feat(runtime): run sessions in Cloud Hypervisor VMs behind a setting

cmd/aether-server/main.go

@@ -26,6 +26,7 @@ import (
 	"github.com/3xDevOps/Aether/internal/guest/helper"
 	"github.com/3xDevOps/Aether/internal/harness"
 	"github.com/3xDevOps/Aether/internal/runtime"
+	"github.com/3xDevOps/Aether/internal/runtime/vm"
 	"github.com/3xDevOps/Aether/internal/scheduler"
 	"github.com/3xDevOps/Aether/internal/server"
 	"github.com/3xDevOps/Aether/internal/serversetup"
@@ -50,6 +51,9 @@ func main() {
 	if len(os.Args) > 1 && os.Args[1] == "guest" {
 		os.Exit(helper.Main())
 	}
+	if len(os.Args) > 1 && os.Args[1] == vm.MonitorCommand {
+		exitOn(vm.Monitor(os.Args[2:]))
+	}
 	if filepath.Base(os.Args[0]) == "aether-internal" {
 		os.Exit(coordcli.Main(os.Args[1:]))
 	}
@@ -251,6 +255,8 @@ type serveOptions struct {
 	runMemory            *runCountValue
 	runPids              *runCountValue
 	containerRuntime     *string
+	sessionRuntime       *string
+	vmKernel             *string
 }
 
 // serveFlags declares the server options on fs. It is the single definition
@@ -312,6 +318,10 @@ func serveFlags(fs *flag.FlagSet) *serveOptions {
 	o.containerRuntime = fs.String("container-runtime", server.DefaultContainerRuntime,
 		"OCI runtime for run and member environment containers: auto uses sysbox-runc when Docker has it registered "+
 			"(Docker inside sessions without --privileged), otherwise the daemon default; any other value is used as given, e.g. runc to opt out")
+	o.sessionRuntime = fs.String("session-runtime", server.SessionRuntimeDocker,
+		"where sessions run: docker, or vm for one Cloud Hypervisor VM per session (needs /dev/kvm, cloud-hypervisor and virtiofsd; "+
+			"every other container stays on Docker)")
+	o.vmKernel = fs.String("vm-kernel", server.DefaultVMKernel, "the guest kernel session VMs boot with -session-runtime vm")
 	return o
 }
 
@@ -393,6 +403,8 @@ func serve(args []string) error {
 		RunPidsLimit:     int64(*o.runPids),
 
 		ContainerRuntime: *o.containerRuntime,
+		SessionRuntime:   *o.sessionRuntime,
+		VMKernel:         *o.vmKernel,
 		SysboxInstaller:  true,
 	})
 	if err != nil {

docs/environments.md

@@ -314,6 +314,46 @@ pressure](failure-handling.md#recover-from-disk-pressure) and [Recover from
 memory pressure](failure-handling.md#recover-from-memory-pressure) have the
 thresholds and what to do.
 
+## Sessions in virtual machines
+
+`--session-runtime` (config key `session-runtime`) defaults to `docker`. With
+`vm`, each new session runs in its own [Cloud
+Hypervisor](https://github.com/cloud-hypervisor/cloud-hypervisor) virtual
+machine instead of a container. Environment terminals, the assistant, the
+agent updater, the development browser and **Save environment** stay on
+Docker.
+This is work in progress; [the plan](plans/2026-11-13-microvm-sessions.md)
+lists what is still to come.
+
+```sh
+sudo aether-server config set session-runtime vm
+sudo systemctl restart aether-server
+```
+
+The host needs `/dev/kvm`, `cloud-hypervisor`, `/usr/libexec/virtiofsd`,
+`qemu-img`, `mkfs.ext4`, `iptables` and systemd, besides Docker, and the
+release's guest kernel `aether-vmlinux-x86_64` at `/usr/local/lib/aether/vmlinux`
+(`--vm-kernel`, config key `vm-kernel`, names another path).
+
+What changes:
+
+- Each VM runs in the systemd unit `aether-vm-<run id>.service` in
+  `aether.slice`, so it keeps running through a server restart or update.
+  Its disks and logs are in `<data dir>/vm/<run id>/`: `monitor.log` holds the
+  VM monitor's output and `console.log` the guest kernel's.
+- The VM gets the CPU limit rounded up to whole vCPUs, held to the exact
+  limit by the unit's `CPUQuota`, and the memory limit as its memory. The
+  process limit applies inside the guest.
+- The checkout, the home and the other mounts reach the VM over one virtio-fs
+  share; read-only mounts stay read-only. Each VM has its own address in
+  `10.200.0.0/16` and reaches the internet through the host.
+- Starting a stopped session builds a fresh root disk from its image and runs
+  the setup script again: packages installed outside the mounts are gone.
+  Docker images inside the session stay on its data disk.
+- Not yet supported, refused at launch with an error naming the work item:
+  the Isolated tier and egress control, and the development browser beside a
+  session.
+
 ## After a session closes
 
 Reporting success or failure leaves an interactive session open for follow-up, and