I'd love to learn more please. I'm interested in sandboxing AI tools/agents regardless of the underlying mechanism (I explored Firecracker VMs briefly as well, terrible cross platform support though).