critical yet most useful applications of AI.
Model recommendation: GPT-5 Mini via Azure is suitable for public or internal texts (C1–C2). If the text contains confidential content (C3, e.g., committee decisions, personnel data), use Mistral Large 2 or GPT-OSS 120B via the GWDG. RAI is not suitable for strictly confidential information (C4).
“Translate the following German text into academic English. Keep the technical terms as they are and mark any places where you are unsure whether the technical term has been translated correctly. [Insert your own text]"
Key benefit: Scientific translations are time-consuming, and DeepL has its limitations when it comes to technical terminology. RAI can provide a solid initial draft, which you can then review for technical accuracy.
Recommended model: For public or internal texts (C1–C2): GPT-5 Mini via Azure (very good English). For confidential texts (C3, e.g., unpublished research papers): Gemma 3 27B via the GWDG (supports over 140 languages). RAI is not suitable for highly confidential information (C4).
“Summarize the following document in no more than 10 sentences. Focus on: key points, agreed-upon actions, and outstanding issues. [Insert document]"
Key benefit: In everyday work, minutes, guidelines, or reports need to be recorded quickly. RAI saves time during the initial review – but the summary should always be checked against the original.
Recommended model: For public or internal documents (C1–C2): GPT-4.1 Mini via Azure (particularly fast, follows instructions well). For confidential documents (C3, e.g., committee meetings, personnel matters): GPT-OSS 120B or Mistral Large 2 via the GWDG. RAI is not suitable for strictly confidential information (C4).
“Rewrite the following draft email to [target audience, e.g., students / international partners] so that the tone is friendly and clear, while retaining all the relevant information. [Insert your draft text]"
Why this is a good idea: The message comes from you – RAI simply helps you craft it in a way that is tailored to your target audience and easy to understand. This is especially helpful in multilingual communication or when it's difficult to strike the right tone.
Model recommendation: If the email contains only public information (C1), any model is suitable. If it contains internal content without sensitive personal references (C2): GPT-5 Mini via Azure. If it contains names with personal feedback or sensitive HR issues (C3): Mistral Large 2 via the GWDG. RAI is not suitable for strictly confidential information (C4).
“I get the following error when running my Python script: [insert error message]. Here is the relevant code snippet: [Insert code]. Please explain what is causing the error and how to fix it.”
Key benefit: Interpreting error messages is a clearly defined task where AI has already demonstrated its effectiveness.” The learning effect is preserved because the explanation promotes understanding rather than simply spoon-feeding the solution.
Model recommendation: Source code is typically public or internal (C1–C2): GPT-4.1 Mini via Azure (specifically optimized for coding, very fast) or O4 Mini for more complex problems. If the code contains confidential research logic (C3): R1 Reasoning LLM or GPT-OSS 120B via the GWDG (very helpful with technical issues). RAI is not suitable for highly confidential information (C4).
“Please rewrite the following text in simple language (level B1) that is easy to understand for international students and non-native speakers. The content must remain in its entirety. [Insert text]"
Key benefit: Accessibility and clarity are essential at an international university. RAI can help make existing texts more accessible. This is a clear benefit that doesn't harm anyone.
Recommended model: For public-facing text (C1, e.g., website content): GPT-5 Mini via Azure. For internal documents (C2): Azure models are also suitable. For confidential documents (C3, e.g., examination regulations): Gemma 3 27B or Mistral Large 2 via the GWDG. RAI is not suitable for strictly confidential information (C4).
Background: HAWKI and GWDG
HAWKI is an open-source platform for generative AI that is used at many universities. HAWKI was developed with the aim of providing university staff with low-threshold, pedagogically supported access to generative AI and fostering discussion about meaningful ways to use it. RAI uses a HAWKI 2 user interface.
RAI offers two model backends: OpenAI models via Microsoft Azure in the EU, and OpenWeights models hosted on servers operated by the Gesellschaft für wissenschaftliche Datenverarbeitung mbH Göttingen (GWDG).
