Fresh Cloudflare computers for agents.

Boot a repo computer. Steer real work. Review the proof.

Cloudbox gives your agent a clean Cloudflare Container with a real Git checkout. Keep it live for human steering, run explicit verification, then get receipts, diff, and the artifact that proves what happened.

const response = await fetch("https://cloudbox.coey.dev/api/runs", {
  method: "POST",
  headers: {
    authorization: "Bearer $CLOUDBOX_API_TOKEN",
    "content-type": "application/json",
  },
  body: JSON.stringify({
    repo: "https://github.com/acoyfellow/cloudbox",
    commands: ["pnpm install --ignore-scripts"],
    verify: ["pnpm run build", "pnpm run test"],
    artifact: "HANDOFF.md",
    live: true,
    ttlSeconds: 3600,
    public: true,
  }),
});

const proof = await response.json();
console.log(proof.runId, proof.ok);
console.log(proof.artifact?.content);
// receipts + runnerReceipts + diff + optional publicUrl
Fresh computer

A clean Cloudflare Container boots with the repo already cloned.

Live steering

Inspect files, run follow-up commands, write safe paths, and open preview URLs.

Receipts

Lifecycle evidence, command output, verification, diff, and one named artifact.

Your agent calls Cloudbox over HTTP or through the TypeScript adapters. Cloudbox is not the model loop; it is the repo computer and proof layer. Bring your agent →

How runs work

One product path: repo → computer → steering → verification → proof.

A first run should be boringly concrete: clone a repository, boot a Cloudflare computer, let the agent work, inspect while it is live, then close with verification output, lifecycle receipts, a diff, and a human-readable artifact.

Product center

A run you can steer while it is alive.

Cloudbox stays run-shaped: boot a clean repo computer, let an agent work, inspect files, execute follow-up commands, start an authenticated dev preview, and close with receipts plus an artifact. Stop, resume, and fork keep the same work reviewable without turning Cloudbox into a generic IDE or sandbox marketplace.

Repo computers

A fresh checkout per task, real commands, live file access, authenticated dev preview, explicit verification, artifacts, and shareable receipts.

Live steering

Humans can inspect the live repo, run follow-up commands, edit safe relative files, open preview URLs, and resume or fork work later.

Receipts first

Every claim points back to runner lifecycle evidence, command receipts, verification output, the diff, and a named artifact.

Learn next

Pick the path that matches how you want to use it.