feat(open-proxy): configure shared token file

Move mosh/unzip into common home packages, add orca-slicer to the thinkpad home, and remove the now-redundant system package list and kernel param.

.agents/skills/update-package-hashes/SKILL.md

@@ -12,7 +12,7 @@ If the user provides only a **package name** (no version), look up the latest ve
 ## Hard Rules — Read First
 
 1. **Never run `nix build .#<pkg>`** or `.#packages.<system>.<pkg>`. That compiles the package. Only realise **FOD sub-attributes** (`.src`, `.goModules`, `.npmDeps`, `.cargoDeps`) — those are pure downloads, not builds.
-2. **Never** use `nix-prefetch-git`, `nix-prefetch-url`, `nix hash path`, `git clone` + manual hashing, `builtins.fetchGit`, or any other ad-hoc method to compute hashes. They produce hashes in formats that don't match what `fetchgit`/`fetchFromGitHub`/etc. expect, and you will waste time chasing mismatches.
+2. **Never** use `nix-prefetch-git`, `nix-prefetch-github`, `nix-prefetch-url`, `nix hash path`, `nix hash file` (on a raw patch/tarball), `git clone` + manual hashing, `builtins.fetchGit`, or any other ad-hoc method to compute hashes. They produce hashes in formats that don't match what `fetchgit`/`fetchFromGitHub`/`fetchpatch` expect (notably: `fetchFromGitHub { leaveDotGit = true; }` is non-deterministic across machines, and `fetchpatch` normalizes patches — strips `index abc..def`, `From <sha>`, signatures — so its hash ≠ `nix hash file` of the raw `.patch`).
 3. There are exactly **two** correct ways to get a hash, both listed below. If neither fits, stop and ask the user — don't improvise.
 
 ## The Only Two Methods
@@ -29,17 +29,27 @@ Copy the `hash = "sha256-..."` line from the output into the package's `src` blo
 
 ### Method B — FOD mismatch trick (for everything else)
 
-For `vendorHash`, `npmDepsHash`, `cargoHash`, `cargoLock.outputHashes.<crate>`, or any `src` using a custom fetcher (`leaveDotGit`, `postFetch`, `fetchSubmodules`, etc. — applies to `llama-cpp` and `llama-swap`), realise the **specific FOD sub-attribute** and read the `got:` line from the error.
+For `vendorHash`, `npmDepsHash`, `cargoHash`, `cargoLock.outputHashes.<crate>`, `fetchpatch` hashes, or any `src` using a custom fetcher (`leaveDotGit`, `postFetch`, `fetchSubmodules`, etc. — applies to `llama-cpp` and `llama-swap`), realise the **specific FOD sub-attribute** and read the `got:` line from the error.
 
 ```bash
 nix build .#<name>.src --no-link 2>&1 | tee /tmp/hash.log         # for src
 nix build .#<name>.goModules --no-link 2>&1 | tee /tmp/hash.log   # for vendorHash
 nix build .#<name>.npmDeps --no-link 2>&1 | tee /tmp/hash.log     # for npmDepsHash
 nix build .#<name>.cargoDeps --no-link 2>&1 | tee /tmp/hash.log   # for cargoHash
+nix build .#<name> --no-link 2>&1 | tee /tmp/hash.log             # for fetchpatch / other input FODs (see note)
 grep -E '^[[:space:]]*got:' /tmp/hash.log | tail -1 | awk '{print $2}'
 ```
 
-Setting the hash to `lib.fakeHash` (preferred when `lib` is in scope), `sha256-AAAA...` (44 A's), or leaving the old one in place all work — the build will fail at the FOD with `got: sha256-...` which is the correct value.
+**`fetchpatch` note:** patches don't have a dedicated sub-attribute, so you must target the package itself. This is safe *only* when the patch hash is wrong (e.g. `lib.fakeHash`) — Nix realizes the patch FOD before compilation starts, so a hash mismatch aborts with `0 built (1 failed)` and zero compile work. If you accidentally fix all FODs correctly, `nix build .#<name>` will start compiling. To guard against this: always start patch hashes as `lib.fakeHash`, run the build, copy `got:`, paste, and only then re-verify with `.src` / sub-attribute builds (never re-run `.#<name>` to confirm).
+
+**GitHub PR patches — `.patch` vs `.diff`:** When fetching a patch from a GitHub pull request, prefer the `.diff` endpoint over `.patch`.
+
+- `https://github.com/<owner>/<repo>/pull/<N>.patch` — a `git format-patch` **mbox** containing each commit in the PR separately. `git apply` (which `fetchpatch` and the Nix `patchPhase` use) does **not** replay commit history; it applies hunks against the working tree. PRs that create a file in one commit and delete/rename it in a later commit will fail with errors like `The next patch would delete the file X, which does not exist`.
+- `https://github.com/<owner>/<repo>/pull/<N>.diff` — a **squashed** unified diff of the PR's net change. Applies cleanly against any base the PR is mergeable against.
+
+Default to `.diff`. Only fall back to `.patch` if you specifically need authorship metadata (rare for Nix patching). If a previously-working `.patch` URL suddenly fails to apply, switching to `.diff` is the first thing to try.
+
+Always use `lib.fakeHash` (or `pkgs.lib.fakeHash` if only `pkgs` is in scope). This is the only reliable way to set a bogus hash — never write a literal `sha256-...` placeholder string. The build will fail at the FOD with `got: sha256-...` which is the correct value.
 
