@
The chat window of the built-in assistant answers strictly one step at a time,
so every question feels considerably slower than working with a comparable
assistant elsewhere. Nothing appears on screen until a whole step is finished,
and while the assistant is busy the window itself does not respond.

Wanted:

• The answer appears while it is being written, instead of all at once when it is done. Also show what the assistant is doing at that moment ("Reading fact
  type Rental...").
• The window stays responsive at all times; the application must never appear frozen while an answer is being prepared.
• When the assistant needs several pieces of information for one answer, it gathers them at the same time rather than one after the other. Only
  operations that change the model are done in sequence.
• Information that is nearly always needed (the model summary, the list of fact types) is already available when the question is asked, so it does not
  cost an extra round trip.
• Repeated questions in the same conversation start answering faster, because the unchanged part of the conversation does not have to be processed again.

Additionally, a question that currently needs four or five separate lookups
should be answerable in one: the assistant should be able to ask for several
elements at once, and to request a whole related part of the model in a single
step.
@