Insight

My remote agent dev setup (phone + old PC + laptop)

2026-01-249 min read

A practical setup guide for running Codex, OpenCode, and Claude Code across devices with Tailscale, tmux, and custom scripts.

What this setup is for

I want to work from anywhere and still run real builds, tests, and agent tasks safely on hardware I control.

The phone is for quick prompts and task handoffs. The old PC is the always-on workhorse. The laptop is for focused work and reviews.

Hardware layout

Old PC stays online at home and runs long jobs. It is the default place for heavy tasks like builds, indexing, or data prep.

Laptop is for active coding sessions and reviews. It connects to the same repo and can take over any job.

Phone is for quick prompts, status checks, and kicking off tasks when I am away from a desk.

Network glue with Tailscale

All machines are on the same Tailscale network so I can SSH to the old PC from anywhere, including my phone.

I keep a short host alias list so I can jump in fast without thinking about IPs.

tmux as the async job runner

Every long task runs inside tmux on the old PC. Each repo gets a named session and each task gets its own window.

This lets me start a job from the phone, close the app, and check progress later from any device.

Agent stack: Codex, OpenCode, Claude Code

I run Codex in the repo for file edits, tests, and structured refactors. It is my default for hands-on work.

OpenCode handles quick codebase questions and small changes when I want a lighter loop.

Claude Code is my backup for alternative plans or second opinions on a risky change.

Custom scripts that keep it fast

I keep a scripts folder with helpers like: open a tmux session, sync the repo, run tests, and start an agent task with a standard prompt.

One script creates a new tmux window, pulls the latest branch, then opens Codex with a short spec template.

A simple daily flow

From my phone, I SSH into the old PC, create a tmux window, and start a Codex task with a short prompt.

From the laptop, I review the diff, run tests locally if needed, and push the final change.

Safety and guardrails

All agent tasks run in a clean working tree. I keep separate branches for each task to avoid messy merges.

I only allow write access where needed, and I always review diffs before pushing.

Quick start checklist

Install Tailscale on all devices and confirm SSH access to the old PC.

Set up tmux sessions per repo and name them clearly.

Install Codex, OpenCode, and Claude Code on the machines where you will run them.

Create a small scripts folder for repeatable tasks and keep prompts short and specific.

More insights

View all

Need hands-on support with software consulting, design systems, or AI workflows?