 **Note:** `.src`, `.goModules`, etc. are sub-attributes of the derivation. They download but do not compile. `nix build .#<name>` (without the `.src` suffix) compiles — never do that.
 

.agents/skills/update-vllm-3090-configs/SKILL.md

@@ -0,0 +1,71 @@
+---
+name: update-vllm-3090-configs
+description: Update only the qwen3.6-27b vLLM 3090 llama-swap configs from club-3090 refs; compare diffs, present a plan, and require approval before editing.
+---
+
+# Update vLLM 3090 Configs
+
+## Scope
+
+Use only for Qwen3.6 27B vLLM 3090 configs in `modules/nixos/services/llama-swap/`.
+Do not use this skill for other models, other Qwen sizes, non-vLLM configs, or package bumps.
+
+Local files:
+- `modules/nixos/services/llama-swap/config.nix`
+- `modules/nixos/services/llama-swap/setup-qwen36-vllm.sh`
+
+Local config keys:
+- `vllm-qwen3.6-27b-tools-text`
+- `vllm-qwen3.6-27b-long-text`
+- `vllm-qwen3.6-27b-long-vision`
+
+## Hash Tracking
+
+Each config entry stores an upstream commit hash comment:
+  `# Upstream: club-3090 <hash> (<date>) - <compose-file>`
+
+When comparing, first extract stored hashes. If a config's hash matches
+upstream HEAD, skip it (report "already synced"). Only full-diff configs
+whose hash differs. Update the hash comment when edits are applied.
+
+## Upstream References
+
+Compare against `club-3090` master:
+- `models/qwen3.6-27b/vllm/compose/single/tools-text.yml`
+- `models/qwen3.6-27b/vllm/compose/single/long-text.yml`
+- `models/qwen3.6-27b/vllm/compose/single/long-vision.yml`
+- `scripts/setup.sh` for the current `GENESIS_PIN="${GENESIS_PIN:-...}"`
+
+Use raw URLs or a temp clone under `_scratch/club-3090`. Prefer a temp clone when checking broad changes:
+
+```bash
+mkdir -p _scratch
+git clone https://github.com/noonghunna/club-3090 _scratch/club-3090 2>/dev/null || git -C _scratch/club-3090 pull --ff-only
+```
+
+## Required Workflow
+
+1. Fetch/update upstream refs under `_scratch/club-3090` or fetch the raw files.
+2. Extract stored upstream hashes from `# Upstream: club-3090 ...` comments in config.nix. Skip any config whose hash matches upstream HEAD (report "already synced").
+3. Compare upstream compose files to the remaining local llama-swap entries. Translate docker-compose semantics into the existing `docker run`/llama-swap format.
+4. Compare upstream `scripts/setup.sh` Genesis pin to local `GENESIS_PIN` in `setup-qwen36-vllm.sh`.
+5. Check upstream compose volumes/entrypoint for sidecar patches. If patches are added, removed, renamed, or invoked differently, update both:
+   - runtime mounts and `python3 /patches/...` calls in `config.nix`
+   - download/install logic and summary in `setup-qwen36-vllm.sh`
+6. Ignore these diffs unless the user explicitly asks otherwise:
+   - `shm_size` / shm-related compose settings
+   - local timing patch `patch_timings_07351e088.py` and its mount/invocation
+   - model served-name differences caused by llama-swap `${MODEL_ID}`
+   - `HUGGING_FACE_HUB_TOKEN`; keep local CUDA device/env choices
+   - upstream relative paths vs local `/mnt/ssd/vLLM/...` paths
+   - docker-compose format vs local llama-swap/Nix format
+7. Before editing, present:
+   - upstream files/commit checked
+   - meaningful diffs found
+   - ignored diffs
+   - exact planned local changes
+   Then wait for explicit user approval.
+8. After approval, edit minimally and update the `# Upstream: club-3090 ...` hash comments. Validate:
+   - `bash -n modules/nixos/services/llama-swap/setup-qwen36-vllm.sh`
+   - `nix-instantiate --parse modules/nixos/services/llama-swap/config.nix`
+9. Summarize changed files and any remaining upstream differences.

