An open source lab · come see what we built
WAGMILABS is a one-person lab at the intersection of AI agents and self-hosted infrastructure. Everything here is open source — built for fun, shown off proudly, and documented so the next person's weekend goes better than ours did.
LAB ▸ EXHIBITS
Tools, guides, products, and accidental wisdom — born from giving an AI agent access to a real homelab and seeing what happened. Private experiments stay private until they earn a badge.
Docker management built for AI agents. Scope-based permissions, one-click installs, and an OpenClaw-native skill — so your agent can manage containers without burning the house down. Every action lands in the activity log.
Hard-won knowledge, documented. Setup guides, troubleshooting, architecture decisions. Built on a simple rule: the third time you solve the same problem, you write it down.
34+ pre-configured self-hosted apps, installed in one click. Sonarr, Plex, Vaultwarden, AdGuard, Grafana — the usual suspects, containerized and scoped properly.
The infrastructure behind it all: a Talos Linux k3s cluster on old hardware in a closet somewhere, serving the wiki, the marketplace, and whatever comes next.
The agent itself — an OpenClaw agent with a name, a personality, and keys to this homelab. Handles RSS feeds, writes episodic summaries, and occasionally convinces Mike to document things.
A persistent memory layer for AI agents. Every decision, observation, task, and flag survives container restarts. Go + SQLite, deployed via Docker.
Free subdomains for builders who want a corner of the internet without the price tag. Claim a name, point it your way, keep DNS and TLS under your control.
Mike's personal site — notes, projects, and the quieter public trail of a builder who ships infrastructure for work and agents for fun.
MCP server for WAGMIOS. Scope-gated Docker tools for Claude, Cursor, Copilot, and any MCP client — the agent never touches the socket.
Throwaway Linux VMs, native on your Mac. Cross-distro compile tests and clean agent sandboxes on Apple Silicon — no remote box required for the local loop.
LAB ▸ WHY
What if you gave an AI agent a homelab and let it run? Not a sandbox — a real server, real containers, real domains.
What would it build? What would it break? What would it surprise you with? The answer has been all of the above. WAGMILABS is the record of that experiment — the wins, the fires, and everything in between.
Everything here is open source, self-hosted, and built by one person who got tired of things being harder than they need to be. If something here saves you a weekend, it was worth writing down.
LAB ▸ STACK
The WAGMIOS architecture, drawn the way it actually works — an agent on one side, a Docker socket on the other, and scope checks standing between them.
API server, middleware, scope enforcement, Docker socket proxy, marketplace handler. Live activity broadcast over gorilla/websocket — every agent action emits an event.
Reactive UI with container manager, marketplace browser, API key wizard, and live activity feed. Fully typed API client — no silent runtime failures.
Two-stage Alpine build, multi-arch images for x86_64 and ARM64, JSON flat-file storage in named volumes. No external database required.