Models in RAI and selection guide
RAI offers several AI models. The plans differ in terms of performance, data limits, and hosting.
- Azure (EU): OpenAI models running on Microsoft Azure servers in the EU (permitted for data classes C1–C2). Microsoft stores input data for 30 days.
- GWDG (Göttingen): Open-weight models running on GWDG servers (permitted for data classes C1–C3; documented approval is required for C3).
Which information may be entered into which system depends on the confidentiality level and the type of system. If in doubt, err on the side of caution: until the matter is clarified, use only public content (C1). The AI systems available in the portfolio are generally not suitable for highly confidential information (C4).
The following overview will help you choose the model that best suits your needs.
GPT-5.4 is one of the newer versions of OpenAI's GPT-5 generation and is available via Azure (EU region). GPT-5.4 represents a further refinement in large language model capabilities, focusing on improved reasoning, multimodal understanding, and more reliable long-context processing. It builds on earlier GPT-5 iterations by enhancing factual consistency, reducing hallucinations, and offering better tool integration for complex workflows such as coding, data analysis, and agent-based systems. With more efficient inference and stronger alignment mechanisms, GPT-5.4 is designed to support both advanced research applications and scalable real-world deployments.
Details:
- Developer / Country: OpenAI / United States
- Company: Azure (EU)
- Data classes: Permitted: C1–C2; not permitted: C3–C4
- Context: 1M input, 128k output
- Text performance: 2 - high
- Programming performance: 2 - high
- Performance with understanding images (vision): currently not supported (2 – high)
GPT-5 Mini is a compact version of OpenAI's GPT-5 generation and is available via Azure (EU region). Compared to earlier GPT-4 models, this model offers significantly improved reasoning, coding, and text-generation capabilities, at a lower cost and with greater speed than the larger GPT-5. It serves as a standard all-rounder for longer texts, summaries, programming tasks, academic writing, and general academic or administrative topics, producing very good English and very solid German.
Details
- Developer / Country: OpenAI / United States
- Company: Azure (EU)
- Data classes: Permitted: C1–C2; not permitted: C3–C4
- Context: 128,000 tokens (≈ 96,000 words ≈ 320 pages)
- Text performance: 2 - high
- Programming performance: 2 - high
- Performance with understanding images (vision): currently not supported (2 – high)
O4 Mini is a dedicated reasoning model from OpenAI designed specifically for tasks involving many computational steps. It is smaller and more affordable than earlier o-models, yet it offers a very high level of performance in mathematics, programming, and scientific problems, and employs more detailed step-by-step reasoning internally. This makes O4 Mini particularly well-suited for complex computational or logical tasks, draft reports, or technical analyses; for simple chats, GPT-5 Mini is usually faster.
Details
- Developer / Country: OpenAI / United States
- Company: Azure (EU)
- Data classes: Permitted: C1–C2; not permitted: C3–C4
- Context: 128,000 tokens (≈ 96,000 words ≈ 320 pages)
- Text performance: 2 - high
- Programming performance: 2 - high
- Performance with understanding images (vision): currently not supported (2 – high)
GPT-OSS 120B is an "open-weight" model from OpenAI, designed for strong reasoning performance and agent-based use cases. It achieves benchmark results comparable to O4 Mini while remaining fully deployable on-premises. On this platform, GPT-OSS 120B runs on servers operated by the GWDG in Göttingen, ensuring that all data remains within their infrastructure. It is particularly well-suited for computationally intensive analyses, planning and coding tasks, as well as research projects where transparency and data sovereignty are important.
Details
- Developer / Country: OpenAI / United States
- Company: GWDG
- Data classes: Permitted: C1–C3; not permitted: C4
- Context: 128,000 tokens (≈ 96,000 words ≈ 320 pages)
- Text performance: 2 - high
- Programming performance: 2 - high
- Performance understanding images (vision): not supported
Gemma 4 31B Instruct is an open-source model developed by Google that is based on Gemini technology and can process text and image inputs. It features a large context window and supports over 140 languages, making it particularly well-suited for multilingual scenarios and the combination of longer documents and illustrations. When it comes to deep mathematical reasoning, however, models like R1 or GPT-OSS usually come out on top.