flake.lock

@@ -8,11 +8,11 @@
         ]
       },
       "locked": {
-        "lastModified": 1776147994,
-        "narHash": "sha256-c5F8jYiB0fjWsP4j/yeszqszA3laflzDj6/pmoJTeG4=",
+        "lastModified": 1778234684,
+        "narHash": "sha256-usIHfvSt7aXvMvRGtcbsue3rA13Z+9TW/7I3WBzLqFY=",
         "owner": "nix-community",
         "repo": "nixos-apple-silicon",
-        "rev": "81439a7fb8067ab43641efd79c84607701da1ccd",
+        "rev": "3d7fe422ef6162154830209b9e50bf69e150cff7",
         "type": "github"
       },
       "original": {
@@ -28,16 +28,16 @@
         ]
       },
       "locked": {
-        "lastModified": 1772129556,
-        "narHash": "sha256-Utk0zd8STPsUJPyjabhzPc5BpPodLTXrwkpXBHYnpeg=",
+        "lastModified": 1779036909,
+        "narHash": "sha256-zXcwYQGCT6pzinK+1dBB2ekTVtfxGZAapb3Evdcu4fY=",
         "owner": "nix-darwin",
         "repo": "nix-darwin",
-        "rev": "ebec37af18215214173c98cf6356d0aca24a2585",
+        "rev": "56c666e108467d87d13508936aade6d567f2a501",
         "type": "github"
       },
       "original": {
         "owner": "nix-darwin",
-        "ref": "nix-darwin-25.11",
+        "ref": "nix-darwin-26.05",
         "repo": "nix-darwin",
         "type": "github"
       }
@@ -51,11 +51,11 @@
         "nixpkgs": "nixpkgs_2"
       },
       "locked": {
-        "lastModified": 1775584659,
-        "narHash": "sha256-NA5oZRunqxD+4LNdU7ZKJHqwuazKyAmBjO4OHXL14X4=",
+        "lastModified": 1778179392,
+        "narHash": "sha256-W6zorvjBYbzMNvqKIqCdpDF4rq3gj50Xximl56YM9/I=",
         "owner": "determinatesystems",
         "repo": "determinate",
-        "rev": "21dcaa011d3d35cf42a04e988eaac9b28c97a707",
+        "rev": "efd54faa68be8cd777b5c28cab11e638998a0853",
         "type": "github"
       },
       "original": {
@@ -67,37 +67,37 @@
     "determinate-nixd-aarch64-darwin": {
       "flake": false,
       "locked": {
-        "narHash": "sha256-qLWfYk9qkb21wKCDWnhMfqBFjcdBBJkNUKBlvdHSLgA=",
+        "narHash": "sha256-z4mCqKI3Qd6weuHrlfzGccJG0giym/VJhKv20ijRSs0=",
         "type": "file",
-        "url": "https://install.determinate.systems/determinate-nixd/tag/v3.17.3/macOS"
+        "url": "https://install.determinate.systems/determinate-nixd/tag/v3.20.0/macOS"
       },
       "original": {
         "type": "file",
-        "url": "https://install.determinate.systems/determinate-nixd/tag/v3.17.3/macOS"
+        "url": "https://install.determinate.systems/determinate-nixd/tag/v3.20.0/macOS"
       }
     },
     "determinate-nixd-aarch64-linux": {
       "flake": false,
       "locked": {
-        "narHash": "sha256-0BmprPIRTopvJ2QdImOMP+TujAPVgRdl0bUL3vhqGIY=",
+        "narHash": "sha256-yW+VNepSRytzfanSssPMJPvwioCcmlZYaBX8++UFkAk=",
         "type": "file",
-        "url": "https://install.determinate.systems/determinate-nixd/tag/v3.17.3/aarch64-linux"
+        "url": "https://install.determinate.systems/determinate-nixd/tag/v3.20.0/aarch64-linux"
       },
       "original": {
         "type": "file",
-        "url": "https://install.determinate.systems/determinate-nixd/tag/v3.17.3/aarch64-linux"
+        "url": "https://install.determinate.systems/determinate-nixd/tag/v3.20.0/aarch64-linux"
       }
     },
     "determinate-nixd-x86_64-linux": {
       "flake": false,
       "locked": {
-        "narHash": "sha256-+Q85cySxr0FB/cr97hk/WWYgeJY+iC4OH+FjGYygIbU=",
+        "narHash": "sha256-+L102C3Hhkd1GlXmRm2eLTLsZKBxEvooiQZFqQRlBf0=",
         "type": "file",
-        "url": "https://install.determinate.systems/determinate-nixd/tag/v3.17.3/x86_64-linux"
+        "url": "https://install.determinate.systems/determinate-nixd/tag/v3.20.0/x86_64-linux"
       },
       "original": {
         "type": "file",
-        "url": "https://install.determinate.systems/determinate-nixd/tag/v3.17.3/x86_64-linux"
+        "url": "https://install.determinate.systems/determinate-nixd/tag/v3.20.0/x86_64-linux"
       }
     },
     "disko": {
@@ -105,11 +105,11 @@
         "nixpkgs": "nixpkgs_3"
       },
       "locked": {
-        "lastModified": 1773889306,
-        "narHash": "sha256-PAqwnsBSI9SVC2QugvQ3xeYCB0otOwCacB1ueQj2tgw=",
+        "lastModified": 1777713215,
+        "narHash": "sha256-8GzXDOXckDWwST8TY5DbwYFjdvQLlP7K9CLSVx6iTTo=",
         "owner": "nix-community",
         "repo": "disko",
-        "rev": "5ad85c82cc52264f4beddc934ba57f3789f28347",
+        "rev": "63b4e7e6cf75307c1d26ac3762b886b5b0247267",
         "type": "github"
       },
       "original": {
@@ -126,11 +126,11 @@
       },
       "locked": {
         "dir": "pkgs/firefox-addons",
-        "lastModified": 1776199335,
-        "narHash": "sha256-ImihxU7ReZZuNdrASq8qzOmmO/UQtkuqQ9V9KKb1dD0=",
+        "lastModified": 1778385775,
+        "narHash": "sha256-n0MUvWA2SML/qBB4hpShQ7i+i961MX4oPtaQfYo0+uU=",
         "owner": "rycee",
         "repo": "nur-expressions",
-        "rev": "95066e56aaa948f170747f57a20c99511a953eed",
+        "rev": "268324916742a48cd03b94fd63f2822d6b66d519",
         "type": "gitlab"
       },
       "original": {
@@ -278,16 +278,16 @@
         ]
       },
       "locked": {
-        "lastModified": 1775425411,
