gh, nginx, and the usual tools, reachable over SSH and at https://name.boxd.sh. Typical boot time is ~50ms.
How it works
boxd new provisions a KVM microVM on a worker, attaches a fresh 100 GB disk, brings up the network stack, and registers a default proxy that forwards https://name.boxd.sh to port 8000. TLS is terminated for you, HTTP redirects to HTTPS, HSTS is set, and WebSockets pass through.
The image is real Ubuntu 24.04 (unminimized) with the runtimes and CLI tools youโd install on a fresh laptop already in place. You land as the boxd user with passwordless sudo.
Pick a name, run the command, start working.
Use it
- SSH
- CLI
- TypeScript
- Python
--json works here too.--name, boxd generates one like blue-river.
Whatโs already installed
Each box ships with a developer-ready Ubuntu 24.04:- Languages: Python 3 (with
uv,pipx), Go,build-essential, Node.js 24 via nvm - Agents: Claude Code (
claude), Codex (codex), pre-wired withAGENTS.mdcontext - Containers: Docker, Docker Compose, Buildx
- Web: nginx pre-installed on port 8000 (start it when you want it), headless Chrome
- Tools:
git,gh,jq,ripgrep,sqlite3,rsync,ffmpeg, ImageMagick,mitmproxy - Editors: vim, neovim, plus full SSH for Cursor/VS Code/Zed/JetBrains
Custom images
Every box boots fromcomputer:latest, a curated Ubuntu 24.04 image we maintain. Same image across the fleet, prewarmed on every worker, which is how fresh boots stay around ~50ms.
Need something different? A different base distro, your own pre-installed toolchain, a specific CUDA stack, a slimmer surface area? Email contact@boxd.sh with what you want baked in. We package it, register it as a template on the cluster, distribute it to every worker, and prewarm it. Your VMs then boot from your image with the same sub-second cold start as the default.
Recipes
Serve something instantly
name.boxd.sh to port 8000. Any HTTP server on 8000 is live.
Change the default port
--port=auto to let boxd detect whichever port your app started listening on.
Stop paying when idle
Boxes auto-suspend after 30s of network idle by default. Resume is sub-millisecond on the next request.FAQ
How is this different from a container?
How is this different from a container?
Each box is a full KVM microVM with its own kernel, network stack, persistent disk, and public IP.
systemd runs as PID 1, Docker works without nesting tricks, and the filesystem survives reboots.What happens to data when I reboot?
What happens to data when I reboot?
The 100 GB disk persists across reboots, suspends, and resumes. Only
destroy removes it.Where is the infrastructure?
Where is the infrastructure?
EU-hosted on sovereign infrastructure.
Next
Fork from a golden
Skip the install step. Warm copies of your app in ~160ms.
Live demos & share URLs
Every box has a real HTTPS URL the moment a port is open.