The issue on our side was probably caused by not having enough resources for another model depoyment. Not that any of the error messages pointed in that direction 
After adding another ML node, a model with instance ID "obs_ai_assistant_kb_inference" was made available and we are able to add stuff to the KB.