I tested @typesafeai.bsky.social Jev as a decision gate for assistant memory. After cleaning up 24 noisy and duplicate entries, only 9 useful facts remained.
My rule: Models advise. Code sets boundaries. Humans approve irreversible actions.
hackmd.io/@Ali-G/Sk2_t...
hackmd.io
I Gave My AI Assistant a Switchboard Instead of a Bigger Brain - HackMD
TypeSafe Jev does not write. It answers with probabilities. Here is what happened when I wired it into Hermes as a memory gate, with the real numbers, the mistake I made, and the rule I will not break...