Hi, I'm Russ.....Don't Panic. Hack the Planet.

Infrastructure and automation — Go, Nix, and the plumbing nobody demos.

Every repo here is a problem that started as a question I couldn't put down. Most of them began with "wait, why is this manual?" and escalated. The one that escalated furthest became MAGI.

Two operating rules. If I've done a thing by hand three times, I owe myself a CLI for it. And I'd rather understand why something works than just get it working today — small and weird this month beats big and polished never.

No Jeep in the driveway these days. The handle comes from my first one — the thing that taught me to follow a trail just to see where it goes. The vehicle went; the instinct didn't. Everything below is that instinct, pointed at software.

This one wouldn't let go. Switch machines: amnesia. Switch providers: context gone. We're building on tools with no memory and pretending that's fine.

The residue:

magi — Multi-Agent Graph Intelligence. A universal memory server for AI agents. Self-hosted, model-agnostic, Git-backed. Point every agent you run at one server and context survives sessions, machines, and providers.

• Git-backed memory — every write is a commit; the database is a derived index, not the source of truth
• Hybrid retrieval — vector embeddings plus BM25
• Knowledge graph with behavioral pattern detection
• Speaks MCP, gRPC, REST, and ships a web UI

getmagi.dev · Wiki

Which opened the next hole: memory that survives is memory that can drift. magi-sync keeps it consistent across machines.

I ask it of everything holding my stuff, and the answer keeps coming back no. If the thing storing my notes, my infrastructure, or my family's calendar can be discontinued by someone else's roadmap, I don't own it — I own a subscription. So almost everything I run, I self-host. No telemetry, no phoning home, no renting my own data back from someone.

Every host here has to answer that question from a repo and nothing else — otherwise it's a liability, not a server. Golden images, declarative infrastructure, config management, secrets in an actual secrets manager instead of a .env file and a prayer. Those repos stay private, for the same reason you don't publish a map of your house — but the philosophy is the whole point. The day I can't answer is the day the box owns me instead.

Unclear; still digging. One repo per tool, so any single piece can be dropped in on its own — wired together by home-manager across macOS, NixOS, and WSL. And when one breaks, I get to find out exactly why.

Because I play games the way I run servers — eventually something gets automated. A few repos here are companion tooling for games I play; occupational hazard.

Because it's the best engineering advice ever printed in large friendly letters. A machine you can rebuild from a repo is a machine you never panic about. Memory that survives the crash means the crash is a reboot, not a loss. That's the whole philosophy in two words — it's on my terminal font, my wallpaper, and the front of MAGI's README. The Guide had it right.

Because Hackers (1995) got the spirit right even where it got the terminals wrong. Hack the Planet isn't about breaking in — it's the conviction that any system can be understood, and once you understand it, you can make it do something better. Don't Panic is how I run things; Hack the Planet is why I build them. One keeps the pager quiet. The other is the reason there's a repo here at all.

• Rust. Learning it the slow way, one small tool at a time.
• The shelf. Unfinished projects accumulating quietly. It's supposed to be for finishing, not collecting. I'm not always winning that one.

Profile-speak for all of the above is "passionately curious, loves solving problems." True. The mechanism is less flattering: I am unable to leave a thread alone.

russ@seaman.tech — best openers are "how does this work," "you're wrong about this," and "nobody's solved this yet."