Hopefully, an answer will be forthcoming to your inquiries...
In the meantime, keep digging.
Hal was previously Ultra Hal Assistant and was capable of many cool things back in the day. While many of these earlier Plugins are still to be found, they may or may not necessarily work with our most current version of HAL.
Hal is NOT able to learn from the GPT-3 Davinci or Curie models but it can produce some engaging and thought-provoking conversations.
The individual brains are stored in the <filename.db> files.
One can explore or edit them with an editor like SQLitestudio, which is FREE.
It takes a bit to learn how to use it correctly but always make a copy or backup BEFORE messing with Hal's brains.
If you're just using HAL's Default Brain then the Brain Editor within Hal should suffice.
Everything is now placed into various Tables instead of individual small files like in years past.
The SLIDER thing determines how much or little HAL will use any learned information. Leaving it just a little bit to the right of center seems to work well for a lot of us. Some like around 30 - 35 for good results.
Word is that a Mobile version for HAL is being considered/worked on but we've no official status report on that as yet.
Hopefully, Robert will weigh in on a few of your concerns.