-        "narHash": "sha256-KY6HsebJHEe5nHOWP7ur09mb0drGxYSzE3rQxy62rJo=",
+        "lastModified": 1780361225,
+        "narHash": "sha256-wnV9ttf4fPWNonBIQmvlrSlNpQYgx5HgWWd007mwIFA=",
         "owner": "nix-community",
         "repo": "home-manager",
-        "rev": "0d02ec1d0a05f88ef9e74b516842900c41f0f2fe",
+        "rev": "e28654b71096e08c019d4861ca26acb646f583d8",
         "type": "github"
       },
       "original": {
         "owner": "nix-community",
-        "ref": "release-25.11",
+        "ref": "release-26.05",
         "repo": "home-manager",
         "type": "github"
       }
@@ -301,12 +301,12 @@
         "nixpkgs-regression": "nixpkgs-regression"
       },
       "locked": {
-        "lastModified": 1775583600,
-        "narHash": "sha256-/shs/3GA4R3rxhhqpPbEMnDZKbCvf3VpwnHB75nkTcI=",
-        "rev": "e9b4735be7b90cf49767faf5c36f770ac1bdc586",
-        "revCount": 24880,
+        "lastModified": 1778177425,
+        "narHash": "sha256-oyHvP5HDRe59opmjTrq2ED9lh+R9FrHyaCGPPNfBqWM=",
+        "rev": "f0ccb960d3ad5bff28acd9cabf8bdef885b5d52f",
+        "revCount": 25858,
         "type": "tarball",
-        "url": "https://api.flakehub.com/f/pinned/DeterminateSystems/nix-src/3.17.3/019d6913-e8c2-7128-ba76-3dc4f6b58158/source.tar.gz"
+        "url": "https://api.flakehub.com/f/pinned/DeterminateSystems/nix-src/3.20.0/019e03bc-3f83-7833-aba3-b691ef4956c7/source.tar.gz"
       },
       "original": {
         "type": "tarball",
@@ -351,16 +351,16 @@
     },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1761597516,
-        "narHash": "sha256-wxX7u6D2rpkJLWkZ2E932SIvDJW8+ON/0Yy8+a5vsDU=",
-        "rev": "daf6dc47aa4b44791372d6139ab7b25269184d55",
-        "revCount": 811874,
+        "lastModified": 1773222311,
+        "narHash": "sha256-BHoB/XpbqoZkVYZCfXJXfkR+GXFqwb/4zbWnOr2cRcU=",
+        "rev": "0590cd39f728e129122770c029970378a79d076a",
+        "revCount": 909248,
         "type": "tarball",
-        "url": "https://api.flakehub.com/f/pinned/NixOS/nixpkgs/0.2505.811874%2Brev-daf6dc47aa4b44791372d6139ab7b25269184d55/019a3494-3498-707e-9086-1fb81badc7fe/source.tar.gz"
+        "url": "https://api.flakehub.com/f/pinned/NixOS/nixpkgs/0.2511.909248%2Brev-0590cd39f728e129122770c029970378a79d076a/019ce32b-8ace-7339-b129-cceaa8dd10c6/source.tar.gz"
       },
       "original": {
         "type": "tarball",
-        "url": "https://flakehub.com/f/NixOS/nixpkgs/0.2505"
+        "url": "https://flakehub.com/f/NixOS/nixpkgs/0.2511"
       }
     },
     "nixpkgs-23-11": {
@@ -397,11 +397,11 @@
     },
     "nixpkgs-unstable": {
       "locked": {
-        "lastModified": 1775710090,
-        "narHash": "sha256-ar3rofg+awPB8QXDaFJhJ2jJhu+KqN/PRCXeyuXR76E=",
+        "lastModified": 1777954456,
+        "narHash": "sha256-hGdgeU2Nk87RAuZyYjyDjFL6LK7dAZN5RE9+hrDTkDU=",
         "owner": "NixOS",
         "repo": "nixpkgs",
-        "rev": "4c1018dae018162ec878d42fec712642d214fdfa",
+        "rev": "549bd84d6279f9852cae6225e372cc67fb91a4c1",
         "type": "github"
       },
       "original": {
@@ -413,12 +413,12 @@
     },
     "nixpkgs_2": {
       "locked": {
-        "lastModified": 1775464765,
-        "narHash": "sha256-nex6TL2x1/sVHCyDWcvl1t/dbTedb9bAGC4DLf/pmYk=",
-        "rev": "83e29f2b8791f6dec20804382fcd9a666d744c07",
-        "revCount": 975711,
+        "lastModified": 1777826146,
+        "narHash": "sha256-wQ/iN5Zp5VIa3ebBibijPnLyKhor+xEbDy4d0goa9Zs=",
+        "rev": "73c703c22422b8951895a960959dbbaca7296492",
+        "revCount": 991389,
         "type": "tarball",
-        "url": "https://api.flakehub.com/f/pinned/DeterminateSystems/nixpkgs-weekly/0.1.975711%2Brev-83e29f2b8791f6dec20804382fcd9a666d744c07/019d6689-cde2-7061-b044-e0ef61ade488/source.tar.gz"
+        "url": "https://api.flakehub.com/f/pinned/DeterminateSystems/nixpkgs-weekly/0.1.991389%2Brev-73c703c22422b8951895a960959dbbaca7296492/019df6c8-934b-7d40-b402-027bb5def30f/source.tar.gz"
       },
       "original": {
         "type": "tarball",
@@ -443,16 +443,16 @@
     },
     "nixpkgs_4": {
       "locked": {
-        "lastModified": 1776067740,
-        "narHash": "sha256-B35lpsqnSZwn1Lmz06BpwF7atPgFmUgw1l8KAV3zpVQ=",
+        "lastModified": 1780203844,
+        "narHash": "sha256-K5sT4jTpGs15ADhviMKNBH38REpPf5Q6mM1+N6cArVE=",
         "owner": "NixOS",
         "repo": "nixpkgs",
-        "rev": "7e495b747b51f95ae15e74377c5ce1fe69c1765f",
+        "rev": "b51242d7d43689db2f3be91bd05d5b24fbb469c4",
         "type": "github"
       },
       "original": {
         "owner": "NixOS",
-        "ref": "nixos-25.11",
+        "ref": "nixos-26.05",
         "repo": "nixpkgs",
         "type": "github"
       }
@@ -501,11 +501,11 @@
         ]
       },
       "locked": {
-        "lastModified": 1776119890,
-        "narHash": "sha256-Zm6bxLNnEOYuS/SzrAGsYuXSwk3cbkRQZY0fJnk8a5M=",
+        "lastModified": 1777944972,
+        "narHash": "sha256-VfGRo1qTBKOe3s2gOv8LSoA6Fk19PvBlwQ1ECN0Evn8=",
         "owner": "Mic92",
         "repo": "sops-nix",
-        "rev": "d4971dd58c6627bfee52a1ad4237637c0a2fb0cd",
+        "rev": "c591bf665727040c6cc5cb409079acb22dcce33c",
         "type": "github"
       },
       "original": {

