OSDev.org

Well I think you explained yourself more clearly and I think I understand what it is you which to accomplish, however, my initial assertion that you write a hypervisor instead of a kernel still stands.

Well as I understand a hypervisor it would still need a kernel to run atop it to control the programs run on it. Why virtualize a second machine? I'm extending the main machine a middle man is just something else to run.

The agent platform eliminates that need by integrating the few features it needs from the middleman and the majority of its features coming from the kernel. Really and truly this could be done on a single kernel, extend the kernel to allow for multiplicty of parts, but this removes any special programming on the devopers part, seeing as the user can specify agent usage, unless they want to init an agent automatically.

BTW:
Anyone thinking of implementing this I would gladly help. Just seeing this in action is gratifying enough. I dont need to be the big programmer behind it (credit for my brain [not me, just my brain] would be nice though).

What you're describing sounds a lot like a more detailed and homogeneous version of Xen, which is a hypervisor. Like your agent model or recursive os or whatever, the original versions of Xen could only run kernels that were specially written for Xen. Basically, the kernels became like a sub kernel to the main Xen hypervisor.