AG: Generative KI

AI-ToolLab: AI Chatbot

LobeChat

To communicate with Large Language Models (LLMs), users need an application that provides an interface (UI) where they can submit their questions and tasks, and where the responses generated by the connected LLMs are displayed.

One such chat UI is the open-source application LobeChat. LobeChat provides an appealing interface for communicating with AI chatbots and agents in the style of ChatGPT. The open-source application enables users to engage in advanced, AI-powered conversations and offers a wide range of features that make it particularly versatile and useful.

LobeChat has been configured by studiumdigitale for the members of Goethe University and made available for testing on servers within Goethe University.

➜ Start AI Chatbot

Applications in Teaching

  • Translation of content into different languages.
  • Creation of suggestions for tasks and sample solutions.
  • Creation of material lists for experiments.
  • Brief or age-appropriate summaries of complex texts
  • Generation of suggestions for the structure of a course
  • Research on the most important information on a topic.
  • Generation and revision of source code.

Advantages of Lobechat

  • Open Source: As an open-source application, LobeChat is freely accessible and can be further developed and adapted by the community. This promotes innovation and allows for continuous improvement of the application.
  • Flexibility: The support for various models and multimodal conversations makes LobeChat a versatile tool that can be used in many different contexts.
  • Customizability: Through the ability to create custom agents, personal preferences or repetitive work steps can be processed more effectively
  • User-friendliness: By integrating voice input and output, LobeChat offers an intuitive and accessible user experience suitable for both technical and non-technical users.
  • Data Protection & Anonymization: All chat and settings data remain in the users’ browser. Requests to external services (Azure, KissKI) are anonymized

Features of LobeChat

Support for various models:
LobeChat supports a wide range of AI models, making it flexible and adaptable for different use cases. Users can select and employ the appropriate model according to their requirements.

Azure OpenAIKissKIstudiumdigitale
ChatGPT models from OpenAIlarge, open models for various tasks small, open models for simple tasks
at Microsoft Azure in the EUin Germany by KissKI / GWDGat Goethe University by studiumdigitale
In addition to the most common OpenAI models, LobeChat provides access to several alternative models hosted by other providers. The list is constantly expanding.

Creation of own assistants

LobeChat offers the possibility to create your own assistants for recurring tasks. With the help of role prompts and other settings, users can develop their own assistants that support them in processing repetitive tasks more efficiently and mapping their own preferences.

Speech input and output

The platform offers both voice input and voice output, which makes interacting with the chat bots and agents particularly intuitive and user-friendly. This makes it possible to conduct natural language conversations and facilitates use for people with different needs and preferences.

Multimodal conversations

LobeChat supports multimodal conversations, which means that in addition to text, other media types (initially only images) can be integrated into the communication. This expands the possibilities of interaction and makes the conversations richer and more informative.

FAQs

How do I choose an alternative model?

At the bottom left, above the text input field there is a small icon representing the currently selected model. The available models can be selected there.

You can also change the model of the respectively selected assistant alternatively: at the very top.

Which model should I use?

It entirely depends on your use case.

  • The Azure OpenAI models are good all-rounders:
    • gpt-4o-mini: the smallest, cheapest model from OpenAI which is preset and offers good performance at a fair price.
    • gpt-4o: this model delivers slightly better results for complex tasks. However, its use is also slower and more expensive.
    • o3-mini: a reasoning model, optimized for scientific, mathematical, and code queries, but also well-suited for general inquiries. The costs are in the mid-range, between gpt-4o-mini and gpt-4o.
  • KissKI provides freely available, open models that can have different usage focuses.
    • LLama, Mistral, DeepSeek, or Qwen are good all-rounder models that can be used as alternatives to OpenAI
    • You can find a list of the models currently offered by KissKI here (we only integrate a selection of the models into LobeChat).
    • The general knowledge of these models may be somewhat limited compared to the OpenAI models, as they are only a fraction of the size.
    • However, these models can still perform special tasks (translating, working with texts, etc.) very well.
    • Some models are also optimized for certain fields or tasks. Therefore, it’s always worth experimenting and trying them out!
  • The locally hosted models are hosted by studiumdigitale at GU.
    • No data flows externally. Therefore, their use is the least problematic in terms of data protection law.
    • Due to very limited hardware resources, we can only host the “smallest” models.
    • Nevertheless, these can be well-suited for specific tasks (working with texts, making revisions, etc.). However, more complex queries may produce lower quality answers.

Conclusion:

  • Trying it out is the key:
    Simply try out which use cases can be best implemented with which models.
  • Do your own research:
    There are many comparisons and benchmarks that compare the quality of the models. If you want to find the “optimal” model for your usage scenario, look for information about the models and compare them.
Where are my chat messages and settings stored?

LobeChat stores all chat messages and settings in the user’s browser. Thus, no messages or settings are stored on the LobeChat server, and users have full control over their data.
Caution: This also means that if you clear your browser cache or switch browsers, you will lose your settings and previous conversations!

Is the use of LobeChat anonymous?

LobeChat runs entirely on servers within GU. The contents of the requests are not cached in the server logs.
However, depending on the chosen model, the requests to the LLMs are forwarded to external providers for processing (e.g., Azure). All requests first go through an LLM proxy server, which does not forward the user’s IP. Thus, communication with external providers is completely anonymous as long as no personal data is contained in the messages.

Important Notes

⚠️ Data protection: It is not permitted to send personal or sensitive data to the LLMs.

📚 Requirement: Completion of the training on the critical use of AI (according to EU AI-VO)

🎯 Fair-Use: Fair-use restrictions apply to GWDG models – avoid stress tests

Support

If you have any problems or questions, write an email to the AI-ToolLabs team.

Credits

LobeChat is developed by LobeHub.
studiumdigitale operates a preconfigured instance of LobeChat for users of Goethe University.
All images, logos, and used files are the property of LobeHub.
Apache License 2.0 | GitHub