Let an assistant work out what is wrong with a computer, without giving it the keys.

Cortt puts a small agent on a device. An assistant asks it questions, gathers the answers, and says what it thinks is wrong. It can only ask for what you have already allowed, and it cannot change anything on its own.