Skip to content
dx docs

Run Claude Code in the cloud

dx runs the Claude Code CLI in a cloud sandbox, called an Orb, on your own Claude Pro or Max plan. Each conversation is a Thread with its own Orb, so several Claude Code sessions can work at once, and they keep working when you close your laptop. You follow and steer them from the dx web app.

What you need

Section titled “What you need”
  • A Claude Pro or Max plan.
  • Claude Code on your own computer, once, to create a token.
  • An Orb provider: an E2B, Cloudflare, Freestyle, or Docker account you connect, or one your dx deployment provides.

Set it up

Section titled “Set it up”
  1. Connect your plan. In dx, open Personal Settings → Model Routing and choose Add → Claude Pro/Max. On your computer, run claude setup-token, sign in with your Claude account, and paste the token it prints. dx checks the token with Anthropic before saving it.
  2. Connect an Orb provider. In Personal Settings → Orb Providers, choose Connect next to a provider and enter its key. dx prepares your account in a few minutes. See Connecting an Orb provider.
  3. Start a Thread. Open a project, choose New Thread, and pick the Claude Code row in the mode picker. Choose a model, such as Claude Opus 5.5 (the default), and an effort, then send your task.

Connecting a model covers the token, models, and plan limits in detail.

How Claude Code runs in the Orb

Section titled “How Claude Code runs in the Orb”
  • In your repository. The Orb clones the project into ~/workspace/repo, and Claude Code starts there. It reads the repository’s CLAUDE.md, or its AGENTS.md when there is no CLAUDE.md.
  • Your instructions. What you write in Personal Settings → Instructions becomes Claude Code’s ~/.claude/CLAUDE.md in the Orb.
  • Installed when needed. dx installs the Claude Code CLI when the Orb lacks it or has a version older than 2.1.310.
  • One session per Thread. Your first message starts a Claude Code session, and later messages resume it.
  • No permission prompts. The Orb is a sandbox, so Claude Code runs with permission checks bypassed and works without stopping to ask.
  • Awake while it works. The Orb stays awake while Claude Code runs. By default it pauses after five minutes without activity and resumes on your next message.

What you can do from dx

Section titled “What you can do from dx”

Every tool call Claude Code makes shows up in the Thread’s transcript. Next to it, the Thread has Files, Changes, a Terminal you share with the agent, a Linux Desktop with Chromium, and Portal links to web servers running in the Orb.

Claude Code also gets dx’s own tools through an MCP server named dx: it can start child Threads, read other Threads, set a schedule, and subscribe to GitHub events. Claude Code’s own web search and page reading replace dx’s versions of those tools.

Plan limits

Section titled “Plan limits”

Claude Code in dx counts against your Claude plan’s limits. Personal Settings → Model Routing shows your 5-hour, weekly, and per-model limits. By default Claude Code stops at a limit. When your Claude account has extra usage turned on, choose Allow Extra Usage to keep it working on paid extra usage.

Source

Section titled “Source”