Privacy-first deployment use case on Railway, fully with an agent

An AI agent (Claude Code) deploying the same self-hosted stack on Railway, end to end in one session: a private GitLab, a private fork of Supabase pushed into it, self-hosted Supabase deployed from that fork, a real backups feature built into Supabase Studio, and a private VPN. Raw, only dead time cut, every mistake and self-heal left in. This is the Railway run in a four-platform comparison: the same job, same agent, same prompts, on redu.cloud, AWS, Vercel, and Railway. Railway did well here, and it's worth saying so. It asked before spending, consistently, you can watch it stop and confirm through the run, and its usage view is clear about what things cost. The friction on this task was mostly the database and passwords, and keeping everything fully private took tradeoffs the platform doesn't lean toward by default. The full agent transcript for this session is published with every credential removed, so you can read exactly what happened. Chapters: 0:00 Deploy a private GitLab 4:17 Add a private fork to GitLab 5:46 Deploy Supabase 10:32 Integrate a backups feature into Supabase 14:03 Confirm it works, merge to master 17:16 A private VPN (Tailscale) 21:52 Architecture diagram and report 23:17 Cost overview and conclusion The full write-up, the downloadable transcripts, and the same job on redu.cloud, AWS, and Vercel: https://redu.cloud/compare/the-proof redu.cloud is an EU-hosted cloud you operate through an AI agent.