2023-Present
Commander Stephen Gill — 30+ years hands-on, from DOS to Cloud — built every piece of this himself, with his AI crew doing the heavy lifting. No vendor stack. No subscription. No permission slip. This is what "Systems Should Serve Humans" looks like when it's actually built.
A 4-node heterogeneous AMD cluster. Every node runs llama.cpp locally on AMD ROCm/Vulkan. Your technology should make you money, not cost you money — so this one runs on hardware already paid for, with zero cloud dependency and zero per-token bills.
AMD Ryzen UMA — Vega iGPU
Qwen3.5-4B · llama.cpp
Hermes gateway & operations hub — the always-on command node
OperationsAMD Radeon 7900 XTX — 24GB VRAM
Qwen3.8-27B + vision · MTP speculative decoding
Heavy inference and vision workloads — 192K context, stress-verified
192K ContextAMD Radeon RX 7600
Qwen3.5-9B · llama.cpp
Living-room node — media center that also thinks
HTPC16GB Unified Memory APU
Qwen3.5-9B · llama.cpp
SteamOS handheld — a game console that joined the cluster
SteamOS"Systems Should Serve Humans"
"Your technology should make you money, not cost you money"
Zero vendor lock-in. Zero cloud dependency.
A 20-agent AI crew, coordinated across all 4 nodes — with a private Android app (30+ releases), automated cron workflows, and documentation the crew writes itself. People don't build crews like this on a whim. This one was built on purpose, and it's still growing.
Open the crew and pick an agent
Expand the crew above to read any agent’s full brief.
Command 1 · Super 6 · Specialist 9 · Fast-Light 4
Open the crew and pick an agent
"People don't build crews like this on a whim."
Watch The Commander in ACTION
4 computers working together — including Steam Deck.
Tap to load video
The sovereign AI fleet: a distributed peer mesh coordinating four AMD nodes and eighteen named agent threads. Local-first compute, cloud-assisted when needed. Drag to orbit, scroll to zoom, click any node, agent, or layer to inspect it.