Details
- Developer / Country: Google / United States
- Company: GWDG
- Data classes: Permitted: C1–C3; not permitted: C4
- Context: 256,000 tokens
- Text performance: 2 - high
- Programming performance: 2 - high
- Performance understanding images (vision): 2 - high
Llama 3.1 8B is a compact language model from Meta that serves as an efficient model for many standard tasks. It is optimized for conversational applications and general text processing, offers a large context window and very low latency, but lags significantly behind larger models such as Mistral Large 2 or GPT-OSS when it comes to complex reasoning and demanding coding tasks. Because of its efficiency, it is particularly well-suited for simple inquiries, brief summaries, initial drafts, or teaching and practice scenarios involving many concurrent requests.
Details
- Developer / Country: Meta/ USA
- Company: GWDG
- Data classes: Permitted: C1–C3; not permitted: C4
- Context: 128,000 tokens (≈ 96,000 words ≈ 320 pages)
- Text performance: 1 - solid
- programming performance: 0 - low
- Performance understanding images (vision): not supported
Mistral Large 2 is the flagship model of the French company Mistral AI (Mistral AI), specifically designed to deliver strong multilingual text processing, coding, and reasoning capabilities. It is suitable for complex analyses, longer academic texts, and challenging programming tasks. In a direct comparison, Mistral Large 2 generally falls between the smaller models (Llama 3.1 8B) and the specialized reasoning models (R1, GPT-OSS) – offering a good balance of quality and speed.
Details
- Developer / Country: Mistral / France
- Company: GWDG
- Data classes: Permitted: C1–C3; not permitted: C4
- Context: 128,000 tokens (≈ 96,000 words ≈ 320 pages)
- Text performance: 1 - solid
- Programming performance: 1 - solid
- Performance understanding images (vision): not supported
This model from the Chinese company DeepSeek is specifically designed for complex logical reasoning as well as mathematical and technical tasks. In benchmark tests, it achieves very high scores in mathematics, programming, and problem-solving. It works most reliably in English and Chinese; German is well supported, but the stylistic quality may be somewhat weaker. R1 is significantly slower than most other models and may exhibit content-related biases when dealing with politically sensitive topics; it is therefore particularly well-suited for technical and scientific questions that require a high level of reasoning.
Details
- Developer / Country: DeepSeek / China
- Company: GWDG
- Data classes: Permitted: C1–C3; not permitted: C4
- Context: 32,000 tokens (≈ 24,000 words ≈ 80 pages)
- Performance text: 3 - very high
- Programming performance: 3 - very high
- Performance understanding images (vision): not supported
Guidelines for responsible use
- Confidentiality
-
Check the confidentiality class and select the appropriate hosting option (Azure: C1–C2; GWDG: C1–C3, C3 only with documented approval).
- Disclosure requirements
-
AI-assisted content must be disclosed if an AI system significantly contributes to the creation of content or if content is published without further human editing or quality assurance (e.g., website content, official correspondence, emails to external recipients).
- Training courses
-
Complete training before using the system for work purposes. Such courses raise awareness about bias, discrimination, and hallucinations, as well as sustainable, digitally sovereign use.
- Report errors
-
If you make a mistake (enter the wrong system or incorrect data), report the incident to the Data Protection and Information Security department as soon as possible.
Training, tutorials, videos
Video tutorials and training courses are available on the AI learning portal in ILIAS. Training is mandatory prior to official use; upon successful completion, a certificate will be issued, which must be presented to the supervisor before the device is used for official purposes.
Frequently asked questions about RAI
Yes. RAI is accessible only on the university network or via VPN. If you are off campus, you must first establish a VPN connection. You can download the VPN client here. Instructions can be found on the TIK website.
