Switch from your existing agent

Use agents-sandbox with the agent setup you already have without writing any config. If you haven't installed agents-sandbox yet, [install it]({% link index.md %}#install) first.

Run it

In any project directory:

agents-sandbox

agents-sandbox copies your existing agent config (e.g. ~/.config/opencode/**) and credentials into the VM by default, so your normal agent, models, and permissions are available immediately, now running in a hardware-isolated VM with your project at /workspace.

Keep your normal workflow. Because your config is mirrored, you can keep managing it where and how you already do, including managing config via the agents' TUI on the host, and it stays in sync with the sandbox.

⚠️ What to know

The fast path shares your host config and credentials (for opencode, auth.json) into the VM. This is fine for trying it out or low-sensitivity work. When you want no secrets exposed to agents, not even the provider API keys, or you want a self-contained, reproducible setup, see [Manage config in the sandbox]({% link manage-config.md %}) and [Secrets]({% link configuration/secrets.md %}).

Next steps

• [Manage config in the sandbox]({% link manage-config.md %})
• [Secrets]({% link configuration/secrets.md %})
• [Agent configuration]({% link configuration/agent.md %})