Getting started

Placeholder page — the scaffolding is real, the prose is a stub. Replace with actual docs before deploying.

Install

npm i -g hotcell              # CLI + daemon + TypeScript SDK, one install

First run

hotcell                       # first run: 30-second guided setup — then your live fleet

Bare hotcell in a terminal opens the interactive fleet monitor; the very first run walks you through daemon setup. For scripts and CI, hotcell start starts the daemon in the background and returns your terminal — nothing ever prompts outside a TTY. hotcelld is the same daemon run in the foreground.

Add a provider key

hotcell keys add openrouter   # stored on the host, never in a sandbox

Run a contained agent

hotcell run --egress --setup "npm i -g opencode-ai" "opencode run 'fix the failing tests'"

One shot: create → run → destroy. The sandbox gets a revocable egress token with a hard spend cap, and reaches its model only through the gateway — the provider key never enters the sandbox. Egress lockdown is kernel-enforced on microVMs (no NIC) and Linux containers, advisory on the microVM-NIC and macOS-Docker paths.

A fleet on one repo

hotcell create -n 5 --name feat --branch --egress \
    --repo https://github.com/you/app --setup "npm i -g opencode-ai"
hotcell terminal <id>         # inside: cd app && opencode
hotcell rm --all              # done — 5 cells gone, your repo untouched

5 isolated cells, each with the repo cloned, its own branch (feat-1 … feat-5), and your agent preinstalled. With --egress, each cell’s git origin is wired through the gateway automatically — git push works keylessly out of the box, and your GitHub key stays on the host.

Check the wiring

hotcell run --egress "printenv OPENROUTER_BASE_URL"   # egress wired — code reaches the model via the gateway, no key inside

Watch the fleet

Bare hotcell opens the full-screen fleet monitor — attach, pause, live cost. (hotcell tui is the explicit form; hotcell top is its alias.)