Skip to content

Runtime Catalog

A DevBox is the machine an agent works in — a full container (or Windows VM) with a coding agent, a real shell, your languages, and an optional graphical desktop. This page is the canonical reference for what’s in a DevBox and what you can add.

Every DevBox ships with an AI coding agent. ClusterCode supports three engines today — Claude Code, Codex, and GitHub Copilot — with more coming.

Add the languages you work in to a DevBox — Python, Go, .NET, Rust, Java, Ruby, PHP, and more — so the agent has the toolchain it needs without writing a Dockerfile. Select them when you configure the image; they’re installed into the DevBox and ready to use.

Every DevBox comes with a real, comfortable shell out of the box:

  • GitHub CLI (gh)
  • Neovim (with NvChad)
  • Oh My Posh (shell prompt)
  • Nerd Font (terminal icons)
  • Shell config — sane history, readline, and tab completion

Infrastructure dependencies are always present:

  • Git, Node.js, tmux, curl

You can open the live terminal on any DevBox and drive it yourself — start an agent, watch one work, jump in mid-task, or pair with it line by line.

DevBoxes can include a real graphical desktop, with VS Code and a Chromium that already has the Claude extension wired in. With the Display capability on, an agent can open your running app, see the rendered UI, and click through it.

See Visual testing for the full capability, and note it works on Windows DevBoxes too.

Each agent authenticates with its own credentials:

AgentRequired credential
Claude CodeAnthropic API key, Claude OAuth token, or platform credits
GitHub CopilotCOPILOT_GITHUB_TOKEN + GH_TOKEN
CodexOPENAI_API_KEY

Credentials are configured in the orchestrator. See Environment Variables.

On Pro and Enterprise plans, you can build your own DevBox images to:

  • Add project-specific dependencies
  • Pre-install language runtimes
  • Include proprietary tools
  • Configure a custom shell environment

See Custom Containerfile. You can also snapshot a configured DevBox into a reusable golden image so future launches start from exactly the state you set up.

For the full per-agent capability matrix, see Supported Agents.