← Back to Security

The Ultimate Mobile Vibe Coding Setup

security

It only took 10 years…

There is something magical about being able to create code and tinker with software “anywhere”. In 2010, I bought my first netbook and thought, “No way, I can program Arduino with this… I can configure Cisco gear with this… I can do ANYTHING!” The battery life was GREAT for the time (4 hours maybe?), and it could even run World of Warcraft.

Since then, I’ve experimented with a few different ideas on how to “do nerdy things” wherever I am:

  • JupyterHub with some customizations that allowed me to self-host VSCode running in a browser
  • Using VNC on an iPad connected to a remote desktop server
  • Chrome Remote Desktop

These were OK, but I ran into a few common problems:

  • Special characters, keyboard shortcuts, and UX designed for desktop did not translate well to a mobile keyboard
  • Cramming a desktop environment into a netbook screen or Chromebook tab did not work well… it’s just too small

Enter Vibecoding

My home lab exists because I want to actually understand the technology that makes “engineering excellence” look easy.

  • I want to understand how Argo works
  • I want to understand how Helm works
  • I want to understand how k8s works

The great thing about running this at home is that now my agents can too.

  • Agents can create PRs
  • PRs trigger container build jobs
  • Merging a PR updates the deployed version of a container in my Helm charts
  • ArgoCD auto-deploys those changes

And because it all lives on the lab network, the entire loop is reachable from my couch — or anywhere else — with nothing more than an iPad.

The Winning Setup

My new favorite thing has two parts:

  • A dusty old stick PC
  • An iPad

Setup the Linux Machine

This is where we will ACTUALLY be running Claude Code

  1. Install Ubuntu and configure SSH
  2. Configure kubectl so that Claude can debug the cluster
  3. Configure gh so that Claude can check Actions workflows and merge PRs
  4. Configure mosh so that SSH clients are “always” connected
  5. Install Claude Code
  6. Create a custom service that runs claude remote-control at boot

The boot service

I want claude remote-control to come up automatically on boot, survive crashes, and stay reachable even if the SSH/mosh session drops. The trick is to run it inside a detached tmux session, and let systemd babysit that.

First, a small wrapper script that owns the tmux session. Drop this at /usr/local/bin/claude-rc-wrapper.sh and chmod +x it:

#!/usr/bin/env bash
set -euo pipefail

PROJECT_DIR="${HOME}/remote-claude"
SESSION_NAME="claude-rc"
LOG_FILE="${HOME}/.claude/rc-session.log"

mkdir -p "$(dirname "$LOG_FILE")" "$PROJECT_DIR"

# Kill any existing tmux session
tmux kill-session -t "$SESSION_NAME" 2>/dev/null || true

# Start tmux with claude remote-control inside it
tmux new-session -d -s "$SESSION_NAME" -c "$PROJECT_DIR" \
  "claude remote-control --remote-control-session-name-prefix minibox --verbose 2>&1 | tee -a $LOG_FILE"

# Wait for tmux to be alive
sleep 2
if ! tmux has-session -t "$SESSION_NAME" 2>/dev/null; then
  echo "Failed to start tmux session" >&2
  exit 1
fi

# Keep the systemd process alive as long as the tmux session is up.
# When claude exits, the session dies, this loop ends, and systemd
# can apply its Restart policy.
while tmux has-session -t "$SESSION_NAME" 2>/dev/null; do
  sleep 5
done

Then the systemd unit that runs it at boot. Drop this at /etc/systemd/system/claude-rc.service:

[Unit]
Description=Claude Code Remote Control (minibox)
After=network-online.target
Wants=network-online.target

[Service]
Type=simple
User=tommy
Group=tommy
Environment=HOME=/home/tommy
Environment=PATH=/home/tommy/.local/bin:/usr/local/bin:/usr/bin:/bin
ExecStart=/usr/local/bin/claude-rc-wrapper.sh
ExecStop=/usr/bin/tmux kill-session -t claude-rc
Restart=on-failure
RestartSec=10
TimeoutStartSec=60
TimeoutStopSec=15

[Install]
WantedBy=multi-user.target

Enable and start it:

sudo systemctl daemon-reload
sudo systemctl enable --now claude-rc.service

Now claude remote-control is always running, reconnects after crashes, and because it lives in tmux I can tmux attach -t claude-rc over SSH any time I want to peek at what the agent is doing.

Access Anywhere

Combining the above with an iPad and a keyboard case (I like this typecase Edge case), I can now:

  • use the Claude iOS app to start a session anywhere
  • use the Termius iOS app to connect directly to the host over SSH for the things Claude can’t/won’t handle

typecase Edge Keyboard Case for iPad — backlit magic-style keyboard with multi-touch trackpad and floating magnetic stand

In practice this means I can kick off a refactor from the couch, watch the PR build and ArgoCD sync from my phone, and only drop into a real SSH session when something needs hands-on debugging. The trackpad and proper keyboard make the iPad feel close enough to a laptop that I stopped reaching for one — and unlike the netbook days, there’s no cramped desktop crammed into a tiny screen, just a chat session driving a machine that already has all my tools.

Honestly, the setup has been pretty great.