Transparent agency over the retreived memories

Being able to control what the platform will add to
the memories and insights
from the conversation.

So that it can be programmed through the cachracter intructions without getting any surprises. 

Currently, that only work reliably with user preferences, but many other information that i ask the system to memorize, like a character’s evolution — I know that memory can’t be updated fast, so I use it for character peristant changes rather than short term status like mood. 

But it’s not always memorized.
And I can’t find documentaiont explaining why.