Claude Code and Codex, with a real terminal around them.
Chat is for conversation. A sandbox gives an engineer a full terminal — git, build tools, the filesystem — with an interactive coding agent running inside it, on infrastructure you control.
Claude Code in a browser tab, ready to work
Start a sandbox and the coding agent is already there — in the image your platform team chose, with the tools installed and the credentials it needs. Nothing to set up locally, and no API key on anyone's laptop.
It runs under the same environment and egress rules as your unattended agents, so an engineer experimenting is inside the rules — not around them.
A real machine around the agent
The agent sits in a terminal on a real machine — filesystem, git, build tools. Run the tests it just wrote. Start the dev server it just fixed. Check out a branch and let it continue there.
Everything a coding agent can do, you can watch, interrupt, and steer.
Work from the browser, or from your terminal
The browser tab is the quick way in, not the only one — the CLI attaches the very same session from the terminal you already live in.
Either way, the session runs on the platform: close the laptop at six, reattach in the morning, and the refactor finished without you. Stopping keeps the workspace.
Your editor, their filesystem
Two-way sync keeps a local directory and the sandbox reconciled — you edit in the editor you already have, while the code runs where the credentials are.
Expose a port, open it in your browser
A dev server inside the sandbox is one click from your browser — expose the port and the app hands you the URL.
It lives on your platform's private network: any enrolled device can open it — handy for showing a teammate — and the public internet cannot.
See it on your own cluster.
Thirty minutes, your infrastructure, your stack. Or skip the call — it is one Helm release onto a cluster you already run.