Skip to content
Fuse
Esc
↑↓navigate↵open⌘Jpreview
On this page

Desktop environments

Bake a rootfs that boots to a working X session, so an environment has a screen to drive

Most Fuse environments are headless: a guest agent, a workload, no pixels. A computer-use agent needs the opposite — a display to screenshot, a pointer to move, a browser to open. The desktop image provides that: it boots straight to a running X session with a window manager, a panel, and Firefox installed, inside the same Firecracker isolation every other environment gets.

The display is Xvfb, a software framebuffer. It needs no GPU and no extra virtual devices, so a desktop environment schedules onto any Firecracker host like any other environment.

Baking the image

The desktop is a second image variant, layered on the standard baked rootfs on the host:

cd host-agent/firecracker
./fc-bake-rootfs.sh           # the base image, if you have not baked it yet
./fc-bake-desktop-rootfs.sh   # layers the desktop on top -> rootfs-desktop.ext4
mkdir -p images && cp rootfs-desktop.ext4 images/desktop.ext4

The bake installs the desktop stack (Xvfb, mutter, tint2, xdotool, scrot, xclip, Firefox ESR, pcmanfm) by resolving it in an ubuntu:22.04 container on the host and extracting the result into the image — the guest itself cannot run apt-get. It finishes by booting the display inside a chroot and capturing a screenshot, so a broken bundle fails at bake time rather than at first boot.

What comes up at boot, in order: fuse-display (Xvfb on :1), fuse-wm (mutter), fuse-panel (tint2), then fused. The agent is deliberately ordered last so nothing can reach the environment before the screen it would be driving exists.

Using the image

Name it from a Fusefile like any other host-baked image:

version: 1
image: desktop

resources:
  cpus: 2
  memory: 4GB # a browser session does not fit the 1GB default

The display geometry defaults to 1024x768. It can be changed at bake time with FC_DESKTOP_GEOMETRY=1280x800x24 ./fc-bake-desktop-rootfs.sh.

Sizing and cost

A desktop image lands in the 1.5–2 GB range against the base image’s few hundred MB. That is fine once it is on a host, but it is 2 GB crossing the wire the first time an environment lands somewhere new. Memory matters more: Firefox needs real RAM, so budget 2–4 GB where a headless environment gets by on the 1 GB default.

Was this page helpful?