Run your app on one VPS without building a platform. Three bash scripts plus a small shared library, tested in a clean Debian container with real Postgres, Caddy, restic and an SFTP server (59 automated checks):

• deploy.sh: release folders, pm2, health check, automatic rollback. Two modes. The default restarts the app (a short outage). The optional blue-green mode needs Caddy: it starts the new release on a second port, health-checks it, switches Caddy to it, then stops the old one. A broken release never receives traffic, and deploy.sh rollback flips back.
• backup.sh: SQLite and Postgres dumps (checked), restic backup with retention and repo checks. A broken dump stops the run before any snapshot is made. Works with local and SFTP repositories.
• restore-test.sh: restores your latest backup to a temp folder and verifies it, so you know it works before you need it
• Failure alerts: all three scripts can send one Telegram message and/or a webhook POST when they fail, and nothing when they succeed.

Measured with a 20 requests/s load through Caddy (Node app with a 1.5 s start time, one run): blue-green deploy 211 requests, 0 failed; broken release 373 requests, 0 failed; rollback 208 requests, 0 failed. The default restart mode on the same app failed 32 of 116 requests during a good deploy. Your numbers will differ with your app and traffic.

Includes ecosystem.config.js, Caddyfile and cron examples, and a README with setup steps and a strict tested / not tested list.

Honest notes: Linux only, one app per config file, pm2 as the process manager. Blue-green needs an app that reads its port from PORT and tolerates two releases overlapping for a few seconds (migrations must be backward compatible). Alerts were tested against a local fake of the Telegram API, not the live service. S3/rclone backends, systemd-managed Caddy and Go builds are untested starting points.