A Sparkbox is a persistent Firecracker microVM built for development and agentic work.
Your VM disk persists across disconnects, pauses, and cold starts. Treat it like a small server: keep projects in your home directory and commit important work to source control.
CPU and memory are drawn from your owner pool rather than reserved independently for every VM. Idle guest memory can be ballooned down and returned when the VM becomes active. Turbo temporarily increases a VM's working allocation, subject to the owner's burst ceiling and node availability.
Run sparkbox pin inside a VM before starting a server, daemon, or long-running job that must remain continuously available. A pinned VM is not idle-paused and is protected from memory-pressure reclamation. Run sparkbox unpin when the work finishes; sparkbox status reports the current state.
Applications listening in a VM can be reached through Sparkbox's authenticated wildcard HTTPS edge. Use sparkbox set-port PORT to change the default forwarded port, and sparkbox make-public or sparkbox make-private to change access. See the proxy guide.
New VMs include current versions of Claude Code, Codex, Pi, and Hivemind. Sparkbox environment guidance is installed at ~/.agents/AGENTS.md and linked into each harness's global instruction location. Repository-level instructions still apply normally.
Only documented commands and URLs are public interfaces. Local metadata endpoints, gateway ports, node services, and files outside your home directory may change as Sparkbox evolves.