Homelab AI Blueprints: A Guided Tour of Nine Self-Hosted, AI-Operated Projects

Homelab AI Blueprints: A Guided Tour of Nine Self-Hosted, AI-Operated Projects

I run a home server I call “the Tower” — a real multi-service environment, increasingly operated not just by me but by a fleet of persistent AI agents, each responsible for one domain. Over time I wrote down the patterns that made that setup actually work, and published them as Homelab AI Blueprints, a public, CC BY 4.0-licensed repo on GitHub.

Each blueprint is self-contained by design — the whole point of splitting them into separate numbered folders is that you can pull exactly the one you want (say, just the password manager, or just the audiobook pipeline) without needing the rest. Each one also ships with two documents covering the same material: a Guide.md written for a human to read and follow, and a paired AI-Brief.md — the same knowledge distilled into something an AI assistant can execute directly, no interpretation required.

Below are the nine blueprints, each with its own full writeup covering what it does, why I built it the way I did, and what was actually hard about it:

1. No Open Ports: Exposing a Home Server Safely with Cloudflare Tunnel — how every other service on the Tower gets a public, secure address without opening a single port on the router.
2. Passwords an AI Agent Can Actually Use: Self-Hosting Vaultwarden — a self-hosted password manager built specifically so a script or an AI agent can pull a credential programmatically, not just so a human has one more app.
3. Building a Fleet of Persistent AI Agents Without an Extra Subscription — how each of the Tower’s AI agents gets a durable identity, memory, and voice using plain files, riding on a subscription I already pay for.
4. Directors, Specialists, and a Shared Knowledge Base: Teaching AI Agents to Work as a Team — the patterns that let a dozen-plus independent agents coordinate without duplicating each other’s knowledge or stepping on each other’s work.
5. Who’s Watching the Watchers: Scheduled Self-Reporting and an Independent Watchdog — how the Tower checks on itself on a schedule, and why a second, physically separate machine is what catches it if the whole thing goes dark.
6. Yours for Life, Actually: A Self-Hosted Audiobook Pipeline for the Whole Family — turning a personal Audible library into a real multi-user audiobook server without sharing an Amazon login.
7. One Library, No Duplicates: Turning Spotify Playlists into a Self-Hosted Music Collection — a pipeline that mirrors Spotify playlists into a self-hosted library with zero duplicate downloads across playlists.
8. Owning Your Inbox: Archiving Years of Email Without Losing Any of It — safely moving years of email off a provider’s servers and onto storage I actually control.
9. A Searchable Memory: Archiving Text History Into a Queryable Database — turning years of SMS history into a real, queryable database an AI agent can search on request.

If there’s a common thread across all nine, it’s this: self-hosting something isn’t just “run it and forget it” — the interesting engineering is almost always in the pieces that don’t show up in a quickstart guide: what happens when a network mount doesn’t support the operation you assumed it would, how you keep a credential out of plaintext without losing the ability to automate around it, how you notice something’s broken when the very thing that would normally tell you is the thing that’s down. Each post above digs into that layer, not just the happy path.