The input field for the Data Key in RAI (or HAWKI 2) prevents pasting from the clipboard. The key must therefore be entered manually – even if you use a password manager. It is important to choose a password that is easy to remember and type into different devices. You should also save it in a password manager so it doesn't get lost. An update that enables copy-and-paste insertion is planned and will be rolled out in the near future.
After logging in, click in the input field. The model currently in use is displayed on the right. Click the small arrow next to the model name to open the model list. GWDG models are labeled accordingly (all names begin with “GWDG”), while Azure models begin with “Microsoft Azure”. You can find a visual guide in the RAI tutorials.
That depends on your specific use case and the sensitivity of your data. For sensitive data (up to C3), use the GWDG models. For general tasks involving public or internal data (C1–C2), Azure/OpenAI models often offer higher quality and faster performance. You can find a detailed overview in the “Model Comparison” section on this page.
The use of this service is free for all university members. However, depending on the model, the university incurs costs per unit of use. Therefore, please use RAI responsibly and in a manner that conserves resources, in line with the principle of fair use.
That depends on the confidentiality level. Azure (EU) is intended for C1–C2. The GWDG variant is intended for C1–C3; documented approval is required for C3. The AI systems available in the portfolio are generally not suitable for C4. Overview of confidentiality levels
No. Input data is not used for training in RAI – neither for the GWDG models nor for the Azure/OpenAI models.
In the Azure (EU) version, input data is stored by Microsoft for 30 days. For GWDG-supported models, operations take place on GWDG servers. Prompts and chat content are not stored there.
First, check what type of data is affected. If personal or confidential data (C2 or higher) has been entered into an unsuitable system, promptly notify your supervisor and report the incident to the Data Protection Officer and the Information Security Office. Reporting an incident promptly helps limit risks and ensure it is handled properly. For purely public content (C1), notification is generally not required.
Your chats in RAI are stored with client-side encryption. When you log in for the first time, you will be prompted to create a personal key. Choose a password that is easy to remember. You will be required to enter it manually on other devices or in private browsing mode. You can reset the key, but doing so will result in the loss of all saved chats.
Yes, if an AI system plays a significant role in creating content (e.g., drafting text) or if AI-generated content is published without further editing (e.g., a website, an official letter, or an email to external parties).
If the use is very limited (e.g., individual formulation suggestions), you are not required to disclose it. When responding to students or external parties, it must be clear that an AI system was used. This policy applies to all AI systems, not just RAI. For details, see the AI guidelines.
Yes. Mandatory training is required for the professional use of AI systems – including RAI. It covers the basics of confidentiality classes, system types, and the university’s AI policy, and raises awareness of issues such as bias, discrimination, and hallucinations, among others. A certificate will be issued upon successful completion of the course, which must be presented to your supervisor before AI can be used for work purposes. You can find the training courses on the AI Learning Portal in ILIAS. One of the reasons for this is the EU AI Regulation (Art. 4), which establishes minimum standards for AI literacy for all employees who work with AI systems.
RAI supports the understanding and analysis of images and PDFs (Vision) in select models. Image generation is currently not available in RAI. You can use additional approved tools from the whitelist to generate images of for other specialized tasks.
Yes, RAI offers a chat export feature. Occasional formatting issues (e.g., truncated lines) may currently occur during PDF export. A solution is being worked on.
Personal API access is currently unavailable. The platform's capacity is limited, and custom integration (e.g., with VS Code or your own applications) is not currently possible. We are aware that there is high demand for API access – this is a top priority for us, and we are actively working on it. As soon as API access becomes available, an announcement will be made.
RAI works best in Chrome and Firefox. Display issues may occasionally occur in Safari. Refreshing the page (page reload) often helps as a workaround. If the problem persists, try using a different browser.
Reasoning models in particular (e.g., O4 Mini, R1) require more processing time and can cause timeouts with complex queries. In this case, try to make the prompt shorter or less complex, or switch to an all-purpose model like GPT-5 Mini. Contact Support if the problem persists.
That is a well-known problem. If a request is still in progress and a new chat is opened at the same time, replies may appear in the wrong chat, or display errors may occur. If this is the case, close the chat and start a new one.