flake.nix

@@ -2,7 +2,7 @@
   description = "NixOS Hosts";
 
   inputs = {
-    nixpkgs.url = "github:NixOS/nixpkgs/nixos-25.11";
+    nixpkgs.url = "github:NixOS/nixpkgs/nixos-26.05";
     nixpkgs-unstable.url = "github:NixOS/nixpkgs/nixos-unstable";
     disko.url = "github:nix-community/disko";
     determinate.url = "github:determinatesystems/determinate";
@@ -11,7 +11,7 @@
       inputs.nixpkgs.follows = "nixpkgs";
     };
     home-manager = {
-      url = "github:nix-community/home-manager/release-25.11";
+      url = "github:nix-community/home-manager/release-26.05";
       inputs.nixpkgs.follows = "nixpkgs";
     };
     apple-silicon = {
@@ -31,7 +31,7 @@
       inputs.nixpkgs.follows = "nixpkgs";
     };
     darwin = {
-      url = "github:nix-darwin/nix-darwin/nix-darwin-25.11";
+      url = "github:nix-darwin/nix-darwin/nix-darwin-26.05";
       inputs.nixpkgs.follows = "nixpkgs";
     };
   };

homes/aarch64-darwin/evanreichard@mac-va-mbp-personal/default.nix

@@ -7,7 +7,7 @@ let
   inherit (lib.${namespace}) enabled;
 in
 {
-  home.stateVersion = "25.11";
+  home.stateVersion = "26.05";
 
   reichard = {
     user = {

homes/aarch64-darwin/evanreichard@mac-va-mbp-work/default.nix

@@ -8,7 +8,7 @@ let
   inherit (lib.${namespace}) enabled;
 in
 {
-  home.stateVersion = "25.11";
+  home.stateVersion = "26.05";
 
   reichard = {
     user = {
@@ -41,6 +41,7 @@ in
 
     services = {
       sketchybar = enabled;
+      open-proxy.server = enabled;
     };
 
     security = {

homes/aarch64-darwin/evanreichard@mac-va-mbp-work/vm-init.sh

@@ -33,6 +33,16 @@ else
     echo "  [✓] VM SOCKS Proxy Already Running"
 fi
 
+# Reverse tunnel so the VM's `open`/`xdg-open` reach open-proxy serve on this host.
+if ! pgrep -f "ssh -N -R 7777:127.0.0.1:7777 adios-cs" > /dev/null; then
+    echo "  [*] VM Open Proxy Starting..."
+    ssh -N -R 7777:127.0.0.1:7777 adios-cs &> /dev/null &
+    disown
+    echo "  [✓] VM Open Proxy Started"
+else
+    echo "  [✓] VM Open Proxy Already Running"
+fi
+
 echo -e "  [*] Connecting..."
 
 # Connect to VM

homes/aarch64-linux/evanreichard@lin-va-mbp-personal/default.nix

@@ -9,7 +9,7 @@ let
   inherit (lib.${namespace}) enabled;
 in
 {
-  home.stateVersion = "25.11";
+  home.stateVersion = "26.05";
 
   reichard = {
     user = {
@@ -20,7 +20,7 @@ in
     services = {
       ssh-agent = enabled;
       fusuma = enabled;
-      swww = enabled;
+      awww = enabled;
     };
 
     security = {
@@ -58,6 +58,7 @@ in
 
   home.packages = with pkgs; [
     orca-slicer
+    reichard.tuxguitar
   ];
 
   dconf = {

homes/aarch64-linux/evanreichard@lin-va-mbp-work-vm/default.nix

@@ -7,7 +7,7 @@ let
   inherit (lib.${namespace}) enabled;
 in
 {
-  home.stateVersion = "25.11";
+  home.stateVersion = "26.05";
 
