Self-hosting Next.js 16 on a VPS with Caddy and systemd
How I deployed my portfolio to a VPS that already ran another app: Next.js standalone output, a systemd service, a Caddy site with automatic HTTPS, and a one-command deploy that rolls back on failure.
Asadullah Afzal
· 2 min read
My portfolio runs on my own VPS rather than a managed platform. It costs nothing extra, I control everything, and setting it up taught me more than clicking "Deploy" would have.
There was one twist: the server wasn't empty. Caddy already held ports 80 and 443 for another app of mine, a scraper dashboard. Here's how I fitted a Next.js 16 site in next to it without touching it.
Look before you install
My first plan was Docker + Nginx + Certbot. Before running anything, I inspected the server:
ss -ltnp # what's listening on which ports?
systemctl list-units --type=service --state=running
cat /etc/caddy/CaddyfileCaddy was serving the dashboard on 443. Installing Nginx would have fought it for the same ports and taken the dashboard down. So instead of bringing a new stack, I matched the existing one: a systemd service behind Caddy.
1. Build a standalone server
Next.js can bundle just what production needs into .next/standalone, with no node_modules install on the server:
// next.config.ts
const nextConfig: NextConfig = {
output: "standalone",
poweredByHeader: false,
};The one gotcha: the standalone server doesn't include public/ or .next/static, so they have to be copied in:
cp -a .next/standalone release
cp -a .next/static release/.next/static
cp -a public release/public2. Run it as a systemd service
The site runs as its own unprivileged user, listening only on localhost, so the outside world can only reach it through Caddy:
[Service]
User=portfolio
WorkingDirectory=/home/portfolio/app
Environment=NODE_ENV=production PORT=3000 HOSTNAME=127.0.0.1
EnvironmentFile=-/etc/portfolio.env
ExecStart=/usr/bin/node server.js
Restart=always
NoNewPrivileges=true
ProtectSystem=fullRestart=always brings it back after a crash or reboot. Secrets (like the contact form's webhook URL) live in /etc/portfolio.env, readable by root only and never committed to git.
3. Add one Caddy site
Caddy gets HTTPS certificates from Let's Encrypt and renews them automatically, so the whole web-server config is a few lines:
www.asadullahafzal.com {
redir https://asadullahafzal.com{uri} permanent
}
asadullahafzal.com {
encode zstd gzip
@static path /_next/static/*
header @static Cache-Control "public, max-age=31536000, immutable"
reverse_proxy 127.0.0.1:3000
}Because the other app was live, I treated this edit carefully: back up the Caddyfile, validate before reloading, and restore the backup automatically if validation fails:
cp /etc/caddy/Caddyfile /etc/caddy/Caddyfile.bak-$(date +%Y%m%d-%H%M%S)
caddy validate --config /etc/caddy/Caddyfile --adapter caddyfile && systemctl reload caddyI recorded the dashboard's response before the change and checked it again after. Identical.
4. One-command deploys with rollback
Updates are a single script: pull, install, build, then swap releases:
mv "$LIVE" "$PREVIOUS"
mv "$STAGING" "$LIVE"
systemctl restart portfolio
for _ in $(seq 1 30); do
curl -fsS -o /dev/null http://127.0.0.1:3000/ 2>/dev/null && exit 0
sleep 1
done
# Didn't come up: put the previous release back
mv "$PREVIOUS" "$LIVE" && systemctl restart portfolioThe build happens before the swap, so the old version keeps serving while the new one compiles, and a bad release rolls itself back.
A small but important detail: the script updates itself with git pull while running. Bash reads scripts incrementally, so I wrapped everything in a main() function called on the last line. Bash parses the whole function before executing it, so the pull can't change the code mid-run.
5. Line endings will bite you on Windows
I develop on Windows. Git can check shell scripts out with CRLF line endings, and then they fail on Linux with baffling errors. A .gitattributes file prevents it:
*.sh text eol=lf
*.service text eol=lf
*.caddy text eol=lfThe result
- HTTPS with automatic renewal,
wwwredirecting to the bare domain - The app answers in about 10 ms on the server
- Deploys are one command, and a broken release rolls back on its own
- The existing app never noticed a thing
The full setup is in the portfolio repository under deploy/.

Written by Asadullah Afzal
Full-Stack Engineer · SEO Expert · AI Educator. I build products that ship, rank, and scale.