The KI-Toolbox

The AI toolbox gives all KIT employees access to various local and external language models (LLMs). There are two separate ways to use them: a web interface for interactive use and an application programming interface (API) for integration into your own applications.

Contact: ki-toolbox∂scc.kit.edu

The web interface is aimed at users without programming knowledge and is suitable for exploratory tasks such as generating, summarizing and translating texts or answering questions. The API is used to integrate LLM functions into scripts, services or automated workflows. It is intended for repeatable, scalable or batch processes and requires programming knowledge.

The AI toolbox is operated within the framework of the applicable KIT guideline. Depending on the model, requests are processed locally or externally. Users are responsible for taking this into account. For more details, please refer to the SCC service description under the section "Data categories, protection classes and model selection" in the terms of use.

ⓘ We regularly offer accompanying webinars on the introduction, new functions and in-depth exploration of individual topics.

You can find current dates in the ZML events calendar

Access

The KI-Toolbox is currently only accessible in the KIT network:

If you are located outside the campus network (e.g. in your home office), you must currently establish an active VPN connection to KIT in order to access the interface.

ⓘ Please note that you can only use the KI-Toolbox once you have successfully completed the AI competence module on KOALA.

Instruction (PDF)

To get started

For advanced users

For professionals

Standard-Funktionen

Login
All KIT employees can log in to the AI Toolbox with their KIT ID and password at: https://ki-toolbox.scc.kit.edu/


 

ⓘ If you are located outside the campus network (e.g. in your home office), you must currently establish an active VPN connection to KIT in order to access the interface.

Please note that you can only use the AI-Toolbox once you have installed the AI-Kompetencemodul at KOALA successfully completed. Access to the AI toolbox will then be activated after approx. 5 minutes.

Start chat    
In the input window, you can start your chat directly with one of the available language models.
 

How to create a new chat:

  1. You can start another new chat at any time via the menu. To do this, click on the "New Chat" in the left sidebar. The main window is emptied and is ready for your entries.

  2. The AI toolbox provides you with various Models. Information on the strengths and applications can be found next to or below the respective model.

  3. Now enter your question or instruction in the Input field. Alternatively, you can also dictate your question by clicking on the microphone symbol to the right of the input field. Then press the Enter key or click on the send arrow. The AI will now process your request and generate an answer in the chat window. You can continue the dialog as you wish by asking further questions. The context of the previous conversation is retained. Below the input field, you will find additional tools that extend the AI's capabilities. These may vary depending on the model selected.

Screenshot des GPT-OSS-Interfaces mit Seitenleiste und Eingabefeld "How can I help you today?"

 


A brief overview of the functions (for more information, see below or in the instructions):

  • Create and use your own prompt collections (available in the chat via / ).

  • Save your own knowledge in the form of files and texts (available in the chat via #).

  • Create your own models and link them with prompts and knowledge.

  • Use voice input and output (see below).

Overview: Models

ⓘ  Note: The models listed in this table are subject to change. This list may not always be up to date.

 

Modell Betriebsart Intelligenz Geschwindigkeit Eingabe Ausgabe Herkunft Preis pro 1M Tokens
standard-local* Lokal
(Personenbezogene Daten erlaubt)		 💡💡💡 ⚡️⚡️⚡️⚡️ Text Text SCC interne Kosten
qwen3-vl:235b-a22b Lokal
(Personenbezogene Daten erlaubt)		 💡💡💡💡 ⚡️⚡️⚡️ Text/Bild Text Alibaba interne Kosten
mixtral:8x22b Lokal
(Personenbezogene Daten erlaubt)		 💡💡💡 ⚡️⚡️ Text Text MistralAI interne Kosten
gpt-oss:120b Lokal
(Personenbezogene Daten erlaubt)		 💡💡💡 ⚡️⚡️⚡️⚡️ Text Text OpenAI (lokal) interne Kosten
Standard-External* Extern
(keine personenbezogenen Daten erlaubt)		 💡💡💡 ⚡️⚡️⚡️⚡️ Text/Bild Text OpenAI $0.25 in/$2 out
GPT-4.1 Extern
(keine personenbezogenen Daten erlaubt)		 💡💡💡💡 ⚡️⚡️⚡️ Text/Bild Text OpenAI $2 in/$8 out
GPT-4.1 mini Extern
(keine personenbezogenen Daten erlaubt)		 💡💡💡 ⚡️⚡️⚡️⚡️ Text/Bild Text OpenAI $0.40 in/$1.60 out
GPT-4.1 nano Extern
(keine personenbezogenen Daten erlaubt)		 💡💡 ⚡️⚡️⚡️⚡️⚡️ Text/Bild Text OpenAI $0.10 in/$0.40 out
o3 Extern
(keine personenbezogenen Daten erlaubt)		 💡💡💡💡💡 ⚡️ Text/Bild Text OpenAI $2 in/$8 out
o4-mini Extern
(keine personenbezogenen Daten erlaubt)		 💡💡💡💡 ⚡️⚡️⚡️ Text/Bild Text OpenAI $1.10 in/$4.40 out
GPT-5 Extern
(keine personenbezogenen Daten erlaubt)		 💡💡💡💡 ⚡️⚡️⚡️ Text/Bild Text OpenAI $1.25 in/$10 out
GPT-5 mini Extern
(keine personenbezogenen Daten erlaubt)		 💡💡💡 ⚡️⚡️⚡️⚡️ Text/Bild Text OpenAI $0.25 in/$2 out
GPT-5 nano Extern
(keine personenbezogenen Daten erlaubt)		 💡💡 ⚡️⚡️⚡️⚡️⚡️ Text/Bild Text OpenAI $0.05 in/$0.40 out
Microsoft Copilot (EDP) Extern
(keine personenbezogenen Daten erlaubt)		         Microsoft  

* A user-defined system prompt is stored for these models.