   reichard = {
     user = {
@@ -17,6 +17,7 @@ in
 
     services = {
       ssh-agent = enabled;
+      open-proxy.client = enabled;
     };
 
     security = {
@@ -30,6 +31,7 @@ in
           enable = true;
           customFastFetchLogo = ./prophet.txt;
         };
+        conduit = enabled;
         btop = enabled;
         claude-code = enabled;
         direnv = enabled;

homes/aarch64-raw-efi/evanreichard@lin-terminal-image/default.nix

@@ -7,7 +7,7 @@ let
   inherit (lib.${namespace}) enabled;
 in
 {
-  home.stateVersion = "25.11";
+  home.stateVersion = "26.05";
 
   reichard = {
     user = {

homes/x86_64-linux/evanreichard@lin-va-desktop/default.nix

@@ -7,7 +7,7 @@ let
   inherit (lib.${namespace}) enabled;
 in
 {
-  home.stateVersion = "25.11";
+  home.stateVersion = "26.05";
 
   reichard = {
     user = {

homes/x86_64-linux/evanreichard@lin-va-terminal/default.nix

@@ -8,7 +8,7 @@ let
   inherit (lib.${namespace}) enabled;
 in
 {
-  home.stateVersion = "25.11";
+  home.stateVersion = "26.05";
 
   reichard = {
     user = {

homes/x86_64-linux/evanreichard@lin-va-thinkpad/default.nix

@@ -9,7 +9,7 @@ let
   inherit (lib.${namespace}) enabled;
 in
 {
-  home.stateVersion = "25.11";
+  home.stateVersion = "26.05";
 
   reichard = {
     user = {
@@ -20,7 +20,7 @@ in
     services = {
       ssh-agent = enabled;
       fusuma = enabled;
-      swww = enabled;
+      awww = enabled;
       poweralertd = enabled;
     };
 
@@ -68,6 +68,10 @@ in
     };
   };
 
+  home.packages = with pkgs; [
+    orca-slicer
+  ];
+
   home.pointerCursor = {
     gtk.enable = true;
     name = "catppuccin-macchiato-mauve-cursors";

homes/x86_64-linux/evanreichard@lin-va-utility/default.nix

@@ -20,7 +20,7 @@ in
     services = {
       ssh-agent = enabled;
       fusuma = enabled;
-      swww = enabled;
+      awww = enabled;
     };
 
     security = {

modules/home/common/default.nix

@@ -2,11 +2,13 @@
 {
 
   home.packages = with pkgs; [
-    sqlite-interactive
     jnv
     jq
+    mosh
     ncdu
-    ripgrep
     reichard.codexis
+    ripgrep
+    sqlite-interactive
+    unzip
   ];
 }

modules/home/programs/graphical/browsers/firefox/default.nix

@@ -1,4 +1,9 @@
-{ config, lib, pkgs, namespace, ... }:
+{ config
+, lib
+, pkgs
+, namespace
+, ...
+}:
 let
   inherit (lib)
     types
@@ -62,40 +67,47 @@ in
     extensions.packages = mkOpt (with lib.types; listOf package)
       (with pkgs.firefox-addons; [
         bitwarden
+        pkgs.firefox-addons."ctrl-number-to-switch-tabs"
         darkreader
         gruvbox-dark-theme
         kagi-search
         sponsorblock
         ublock-origin
-
-        # bypass-paywalls-clean
       ]) "Extensions to install";
   };
 
   config = mkIf cfg.enable {
     programs.firefox = {
       enable = true;
+      configPath = ".mozilla/firefox";
 
       inherit (cfg) policies;
 
       profiles = {
         ${config.${namespace}.user.name} = {
-          inherit (cfg) extraConfig extensions;
+          inherit (cfg) extraConfig;
           inherit (config.${namespace}.user) name;
 
+          extensions = {
+            packages = cfg.extensions.packages;
+            force = true;
+          };
+
           id = 0;
 
           settings = mkMerge [
             cfg.settings
             {
               "browser.aboutConfig.showWarning" = false;
+              "extensions.autoDisableScopes" = 0;
+              "extensions.activeThemeID" = "{eb8c4a94-e603-49ef-8e81-73d3c4cc04ff}";
               "browser.aboutwelcome.enabled" = false;
               "browser.sessionstore.warnOnQuit" = true;
               "browser.newtabpage.activity-stream.showSponsoredTopSites" = false;
               "browser.shell.checkDefaultBrowser" = false;
               "general.smoothScroll.msdPhysics.enabled" = true;
               "intl.accept_languages" = "en-US,en";
-              "ui.key.accelKey" = "224";
+              "ui.key.accelKey" = 91;
 
