NorwAI has been approached by several public organizations and private enterprises seeking an alternative to international models. 

These entities have primarily raised two concerns regarding existing commercial models: 
(i)    handling sensitive and copyrighted data 
(ii)    the lack of quality in Norwegian language generation. 

This suggests that Norwegian language models should be made available in various ways: 

• General, centralized Norwegian Language Models administered as part of a national infrastructure.

Some desire a secure Norwegian model that can produce better and more reliable Norwegian texts. These models should be accessible via APIs and adhere to strict Norwegian standards. 

• Task- or domain-specific models (fine-tuning)

Some inquire whether NorwAI can customize models for specific purposes or tasks. Others prefer to make these adaptations themselves and run the models locally without sharing their own data. 

• Models combined with proprietary data.

The technique known as Retrieval Augmented Generation (RAG) enables the extraction of relevant data from local entreprise systems and combines it with the language model to generate accurate responses. Local data is often sensitive, not used for model training, and typically more up-to-date and specific than the data in training sets.