Instant Firecracker microVMs — a terminal, a real browser, coding agents preinstalled, and an AI that drives them. Open source. Fork it and run your own with your own keys.
Headless microVMs with a serial console + an AI that drives them. python3 · node · claude, milliseconds to boot, opt-in internet.
Full Linux desktops over VNC — a real browser plus claude, codex, cursor & pi preinstalled.
An AI that sees the screen and drives the mouse & keyboard.
One OpenAI-compatible endpoint for every model — Claude on Anthropic, the rest via OpenRouter.
Persistent volumes that outlive the machine — save a computer, restore it into a fresh one.
Real hardware-virtualized isolation — a kernel per machine, not a shared container.
Each VM runs unprivileged and chrooted, with per-machine CPU, memory and PID limits.
A shell resumes from a memory snapshot in ~3 ms. No cold boot, no waiting.
Machines clean themselves up on a TTL — nothing lingers. Or flip “keep alive” and one runs until you stop it.
It's open source (Apache-2.0) and self-hosted — bring your own Anthropic + S3 keys, point
it at a Linux box with /dev/kvm (x86_64 or arm64), or run the
whole thing on an Apple Silicon Mac with one command. Nothing phones home.