               # "devtools.chrome.enabled" = true;
               # "xpinstall.signatures.required" = false;

modules/home/programs/graphical/wms/hyprland/config/hyprland.conf

@@ -1,154 +0,0 @@
-exec-once = uwsm app -- waybar
-exec-once = uwsm app -- $terminal
-exec-once = uwsm app -- firefox
-
-general {
-    gaps_in = 5
-    gaps_out = 12
-
-    border_size = 2
-
-    # https://wiki.hyprland.org/Configuring/Variables/#variable-types for info about colors
-    col.active_border = rgba(33ccffee) rgba(00ff99ee) 45deg
-    col.inactive_border = rgba(595959aa)
-
-    # Set to true enable resizing windows by clicking and dragging on borders and gaps
-    resize_on_border = false
-
-    # Please see https://wiki.hyprland.org/Configuring/Tearing/ before you turn this on
-    allow_tearing = false
-
-    layout = dwindle
-}
-
-# https://wiki.hyprland.org/Configuring/Variables/#decoration
-decoration {
-    rounding = 10
-
-    active_opacity = 1.0
-    inactive_opacity = 1.0
-
-    shadow {
-        enabled = true
-        range = 4
-        render_power = 3
-        color = rgba(1a1a1aee)
-    }
-
-    blur {
-        enabled = true
-        size = 3
-        passes = 1
-        vibrancy = 0.1696
-    }
-}
-
-# https://wiki.hyprland.org/Configuring/Variables/#animations
-animations {
-    enabled = yes, please :)
-
-
-    bezier = easeOutQuint,0.23,1,0.32,1
-    bezier = easeInOutCubic,0.65,0.05,0.36,1
-    bezier = linear,0,0,1,1
-    bezier = almostLinear,0.5,0.5,0.75,1.0
-    bezier = quick,0.15,0,0.1,1
-
-    animation = global, 1, 10, default
-    animation = border, 1, 5.39, easeOutQuint
-    animation = windows, 1, 4.79, easeOutQuint
-    animation = windowsIn, 1, 4.1, easeOutQuint, popin 87%
-    animation = windowsOut, 1, 1.49, linear, popin 87%
-    animation = fadeIn, 1, 1.73, almostLinear
-    animation = fadeOut, 1, 1.46, almostLinear
-    animation = fade, 1, 3.03, quick
-    animation = layers, 1, 3.81, easeOutQuint
-    animation = layersIn, 1, 4, easeOutQuint, fade
-    animation = layersOut, 1, 1.5, linear, fade
-    animation = fadeLayersIn, 1, 1.79, almostLinear
-    animation = fadeLayersOut, 1, 1.39, almostLinear
-    animation = workspaces, 1, 1.94, almostLinear, fade
-    animation = workspacesIn, 1, 1.21, almostLinear, fade
-    animation = workspacesOut, 1, 1.94, almostLinear, fade
-}
-
-# Ref https://wiki.hyprland.org/Configuring/Workspace-Rules/
-# "Smart gaps" / "No gaps when only"
-# uncomment all if you wish to use that.
-# workspace = w[tv1], gapsout:0, gapsin:0
-# workspace = f[1], gapsout:0, gapsin:0
-# windowrulev2 = bordersize 0, floating:0, onworkspace:w[tv1]
-# windowrulev2 = rounding 0, floating:0, onworkspace:w[tv1]
-# windowrulev2 = bordersize 0, floating:0, onworkspace:f[1]
-# windowrulev2 = rounding 0, floating:0, onworkspace:f[1]
-
-# See https://wiki.hyprland.org/Configuring/Dwindle-Layout/ for more
-dwindle {
-    pseudotile = true # Master switch for pseudotiling. Enabling is bound to mainMod + P in the keybinds section below
-    preserve_split = true # You probably want this
-}
-
-# See https://wiki.hyprland.org/Configuring/Master-Layout/ for more
-master {
-    new_status = master
-}
-
-# https://wiki.hyprland.org/Configuring/Variables/#misc
-misc {
-    force_default_wallpaper = -1 # Set to 0 or 1 to disable the anime mascot wallpapers
-    disable_hyprland_logo = false # If true disables the random hyprland logo / anime girl background. :(
-}
-
-
-#############
-### INPUT ###
-#############
-
-# https://wiki.hyprland.org/Configuring/Variables/#input
-input {
-    kb_layout = us
-    kb_variant =
-    kb_model =
-    kb_options =
-    kb_rules =
-
-    follow_mouse = 1
-
-    sensitivity = 0.0 # -1.0 - 1.0, 0 means no modification.
-
-    touchpad {
-        scroll_factor = 0.5
-        disable_while_typing = true
-        natural_scroll = true
-        clickfinger_behavior = true
-        tap-to-click = false
-    }
-}
-
-# https://wiki.hyprland.org/Configuring/Variables/#gestures
-gesture = 4, horizontal, workspace, invert
-
-# Thinkpad Trackpoint
-device {
-    name = tpps/2-elan-trackpoint
-    sensitivity = -0.3
-}
-
-##############################
-### WINDOWS AND WORKSPACES ###
-##############################
-
-# See https://wiki.hyprland.org/Configuring/Window-Rules/ for more
-# See https://wiki.hyprland.org/Configuring/Workspace-Rules/ for workspace rules
-
-# Example windowrule v1
-# windowrule = float, ^(kitty)$
-
-# Example windowrule v2
-# windowrulev2 = float,class:^(kitty)$,title:^(kitty)$
-
-# Ignore maximize requests from apps. You'll probably like this.
-windowrulev2 = suppressevent maximize, class:.*
-
-# Fix some dragging issues with XWayland
-windowrulev2 = nofocus,class:^$,title:^$,xwayland:1,floating:1,fullscreen:0,pinned:0

