Support and FAQ

Warning

This service is currently in beta phase and is updated regularly. The same applies to the documentation.

If you run into problem using this service please contact our KISSKI support

We also need your help providing publicly available Arcana links. If you have one that you think is relevant to a larger group of users, please also reach out to us so we can publish it.

FAQ

Which models support RAG (Retrieval-Augmented Generation)?

All internal models work with the GWDG-Tools and Arcanas. But some models are better at using Tools then others!

We recommend using: Qwen 3 30B A3B Instruct 2507 for the best results with tool calling and document retrieval.

We do NOT recommend to use: Meta Llama 3.1 8B Instruct as its is not optimized for Tool-Calling!

Info

External models do not support Arcanas/RAG!

What is the maximum file size I can upload?

You can upload files up to 300 MB per file.
For optimal performance, we recommend keeping files under 50 MB - larger files may take longer to process.

Which file types are supported?

The following formats are fully supported:

  • Text: .txt
  • Markdown: .md
  • Word Documents: .docx, .dotx, .docm, .dotm
  • PowerPoint Presentations: .pptx, .potx, .ppsx, .pptm, .potm, .ppsm
  • PDFs: .pdf
  • HTML: .html, .htm, .xhtml

All files are converted into a machine-readable (Markdown) format during indexing.

Is my data stored? Where is it stored?

Your documents and indexed data remain exclusively on GWDG systems at all times.
We do not store, share, or transmit your data to third parties.
Your privacy and data security are our top priorities.

Can I share my Arcana?

Yes - Arcanas can be freely shared with other users.
To share, simply provide the Arcana ID or the full Arcana name.

Info

By default, every Arcana is public. This means anyone who knows your username and the exact Arcana name can access it.
However, they cannot view or edit your uploaded files - they can only interact with the Arcana via the chat interface.

How do I create a private Arcana?

To keep your Arcana private, enable the “Secure Arcana Name” option when creating it.
This adds a random, unique security token to the Arcana name, making it accessible only to you or anyone who knows the full, exact name.

How does RAG (Retrieval-Augmented Generation) work?

RAG enhances large language models (LLMs) by combining document retrieval with generative AI.

Here’s how it works:

  1. Your query is matched against a knowledge base (your uploaded documents).
  2. The system retrieves the most relevant passages (5 Passages).
  3. The LLM uses this context to generate accurate, informed, and contextually relevant responses.

This allows the model to draw from external knowledge beyond its training data, resulting in:

  • Higher factual accuracy
  • More coherent and relevant answers
  • Reduced hallucinations

How are my requests processed?

Every step of your request is handled entirely on GWDG infrastructure:

  • Document embedding generation
  • Retrieval of relevant content
  • Final inference and response generation

Your data and queries never leave GWDG systems, and are never shared with third parties.

Why is the indexing taking so long?

Indexing involves advanced machine learning algorithms to convert complex documents (like .docx, .pdf, .pptx) into a structured, machine-readable format (Markdown).

This process can take time, especially for:

  • Large files (>50 MB)
  • Documents with complex layouts, images, or tables

To speed up indexing, keep file sizes under 50 MB and avoid overly complex formatting.

Further Questions?

Let us know if you have any further questions!