Setting Up a Remote Environment for Agentic Workloads

A remote development box gives coding agents a stable place to run while you disconnect, reconnect, or change devices. We set up an Ubuntu server, protect SSH access, install the development tools agents need, and keep several agent sessions available through tmuxctl. We use AWS examples, but the same setup works on any Ubuntu host that you can reach over SSH.

The machine stays on the other side of SSH while your laptop, VS Code, or phone controls it. tmuxctl keeps terminal sessions alive across disconnects and gives each session a memory boundary, so one busy agent has less chance of affecting the others. Codex, Claude Code, OpenCode, and Grok Build then run in those persistent sessions with the same Git and project files.

We use three controls for the security boundary:

• The provider controls access to the server.
• SSH controls login.
• The host firewall controls network services.

We bind development services to loopback when possible, then use SSH port forwarding to reach them without publishing every port to the internet. If you need public access, add that port deliberately and review the provider security group and host firewall together.

You can also use a phone-based path. PocketShell imports an SSH host through a QR code, attaches to a tmux session, and turns voice dictation into a prompt. Treat the QR payload like a private key and create it only in a private place.

Links

Use this earlier workshop for another remote-server example:

• Serving open models with vLLM on RunPod - an earlier remote-server workshop with SSH and agent setup.