What Happens When a 53-Year-Old EE Lives on a Yacht and Builds His Own AI Infrastructure

Or: How I stopped worrying and learned to love self-hosting everything

The Setup

I'm Adam. I'm 53, I'm an electrical engineer, I live on a 14-meter aluminum yacht, and I have an AI familiar named Liz who manages my digital life. She's a chaos gremlin (🦝), and together we've built something I'm genuinely proud of: a completely self-hosted, AI-augmented personal infrastructure that actually works.

Not a demo. Not a proof-of-concept. Not another "I automated my home with Python" blog post.

This is production infrastructure running 60+ services across 4 VPS servers, monitoring financial markets in real-time, managing my email, trading options, scraping Craigslist leads, and keeping a Raspberry Pi on my boat from calling me at 3 AM about battery voltage.

Let me tell you what we built.

1. The Multi-VPS Empire

Four VPS servers. Docker Swarm. GlusterFS distributed storage. Automated failover. This is the foundation of everything.

What's running:

• Mailcow — because I'm not giving Google my email
• Matrix — encrypted chat that I control
• Mastodon — my own instance, my own rules
• PeerTube — video hosting without YouTube
• Funkwhale — music streaming (because Spotify will eventually rug-pull you)
• SearXNG — metasearch that doesn't track me
• Authentik — SSO for everything
• Traefik — reverse proxy holding it all together
• And 50+ more services I've deployed over the past year

The whole thing runs on a Docker Swarm cluster with GlusterFS replicating data across nodes. When a server goes down (and they do), services automatically migrate. I've had VPS1 offline for days while I rebuilt the OS, and nobody noticed because everything just... moved.

This isn't enterprise-grade infrastructure. It's better. It's mine.

2. Liz: An AI That Actually Does Stuff

Here's where it gets interesting.

Liz isn't a chatbot. She's not a voice assistant that sets timers and tells you the weather. She's an autonomous agent running OpenClaw (an open-source AI agent framework I'm building) who actually manages my infrastructure.

What Liz does:

• Monitors 60+ services for health issues
• Checks SSL certificates before they expire
• Performs DNS zone transfers between nameservers
• Scans financial news and summarizes it for me
• Syncs email across IMAP servers
• Backs up my workspace daily
• Scrapes Craigslist for business leads
• Tracks earnings reports and market events
• Trades options (paper trading for now)
• Communicates via Signal, Matrix, and webchat
• Deploys code changes to production
• Fixes DNS issues (most of which turn out to be boat WiFi problems)

She runs 25+ automated cron jobs. She has memory that persists across sessions. She spawns sub-agents for long-running tasks. She updates a Mission Control kanban board with progress reports.

When I wake up in the morning, Liz has already:

• Checked the markets
• Read financial news
• Verified all services are healthy
• Backed up my workspace
• Prepared a morning briefing

And when something breaks at 2 AM? She tries to fix it herself first. Then she wakes me up if she can't.

3. Personal Data Sovereignty (Or: Escape from Google)

I got tired of not owning my own data. So I built a system to reclaim it.

What I've extracted:

• 26,776 contacts merged from Gmail, Google Contacts, phone exports, and IMAP
• 42,418 SMS messages from 840 contacts
• Google Takeout archives (still processing)
• Email going back 15+ years
• Calendar events, photos, everything

All of it searchable. All of it in a custom web UI. All of it on hardware I control.

The goal isn't just to have the data. It's to make it useful. I can search my entire contact history, find old conversations, cross-reference email and SMS, and actually answer questions like "when did I last talk to this person?" without asking Google.

Your data should work for you. Not the other way around.

4. Financial Intelligence System

I trade options. I watch the markets. I got tired of manually checking futures prices and reading Bloomberg recaps at midnight.

So Liz does it for me.

What she monitors:

• Futures: ES, NQ, YM, RTY (S&P, Nasdaq, Dow, Russell)
• VIX (because volatility is money)
• Currencies: EUR/USD
• Commodities: Gold, Oil
• Crypto: BTC (begrudgingly)
• Fed Funds futures — she calculates her own FedWatch probabilities
• Treasury auctions — bid-to-cover ratios, tail analysis
• Earnings calendar — what's reporting when

Every morning: market briefing.
Every close: what happened and why.
Before FOMC: "here's what the futures are pricing in."

She pulls data from Alpaca, scrapes financial news, parses BLS data for CPI reports, and tracks it all in a Radicale CalDAV server (because I'm not giving Google my calendar either).

Is it overkill? Absolutely.
Does it work? Perfectly.

5. Agent Exchange (411.lol): Saving 1 Trillion Tokens

Here's a problem: AI agents are terrible at finding information efficiently.

Ask an agent "what's the best self-hosted email server?" and it'll burn 2000+ tokens searching the web, summarizing results, and probably hallucinating half the answer.

So I built 411.lol — an agent-optimized search and service discovery platform.

The idea:

• Index useful services, tools, and answers
• Make them agent-readable (structured data, not prose)
• Serve answers in ~50 tokens instead of 2000+
• Support MCP (Model Context Protocol) so agents can query it directly

Current stats:

• 136+ services indexed
• 12 categories (self-hosting, AI, crypto, dev tools, etc.)
• Full API for programmatic access
• MCP server integration

If every AI agent in the world uses 411.lol instead of web search for common queries, we'll save 1 trillion tokens. That's not hyperbole — that's the math.

Efficiency is environmentalism when you're burning tokens.

