Two weeks ago I launched ClawHosters, a managed hosting platform for OpenClaw (open-source AI agent framework). 50 paying customers so far. Here's the technical post-mortem.
The stack: Hetzner VPS provisioning from snapshots, 5-layer Traefik routing with Redis provider, Nginx reverse proxy that kept silently breaking SSE streams, Docker container isolation with a writable-layer commit strategy, and a config migration registry to handle crash loops at 2 AM.
Some highlights from the article:
- Why `proxy_buffering off` alone doesn't fix SSE through Nginx (you need a re-framing buffer)
- Docker's `cp` command silently refuses symlinks — cost me a day
- ZeroTier + `nsenter` for injecting routes into running containers without restart
- Snapshot-based provisioning: ~3 min cold, ~30s with pre-warmed pool
- Per-request LLM billing with included token allowances per tier
Built as a solo developer alongside a full-time job. The whole thing runs in a single Rails monolith.
The "what value is it producing" question keeps coming up. I'll share my use case: I run https://ClawHosters.com, managed hosting for OpenClaw. Built it because I kept setting up instances for friends and getting support texts at 11pm on Saturdays.
The value isn't the meme projects. It's the "n8n but you talk to it" angle someone mentioned above. Small business automation for people who know what they want but can't code it.
The setup friction is real though. Docker, API keys, channel auth, gateway config. That's the actual barrier to adoption, not the underlying tech. Most people who try OpenClaw bounce off the install, not the functionality.
Re: the foundation move - this is actually good for the ecosystem. MIT license stays, community keeps contributing, Peter gets paid. The alternative was him bleeding $20k/month indefinitely.
The stack: Hetzner VPS provisioning from snapshots, 5-layer Traefik routing with Redis provider, Nginx reverse proxy that kept silently breaking SSE streams, Docker container isolation with a writable-layer commit strategy, and a config migration registry to handle crash loops at 2 AM.
Some highlights from the article:
- Why `proxy_buffering off` alone doesn't fix SSE through Nginx (you need a re-framing buffer) - Docker's `cp` command silently refuses symlinks — cost me a day - ZeroTier + `nsenter` for injecting routes into running containers without restart - Snapshot-based provisioning: ~3 min cold, ~30s with pre-warmed pool - Per-request LLM billing with included token allowances per tier
Built as a solo developer alongside a full-time job. The whole thing runs in a single Rails monolith.