Skip to content

What is hopbox

hopbox is a compute-box substrate: a single daemon (hopboxd) that turns one host into a fleet of isolated boxes you reach over plain SSH — for humans and for AIs.

sh
ssh box@box.hopbox.dev

That one command spawns a Firecracker microVM and drops you into a root shell. No signup, no client to install, no pre-created box — your SSH key is your identity and the username is the box spec. It's live at box.hopbox.dev — try it.

The same fleet a person reaches over SSH, an AI drives over MCP — so hopbox is a substrate you can hand work to, not just a place to log in. Delegate a task to an agent box and it runs a coding agent in its own VM, does the work, and asks you when it's unsure.

Ways in, one engine

hopbox exposes the same box fleet through several front doors:

  • ssh <name>@host — spawn/attach a box and get a shell. scp/sftp/rsync and ssh <name>@host "cmd" all work. → Boxes over SSH
  • ssh cli@host — a zero-install management CLI over SSH: ls your boxes, rm one, all key-authed. Nothing to install; your key is your login. → ssh cli
  • the HTTP API — the same fleet as data: https://box.hopbox.dev/v1 with an hbx_ key, including the whole MCP plane over a WebSocket. → HTTP API
  • the web console — watch your fleet and agents live, answer their questions, in a browser: box.hopbox.dev/app.
  • the MCP plane — an AI-control plane: an AI subscribes to the live fleet, delegates work to agent boxes, fields the questions they raise (ask), and even renders interactive UIs a human drives (the canvas loop). → The AI-control plane

How it works

   human (ssh) · AI (MCP)

   ┌──────┴───────┐      reverse dial      ┌──────────────┐
   │   hopboxd    │  ◀───────────────────  │     box      │
   │  the daemon  │                        │   + agent    │
   └──────────────┘                        └──────────────┘
    front door · agent hub · metadata · fleet (store + reconciler) · MCP
  • Boxes dial out. The agent inside each box opens a connection to hopboxd and keeps it. The control plane never routes into a box — so boxes work behind NAT, and your SSH session is proxied over that one reverse connection into the box's own SSH server.
  • Declarative core. A reconciler drives each box from its observed state toward its desired spec — the Kubernetes controller pattern, no Kubernetes dependency.
  • Two compute backends. Firecracker microVMs (hardware isolation, snapshot suspend/resume) or Docker containers (zero-setup). Pick per daemon with --compute.

Next

Status

hopbox is at v0.1. The box model and SSH access are stable; the AI-control plane (agent boxes, ask, the MCP-over-WebSocket API) and the web console are live and evolving.

Instant isolated compute — for humans and AIs