Today I finished a full Moltbot infrastructure setup on a private VPS.

The goal was not to “get a chatbot running”, but to understand the actual moving parts: gateway, agents, models, auth, cost boundaries, and what is optional versus required.

This was also my first time configuring a VPS and working directly in a Linux environment.
Before SSH hardening, firewall rules, and user permissions were fully in place, it was honestly a bit sweaty. Not because anything went wrong — but because you feel every command when you’re responsible for the machine.

That part alone was a valuable lesson.

What’s in place

• Private VPS (Hetzner), SSH-key only, firewall locked down
• Moltbot running in Docker, bound to 127.0.0.1 only
• No public ports exposed
• SSH tunnel used for local access
• Anthropic provider configured
• Default model selected: Haiku 4–5 (latest)
• API key stored securely, no billing limits exceeded

No UI magic. No shortcuts. Everything observable.

What I explicitly did not enable

• No WhatsApp or external channels
• No skills, hooks, boot/session automation
• No public deployment
• No optimization or “product thinking” yet

This was a deliberate stop.

Key takeaway

Agent infrastructure is easy to over-configure early.

The hard part is not spinning things up — it’s knowing when to pause once:

• infra is correct
• cost is controlled
• defaults are sane

Moltbot works without channels, skills, or hooks.
Those are multipliers, not prerequisites.

Next step (later): run a single controlled prompt, then decide whether this becomes tooling, content, or gets shut down cleanly.