 

  • Intelligence: The more 💡, the "smarter" the model.
  • Speed: The more ⚡️, the faster the model delivers results.
  • Price: "in" = upload/processing, "out" = output/response.

Important: Personal data (Art. 4 No. 1 GDPR) may only be processed via the models hosted locally at KIT. Processing personal data via the Azure OpenAI models is prohibited.

 

Working with files
You can create questions or tasks based on files (alternatively, longer texts can also be copied into the input field). Please note that such requests take a little more time and the context length varies depending on the model.
 

You can upload files and images live in the chat. To do this, use the Plus-Symbol (+) in the input window.

Toolbox upload

">If you use files repeatedly in your requests, it is advisable to save them permanently as "Knowledge" for you personally (see "Advanced functions").
UI-Screenshot: Seitenleiste Workspace, Knowledge-Tab, Button '+ New Knowledge', 'No knowledge found'.

You can call up saved knowledge at any time in the input window with the Hash key "#".

Screenshot der GPT-5 KI-Toolbox mit Wissensspeicher (Test) und Eingabefeld.

Ⓘ Tipps for working with documents in der AI-Toolbox 

Using voice input and output
If you want to use the audio functions, it is recommended that you specify the language settings individually (usually "German"). Alternatively, automatic recognition is used, but this does not always lead to optimal results.
 

To do this, go to your personal settings (profile picture top left or bottom right) and navigate to "Settings" > "Audio" > "Language" and enter "en". You can leave the other values as they are. If necessary, you can set a different preference under "Set voice". Make sure that the voice matches the set language (e.g. "de" and "German").

Toolbox Audio
 

You now have several options within a chat:

  • Dictate (microphone symbol): Instead of typing entries, you can have them recorded and edit them again in writing before sending them.

  • Voice mode (Soundwave symbol): Here you can speak directly to the model and have the answers read out to you.

  • Reading: Regardless of the mode in which you have started a conversation (text, dictation, voice mode), you can always activate and deactivate reading aloud by clicking on the loudspeaker symbol under the AI output.

Toolbox Dictate

 

Advanced functions

Create your own prompts
You can save frequently used prompts so that you can easily reuse them in your chats in the future.
 
To do this, navigate to"Workspace" >"Prompts" via the left-hand main menu
You can create new prompts using the"+ New prompt" button. Within the input mask for prompts, you can also select via"Access" whether and with which group this prompt should be shared.
 
Toolbox Prompts
 
You can access the prompt stored in this way directly in the chat with a slash ("/").
Prompts Toolbox
Create your own knowledge base
It is worth creating a knowledge store if you want to use documents repeatedly in different chats as a knowledge base. You create a kind of private library that you can access as required.
 
Navigate to"Workspace" >"Knowledge" via the left-hand main menu
Use the"+ New knowledge" button to create a new knowledge store, which you can fill with knowledge in the form of files or text in the next step. Enter the name and function of the knowledge store. Then click on"Create knowledge". You can now upload your documents using the plus symbol (+) on the right.
 
Knowledge AI-Toolbox
To edit an existing knowledge base, select it directly. Within the knowledge base, you can then add entries using the plus symbol and use "Access" to control who you want to share this knowledge base with.
 
You can refer directly to the knowledge created in this way in chats using the hash symbol "#" to retrieve the saved data. The AI can then access it when processing requests.
Toolbox generate knowledge

ⓘ PDF instructions for working with documents in the AI.toolbox
Creating your own custom models (assistants)
AI chatbots are digital assistants that you work with in natural language: You can ask questions, write texts, structure information or prepare tasks. The assistants respond to your instructions, remember rules and can even work with your own documents. This allows you to complete routine tasks faster and achieve consistent results.
 
A chat assistant needs the following core components an assigned knowledge store and a system prompt.
 
How to create your own assistant: Navigate to "Workspace" > "Models" via the left-hand main menu. You can create a new model using the "+ New model" button.
 
Toolbox Models
 
You should configure the following points in the editing screen:

 

  • Model name*: Mandatory information.
  • Base model*: One of the available language models must be selected.
  • Description: It is recommended to set a meaningful description for all subsequent users.
  • Visibility: In this section, you can control whether and for whom the model is shared: privately (only for you) or specific groups to which you belong.
  • Model parameters: A system prompt will be very useful for most cases. Advanced parameters are optional (depending on the use case).
  • Prompt suggestions: Are optional, but can help later users to better implement the use case.
  • Knowledge: In many cases, a knowledge base will be essential for the use case. If no knowledge base is required, it may not be necessary to create a custom prompt, as the use of a (saved) prompt is already sufficient in such cases. It should also be noted that if custom prompts are shared, the associated knowledge must also be shared with the same group.
  • Tools: Can be added optionally depending on the use case.
  • Skills: Are only to be understood as "tags". The actual rights that a user has on the basis of a custom model depend on the selected base model.
ⓘ Instructions for working with chatbot assistants