dotfiles

One flake for the whole distribution. NixOS + Home Manager configure every personal machine from one reviewed source of truth.

Hosts

The flake exports exactly three NixOS host configurations:

Host	Hardware	Role
coordinator	Framework Desktop, AMD Strix Halo	controller node, daily driver
nas	UGREEN NASync DXP2800 GT, AMD R2514	Ethernet-only headless media/storage behind the coordinator relay; Immich ML stays on coordinator
zenbook-duo	ASUS Zenbook Duo, Intel	thin-client laptop

flake.nix        three NixOS hosts wired through one mkHost
modules/         common.nix (every host) + strix.nix (AMD Strix Halo coordinator)
hosts/           one module per machine
home/            home-manager: typed nix (home.nix, nvim.nix) + RAW out-of-store
                 configs (niri KDL, kitty, fish, nvim lua) linked via mkOutOfStoreSymlink
overlays/ pkgs/  custom packages (mactahoe themes, backlog-md, …)
docs/local-ai/  current local-AI appliance docs, model roster, monthly tallies
docs/old/       archival stub indexing retired docs in Git history

⚠️ Qwen Token Plan — use it or lose it

Bought 2026-08-07 · $51 (from $75) · 92 days · expires 2026-11-07. 10,000 credits every 7 days. They do NOT roll over.

The budget is a rolling weekly allowance, not a quarterly pot: at ~13 weeks that is ~130,000 credits total, and every week you underspend is burned. To get the $51 back you need to average ~1,400 credits a day — spending nothing Monday to Friday and then hammering it on Saturday does not recover the week.

• Concurrency is the lever. The plan explicitly allows 3–4 agents at once. One serial pi session cannot spend 10,000 credits a week; three or four parallel ones can. Reach for the fleet, not a bigger single prompt.
• The advertised 3,000 credits / 5 hours burst cap is Temporarily removed by Alibaba as of purchase, so nothing throttles a hard day right now.
• Models on the plan: qwen3.8-max, qwen3.7-max, glm-5.2, deepseek-v4-flash-0731 — all four verified live against the endpoint on purchase day and available in pi under the Qwen Token Plan provider.

Wiring: key at secrets/qwencloud-token.age (coordinator-only), delivered to /run/agenix/qwencloud-token, read per request by pi via ~/.pi/agent/models.json (generated by home/pi.nix). It is never exported into the environment.

Branches

• main — canonical and NixOS-only.
• archive/* — read-only historical lineage; never used for deployment.

Usage

# on a NixOS host
sudo nixos-rebuild switch --flake .#<host>