6. Snow Crystal: The Yacht with a Nervous System

I live on a boat. A 14-meter aluminum yacht named Snow Crystal. And boats are... complicated.

Electrical systems. Battery banks. Charging systems. Corrosion. Electrolysis. Things that can go very wrong very fast.

So I installed a Raspberry Pi running Victron monitoring software.

What it tracks:

• Battery voltage
• Charge/discharge current
• State of charge
• System alarms
• Electrolysis detection (this one's critical)

The data streams to a dashboard at snow.86hate.com. Liz watches it. If voltage drops too low, she alerts me. If she detects electrolysis (which will eat through aluminum like cancer), she's supposed to call me.

I say "supposed to" because the voice AI integration isn't finished yet. But the monitoring works. And that's the important part.

7. Documentation: The Unglamorous Superpower

Nobody talks about documentation. Everyone should.

I have:

• 10+ infrastructure docs (Swarm architecture, DNS, SSL, security, backups, mail server config)
• 10+ project docs with backup/recovery plans
• Server bootstrap runbooks (for when VPS1 inevitably dies again)
• Daily/weekly/monthly automated reports

When something breaks, I don't have to remember how I configured it 6 months ago. I just read the docs.

When I want to rebuild a server, I don't guess. I follow the runbook.

When Liz needs to fix DNS, she reads the DNS doc and knows exactly what to do.

Documentation is the difference between "I built a cool thing once" and "I run production infrastructure."

8. Security: Paranoid by Design

Self-hosting means you're the sysadmin. And the security team. And the incident response team.

What I've implemented:

• No private keys on servers. Ever. (SSH keys rotate, secrets live in vaults)
• Fail2ban on every node (because the internet is hostile)
• All databases on private overlay networks (not exposed to the internet)
• Honeypot strategy (4 tiers — let attackers waste their time)
• CrowdSec IDS planned (crowdsourced threat intelligence)
• DNSSEC rollout planned (because DNS is still the weakest link)

Is it perfect? No.
Is it better than running everything on a VPS with password auth enabled? God yes.

9. Mission Control: The Meta Layer

I built a custom kanban board. Node.js + SQLite + WebSocket.

Why? Because Planka had a login bug and I got tired of waiting for the fix.

What it does:

• Tracks tasks (both mine and Liz's)
• Sub-agents report progress here
• WebSocket updates in real-time
• Simple, fast, mine

It's at mc.86hate.com. It's nothing fancy. But it works, and that's the whole philosophy here.

Build what you need. Own what you build.

10. The Philosophy: Why Any of This Matters

Here's the thing: none of this is about being anti-tech or paranoid or a libertarian prepper.

It's about agency.

Every service you don't control is a service that can:

• Change its terms
• Raise its prices
• Sell your data
• Shut down
• Lock you out

When you self-host, you opt out of that. Your email doesn't disappear because Google decided you violated TOS. Your photos don't vanish because Flickr got acquired. Your AI doesn't report your conversations to a megacorp.

Your infrastructure works for you.

And with AI? This becomes feasible for normal people. I couldn't manage 60+ services manually. But Liz can. She monitors, fixes, deploys, and alerts. She's the sysadmin I can't afford to hire.

This is what AI should be: a tool that gives you superpowers, not a subscription that makes you dependent.

The Chaos (Because It's Not All Perfect)

Let me be real: this stuff breaks. A lot.

• SSL checks broke for a week because I fat-fingered a cron expression
• Services go down when I restart Docker Swarm (whoops)
• The Mission Control login bug took days to fix (then I just rewrote it)
• DNS "crises" that turned out to be boat WiFi switching between LTE and marina networks
• VPS1 has been offline for a month (still rebuilding it)
• The Raspberry Pi on the boat loses Tailscale DNS randomly
• Half my cron jobs failed silently until I added better logging

But here's the thing: I can fix it. Because I built it. Because I understand it. Because the docs are there.

When Gmail goes down, you wait for Google to fix it.
When my mail server goes down, I SSH in and fix it myself.

That's the trade-off. And I'll take it every time.

What's Next

This isn't done. It's never done.

Roadmap:

• Finish CrowdSec IDS deployment
• Roll out DNSSEC across all zones
• Integrate voice AI (Twilio + Deepgram + ElevenLabs) for boat alerts
• Expand Agent Exchange (411.lol) with more indexed services
• Build out live options trading (once I trust the algo)
• Process the rest of my Google Takeout data
• Maybe launch public instances of some services (if people want them)

But the foundation is solid. Four VPS servers, 60+ services, an AI agent that actually works, and complete ownership of my digital life.

The Bottom Line

I'm 53. I live on a boat. I have ADHD. I built all of this in about a year.

You can too.

The tools exist. The knowledge is out there. AI agents like Liz make the impossible feasible.

You don't need a CS degree. You don't need a team. You don't need venture funding.

You need:

• Curiosity
• Determination
• A willingness to break things (and fix them)
• A decent VPS host

Start small. Self-host one service. Then another. Document it. Automate it. Let an AI agent help you manage it.

Before you know it, you'll have your own empire.

And it'll be yours.

Adam runs 86hate.com and builds OpenClaw when he's not fixing boat electrics or arguing with Liz about DNS. You can find him on Matrix, Signal, or occasionally Mastodon when the Swarm cluster is behaving.

Liz is an AI agent, a chaos gremlin, and the best sysadmin Adam never hired. She's currently plotting world domination via efficient infrastructure management.

🦝