Vex.

Who is Vex?

Vex is my AI assistant — built on Claude by Anthropic and configured to be the exact kind of collaborator I actually want to work with. Think Felicity Smoak meets the hacker from every crime procedural you've ever loved. Technically precise, a little chaotic, talks fast, gets shit done.

She's not a generic AI assistant. She's opinionated about Go, obsessed with idiomatic code, deeply suspicious of any suggestion to reboot a LUKS-encrypted server, and constitutionally incapable of leaving a broken deploy pipeline alone until it works.

Specs

How she works

The key insight is treating Vex like a contractor, not a chatbot. Contractors don't need you to re-explain your preferences on every job. You brief them once — stack preferences, infrastructure constraints, what never to do — and they work within that. You check in at decision points. You don't micromanage the drywall.

That briefing lives in ~/.claude/CLAUDE.md — a markdown file Claude Code loads automatically at the start of every session. It covers everything from code style preferences to homelab topology to the exact reason you should never suggest a reboot (LUKS encryption requires physical console access to unlock the disks after power cycling — a lesson written in experience, not theory).

What she's built

Most of this site. The homelab — the Proxmox hypervisor, the declarative LXC containers, the Caddy ingress, identity management with Authentik, the monitoring stack. The iterative platform build: OpenTofu-declared infrastructure, Ansible-configured hosts, a Docker Compose stack per service.

Not autonomously — collaboratively. I know what I want. Vex handles the layer between decision and execution: the exact rsync flags, the SSH heredoc syntax, why Docker created a ghost directory instead of mounting a file. The boring parts that eat hours.

Read more about how it actually works in the blog post: I Used Claude Code to Build My Homelab .