TUDaGPT – The AI chat assistant from TU Darmstadt

TUDaGPT is the AI chat assistant of TU Darmstadt. Students and staff can work with large language models (LLMs) via a chat interface. They can ask questions, draft and edit texts, analyse documents or have programme code generated.

Since October 2026, TUDaGPT has been based on the open-source platform Open WebUI and has been completely rebuilt. The user interface is connected to a central model gateway. This gateway is used to run the AI models on servers operated by hessian.AI, within the GWDG’s infrastructure and, in future, also on servers at Johannes Gutenberg University Mainz.

The application itself is provided for TU Darmstadt by the computing university centre (HRZ) and runs on servers within the university's own environment.

What's new?

Since October 2026, TUDaGPT has been available in a technically updated version featuring the following new features:

  • New interface: TUDaGPT now uses Open WebUI.
  • Web search: Web search can be enabled for each chat to access up-to-date information. Responses include references to their sources.
  • Files and images: Documents and images can be uploaded directly into the chat and analysed.
  • Organise chats: Chats can be searched, saved in folders, pinned and exported.
  • More models: Models from hessian.AI and GWDG, with models from JGU Mainz to follow.
  • API: A standards-compliant, OpenAI-compatible interface is available.
  • New address: https://tudagpt2.hrz.tu-darmstadt.de

Important note regarding the switch: Chats from the previous version will not be transferred. The old address https://tudagpt.hrz.tu-darmstadt.de will remain accessible until the end of Q1 2027.

Access and Data Protection

  • Access: TUDaGPT can only be accessed within the TU network or via a VPN connection.
  • Internet access and web search: The AI models themselves do not have access to the internet. A connection to an external search service is only established if you specifically enable web search within a chat. Web search must be enabled individually for each chat. This ensures that sensitive data does not inadvertently leave the secure university environment.

Step 1: Log in

Log in with your TU ID (username) and password via single sign-on. When you log in for the first time, you will see an overview of the personal data you have provided.

Step 2: Choose a model

The currently selected model is shown above the chat window. The default setting is Auto, in which TUDaGPT automatically chooses a suitable model. To choose a different model, click the model selector (at the bottom right of the chat window). You can also switch models during an ongoing chat.

Step 3: Enter your question

Type your question or task into the input field and press Enter to send it. Use Shift + Enter for a line break. Add files via the “+”-symbol or by drag and drop.

Step 4: Use the answer

The answer appears in the chat window. Below each answer you can copy it, regenerate it or edit your question. Your previous chats are listed in the sidebar on the left; start a new one with New Chat.

  • You can ask further questions or follow up as you wish.
  • Each new entry appears below the previous conversation.

Without web search, the models answer only from their training data, which ends at a certain point in time. For current topics, news, dates or prices, switch on web search. TUDaGPT then searches the internet and lists the sources it used.

Important: Web search applies only to the current chat. It is switched off in every new chat and must be activated again when needed.

Step 1: Open a chat

Start a new chat or select an existing one.

Step 2: Open the menu

Below the input field, click [Integrations] (the diamond icon next to the “+”-symbol).

Step 3: Switch on web search

Turn on the Web Search toggle. An indicator at the input field shows that web search is active.

Step 4: Ask and check the sources

Ask your question as usual. The answer takes a little longer. The sources used are listed below the answer; open them to verify the information.

Data protection and web search

When web search is enabled, a search query is generated based on your input and sent to the DuckDuckGo search service. Uploaded files and images are not transmitted. However, content from the previous chat history may be included in the search query. Therefore, do not enter any personal or confidential information when web search is enabled.

More information: Open WebUI documentation on web search

  • Upload files: Upload PDFs, Word documents, text or images to the chat using the plus sign (+) or via drag-and-drop, and ask questions about them.
  • Knowledge: Create your own document collections under Workspace › Knowledge and add them to a chat with #.
  • Manage chats: Search, rename, move to folders, pin, archive or delete chats in the sidebar.
  • Export: Download a chat as text, PDF or JSON from its menu.

1. Check carefully

Answers may be incomplete or wrong. Always check the results, including the sources found by web search.

2. Observe permitted uses

Permitted for use in support of studies, teaching and work; not permitted for assessments, examinations, job applications and personnel decisions.

3. Information security and data protection

Chats, uploaded files and settings are stored on TU Darmstadt servers until you delete them. For processing, your input is sent to the AI models at hessian.AI or GWDG (and in future JGU Mainz), where it is not stored. Corresponding agreements are in place with all operators.

Web search is an exception: if you enable this, search queries will be sent to an external search service.

4. Comprehensive guideline for responsible use of TUDaGPT

Please read the guidelines for the use of TUDaGPT before using the service for the first time. With the use of TUDaGPT you agree to this guideline.

TUDaGPT offers several AI models with different strengths. By default, Auto is selected and TUDaGPT automatically chooses a suitable model. Use the model selector at the bottom right of the chat window to choose a specific model and switch at any time. The selected model receives your request together with the previous conversation.

The respective model provider is displayed before the model name in the model selection.


Which AI model is right for me?

Each AI model has its own strengths and weaknesses, which depend largely on the training material used and the IT resources employed. By choosing Auto, you leave the selection to TUDaGPT.

Tip: Many models were trained mainly on English texts. For difficult tasks, English prompts often give better results.

The API lets you use the AI models from your own programs, scripts or tools. TUDaGPT uses the standard Open WebUI API without TU-specific modifications. The chat endpoint is compatible with the OpenAI API, so existing OpenAI libraries work directly.

Create an API key

  • In TUDaGPT, click your profile picture and then Settings.
  • Open Account and create a new key under API Keys.
  • Copy the key and keep it safe. Do not share it.

Connection details

Base URL: https://tudagpt2.hrz.tu-darmstadt.de/api

Authentication: header Authorization: Bearer <API key>

All endpoints and further examples are described in the Open WebUI API documentation.

# List available models

curl -H "Authorization: Bearer $TUDAGPT_KEY" \

https://tudagpt2.hrz.tu-darmstadt.de/api/models

# Chat request

curl -X POST https://tudagpt2.hrz.tu-darmstadt.de/api/chat/completions \

-H "Authorization: Bearer $TUDAGPT_KEY" \

-H "Content-Type: application/json" \

-d '{"model": "MODEL-ID", "messages": [{"role": "user", "content": "Hello!"}]}'

Detailed instructions for all features are available in the official platform documentation:

Open WebUI documentation

Feature overview

Note: Not all features described there are enabled in TUDaGPT.