HX AI's autonomous task layer. Risky actions require approval.
Dispatch is sandboxed. Destructive commands and network egress are denied unless an approval gate explicitly grants them.