launch 8 Blender sessions without starving your laptop.

headless Blender in the cloud for your agents. every session gets its own sandbox, so none of them uses your CPU or memory.

the default cube

connect your agent.

sign in with GitHub, then you and your agent drive sessions from the terminal.

pipx install "git+https://github.com/rehan-remade/defaultcube#subdirectory=oss/sdk"
cube login
cube create -n 8 --name demo
cube render ses_... -o monkey.png --engine CYCLES

your agent can call the same /v1 routes that the CLI uses.

curl https://api.defaultcube.dev/v1/sessions \
  -H "Authorization: Bearer $CUBE_API_KEY" \
  --json '{"name": "demo"}'

how it works.

your agent asks for a session, and defaultcube boots headless Blender in a fresh sandbox. the scene and Python globals persist between calls.
cube create -n 8 starts 8 sessions at once. each one is its own Blender process, so agents never share a scene.
cube render downloads the image, so the model checks its own work. with --gpu, a 640×480 Cycles render takes 6.5 to 9 s on a warm RTX PRO 6000, 45 to 70 s cold.
after 15 idle minutes, a session pauses and keeps its memory and undo. the next call resumes it in under a second.
an idle session holds about 370 MB of memory on our machines. your laptop holds none of it.

open source.

Apache-2.0. self-host it with your own E2B and fal keys. GitHub

make install
make template
CUBE_BACKEND=e2b make serve