boxes.dev

Developers running AI coding agents like Claude Code or Codex today are tied to their laptop: agents run as local worktrees on the same machine, competing for resources, ports, and database state, and work stops the moment the laptop closes. Running multiple agents in parallel means juggling ports and risking one agent's changes (dependencies, schema edits) breaking another's environment. There's also no easy way to kick off or check on an agent from a phone, or to start work directly from tools like Slack or Linear without being at a keyboard. The site argues existing alternatives (Devin, Cursor, raw Claude Code/Codex) only partially solve this, with limited mobile support, partial Slack/Linear integration, or no full isolated cloud computer per agent.

boxes.dev provisions a full cloud Linux computer (a 'devbox') per agent, pre-loaded with the user's actual dev environment, so each piece of work gets its own filesystem, processes, ports, and database rather than a shared worktree. Setup is handled by pasting a prompt into Claude Code or Codex, which inspects the local environment and drafts a plan that a second agent then executes to replicate it on a cloud Template box; a snapshot is taken so new devboxes can spin up from it in seconds. Agents can be started from Slack mentions, assigned Linear issues, cron schedules, or a single API call, and run using the user's own existing Codex or Claude Code subscription rather than a resold token plan. Idle devboxes sleep automatically and are free, while users can manage, review diffs, approve requests, or open a terminal from desktop, mobile, or CLI clients. Team features include shared live session links, public preview URLs for running apps, SSO/SCIM, RBAC, audit logs, and spend/concurrency limits.