The agent needs swap space files

why do AI agents and LLMs not have swap space by default? why do we insist that they keep massive context windows in active memory? this isn’t a new concept - machines have used paging space for a very long time.

i regularly have my agent go into a paging space directory, create a temporary file, dump everything from active memory that it’s just hanging on to but not using. it makes it much less likely it’ll start becoming confused.