sandboxed.sh

sandboxed.sh is a self-hosted cloud orchestrator for AI coding agents. It is built for long-running coding tasks that can outlast a session, for cases where code must stay on a user’s machine, and for projects that need custom agent behavior rather than generic defaults. The site summarizes its purpose as running AI coding agents on your own server, with the promise of starting a task, closing the laptop, and checking back later.

It connects to a web dashboard, iOS app, or CLI. From there, a user tells an agent what to do, and it continues until the job is done, whether that takes 10 minutes or 10 hours. Each project gets its own isolated container, and sandboxed.sh manages starting and monitoring tasks, streaming progress, managing environments, and syncing configuration from a git repository of skills, commands, and rules. Under the hood, it runs Claude Code, OpenCode, Codex, Gemini, or Grok for the actual agent work. The page also says the system handles the infrastructure around AI rather than doing AI itself.

The tool is described for people who have tasks that take hours, who need code to remain on their own machines, or who want agents that follow team-specific patterns. It is also presented for users running multiple projects with different needs, since container workspaces keep them isolated from one another. Setup is described as taking five minutes if a server and domain are available, and the installation notes mention backend deployment on the user’s server, with dashboard options to host on Vercel, run locally, or use an iOS app. The page mentions a server cost of about $30 per month and says it handles systemd, nginx, and SSL during setup. No license is stated.