# FAQ & Troubleshooting ## General **Q: Who can use uniGPT?** A: All students and employees of the University of Münster with a valid university account. **Q: Is uniGPT free?** A: Yes, uniGPT is free for all university members. However, usage limits apply to ensure fair access. **Q: Where do I find the usage policy and privacy statement?** A: They are linked below the chat input field on [gpt.uni-muenster.de](). ## Data Privacy **Q: Is my data safe?** A: Yes. uniGPT is operated by the University of Münster in compliance with the GDPR. The level of data protection depends on which model you use: - **Internal models** (e.g., DeepSeek V4 Flash, gpt-oss-120b, gemma-4-31B-it, medgemma-1.5-4b-it, Qwen3.5-35B-A3B): Your data is processed entirely on university infrastructure and **never leaves the university**. You may transmit information classified up to **confidential (V3)** according to the [university's information classification guideline](). Information classified as **strictly confidential (V4) must not** be processed. - **External models** (e.g., GPT-4, Gemini, Claude): Your data is transmitted to external providers (Google Cloud EMEA Limited, OpenAI Ireland Limited, Microsoft Ireland Operations Limited). Only information classified as **public (V1)** may be transmitted — this includes all information that could also be published unrestrictedly elsewhere. However, important safeguards apply: - **No training on your data** — your chats will not be used for training any models, including those from Google, OpenAI, or Microsoft. - Short-term storage by providers may occur only as technically necessary for generating the response. This means you get **significantly better data privacy** than with free consumer offerings like ChatGPT or Gemini, where your data may be used for model training. > [!WARNING] > > **Information classification matters** > > | Model type | Maximum classification allowed | > | --- | --- | > | **Internal models** | Confidential (V3) | > | **External models** | Public (V1) only | > | **Both** | Strictly confidential (V4) is **never** allowed | > > When in doubt, use internal models and consult the [information classification guideline](). **Q: Which models should I use for sensitive data?** A: Always use **internal models** (DeepSeek V4 Flash, gpt-oss-120b, gemma-4-31B-it, medgemma-1.5-4b-it, Qwen3.5-35B-A3B) for any data that is confidential or subject to GDPR. ## Common Issues **Q: The model is not responding / I get an error "terminated".** A: This can happen when a model (especially gpt-oss-120b) is overloaded. Try again after a few minutes or switch to a different model. **Q: My balance is 0 but it should have reset.** A: Try logging out and logging back in. Your balance resets every 4 hours. **Q: The model seems to "think" for a long time without showing anything.** A: Some models support a "thinking" (reasoning) mode. There is a known issue where the thinking process is not always displayed — the model is working, and the response will appear once reasoning is complete. **Q: Where are my old chats from the previous interface?** A: Legacy chats are available at [gpt-legacy.uni-muenster.de](). You will need to log in again. **Q: I get the error "The agent is configured to use MCP tools, but none are available."** A: MCP is currently disabled. Refreshing the chat usually resolves this error. **Q: I get the error "team not allowed to access model".** A: The requested model is not enabled for your account/group. Models are enabled per team — contact the support channel to request access. Currently **DeepSeek-v4-flash** and **Qwen3.8-27B** are available to all employees via the self-service portal. **Q: Can external researchers or doctoral students get an API key?** A: The self-service AI-coding portal requires membership in the `u0mitarb@uni-muenster.de` group. If you are not a member (e.g., external doctoral students), please request a **research-project API key** instead via the support channel. **Q: Can I access external model with the uniGPT API?** A: This is currently not supported. Only internal models are supported through the API. ## Getting Help If you encounter issues or have questions, reach out via our support channel: [**uniGPT Support**]()