you don't want a future where your openclaw assistant -- or the API it is using to think -- can simply turn over your messages and totality of your life. look for local AI inference inside confidential VMs