Chat AI now has a new, unique function called “Arcana”, which can add additional information to a user request. This is the first service of its kind, since it uses a completely private, serverside RAG with easy access, and it allows users to upload new knowledge and share it with others.

Using the Arcana-Manager, allows the user to upload many pages and multiple files of documentation, to create a so-called “Arcana”, which is indexed and can be accessed via the updated Chat-AI page, not only by the creator, but also with others if needed. An Arcana has an ID and a cryptographic Key, which can be shared or an access link can be created, which contains both information.

Using this, Models in Chat AI can use Retrieval-Augmented Generation (RAG) for creating the answers, which contain better and more curated information close to the request. The relevant information are selected from the Arcana and forwarded into the model together with the request. Only the models that have the option to specify the ID and Key have the possibility to access the Arcana.

In contrary to other services, the RAG implementation is on the serversied and fully private. No documents and request are saved in simple form, similarly to Chat AI. Once preprocessing is completed, the Arcana is available and secured with a newly generated hey. This key can, together with the Arcana-ID, be sent to other users so that the Arcana can be shared.

Try out Arcana for your own courses, lecture material, and documentation. This first version supports Text, Markdown and PDF files. More formats might be supported in the future. An open-source version of the code will also be published in the future.

More information

• Check out the user documentation for the Arcana Framework.
• Try out Arcana based on the HPC documentation.
• Use the RAG-manager interface to start working with an Arcana.