What they did
Hafiz, a Laravel developer, moved his projects onto a single €5 a month VPS and ran Claude Code inside a tmux session for each one, attaching from a Mac or an iPhone. Staging sites run on the box behind a Cloudflare Tunnel with an Access login, and production goes out through a deploy.sh per project. Each project's CLAUDE.md tells the agent to pull when a session starts and push after every commit, and the deploy script refuses to run with unpushed commits. He compared the setup with Taylor Otwell's plan to work only from Laravel Cloud and Amp's cloud machines.
How it went
After 52 days without a reboot the box used 1.3 GB of 3.9 GB of RAM with two Claude Code instances open, each at about 330 to 355 MB, and the load average sat near zero. Daily use turned up a list of failures: Livewire buttons silently broken behind the tunnel, a queue worker holding stale config, a deploy that deleted a production credential, an expired Tailscale key, Reddit blocking the host's IP range, and Playwright missing its browser path inside the agent's shell. He is not fully cloud-only; some work stays on the Mac.
Worth knowing
The €5 covers development only; production runs on servers he already paid for. A persistent box means isolating projects by hand: he gave one bot its own system user, PHP-FPM pool and queue unit. Watch for kernel updates that never get booted and Tailscale's 180-day key expiry on servers.