modules/home/programs/graphical/wms/hyprland/config/hyprland.lua

@@ -0,0 +1,217 @@
+-- Hyprland config (lua backend, Hyprland 0.55+).
+-- `mainMod`, `menuMod`, and the monitor(s) are injected by Nix above this file.
+-- See https://wiki.hypr.land/Configuring/Start/
+
+local terminal = "ghostty"
+local menu     = "wofi --show drun"
+
+-------------------
+---- AUTOSTART ----
+-------------------
+
+hl.on("hyprland.start", function()
+    hl.exec_cmd("uwsm app -- waybar")
+    hl.exec_cmd("uwsm app -- " .. terminal)
+    hl.exec_cmd("uwsm app -- firefox")
+end)
+
+-----------------------
+---- LOOK AND FEEL ----
+-----------------------
+
+hl.config({
+    general = {
+        gaps_in  = 5,
+        gaps_out = 12,
+
+        border_size = 2,
+
+        col = {
+            active_border   = { colors = { "rgba(33ccffee)", "rgba(00ff99ee)" }, angle = 45 },
+            inactive_border = "rgba(595959aa)",
+        },
+
+        resize_on_border = false,
+        allow_tearing    = false,
+
+        layout = "dwindle",
+    },
+
+    decoration = {
+        rounding = 10,
+
+        active_opacity   = 1.0,
+        inactive_opacity = 1.0,
+
+        shadow = {
+            enabled      = true,
+            range        = 4,
+            render_power = 3,
+            color        = 0xee1a1a1a,
+        },
+
+        blur = {
+            enabled  = true,
+            size     = 3,
+            passes   = 1,
+            vibrancy = 0.1696,
+        },
+    },
+
+    animations = {
+        enabled = true,
+    },
+
+    dwindle = {
+        preserve_split = true,
+    },
+
+    master = {
+        new_status = "master",
+    },
+
+    misc = {
+        force_default_wallpaper = -1,
+        disable_hyprland_logo   = false,
+    },
+})
+
+----------------------
+---- ANIMATIONS ------
+----------------------
+
+hl.curve("easeOutQuint",   { type = "bezier", points = { { 0.23, 1 },    { 0.32, 1 } } })
+hl.curve("easeInOutCubic", { type = "bezier", points = { { 0.65, 0.05 }, { 0.36, 1 } } })
+hl.curve("linear",         { type = "bezier", points = { { 0, 0 },       { 1, 1 } } })
+hl.curve("almostLinear",   { type = "bezier", points = { { 0.5, 0.5 },   { 0.75, 1 } } })
+hl.curve("quick",          { type = "bezier", points = { { 0.15, 0 },    { 0.1, 1 } } })
+
+hl.animation({ leaf = "global",        enabled = true, speed = 10,   bezier = "default" })
+hl.animation({ leaf = "border",        enabled = true, speed = 5.39, bezier = "easeOutQuint" })
+hl.animation({ leaf = "windows",       enabled = true, speed = 4.79, bezier = "easeOutQuint" })
+hl.animation({ leaf = "windowsIn",     enabled = true, speed = 4.1,  bezier = "easeOutQuint", style = "popin 87%" })
+hl.animation({ leaf = "windowsOut",    enabled = true, speed = 1.49, bezier = "linear",       style = "popin 87%" })
+hl.animation({ leaf = "fadeIn",        enabled = true, speed = 1.73, bezier = "almostLinear" })
+hl.animation({ leaf = "fadeOut",       enabled = true, speed = 1.46, bezier = "almostLinear" })
+hl.animation({ leaf = "fade",          enabled = true, speed = 3.03, bezier = "quick" })
+hl.animation({ leaf = "layers",        enabled = true, speed = 3.81, bezier = "easeOutQuint" })
+hl.animation({ leaf = "layersIn",      enabled = true, speed = 4,    bezier = "easeOutQuint", style = "fade" })
+hl.animation({ leaf = "layersOut",     enabled = true, speed = 1.5,  bezier = "linear",       style = "fade" })
+hl.animation({ leaf = "fadeLayersIn",  enabled = true, speed = 1.79, bezier = "almostLinear" })
+hl.animation({ leaf = "fadeLayersOut", enabled = true, speed = 1.39, bezier = "almostLinear" })
+hl.animation({ leaf = "workspaces",    enabled = true, speed = 1.94, bezier = "almostLinear", style = "fade" })
+hl.animation({ leaf = "workspacesIn",  enabled = true, speed = 1.21, bezier = "almostLinear", style = "fade" })
+hl.animation({ leaf = "workspacesOut", enabled = true, speed = 1.94, bezier = "almostLinear", style = "fade" })
+
+---------------
+---- INPUT ----
+---------------
+
+hl.config({
+    input = {
+        kb_layout  = "us",
+        kb_variant = "",
+        kb_model   = "",
+        kb_options = "",
+        kb_rules   = "",
+
+        follow_mouse = 1,
+        sensitivity  = 0.0,
+
+        touchpad = {
+            scroll_factor        = 0.5,
+            disable_while_typing = true,
+            natural_scroll       = true,
+            clickfinger_behavior = true,
+            tap_to_click         = false,
+        },
+    },
+})
+
+-- 4-finger horizontal swipe to switch workspaces. The old `invert` modifier was
+-- removed in the 0.51 gesture rework; flip the physical swipe direction if needed.
+hl.gesture({ fingers = 4, direction = "horizontal", action = "workspace" })
+
+-- Thinkpad Trackpoint
+hl.device({ name = "tpps/2-elan-trackpoint", sensitivity = -0.3 })
+
+---------------------
+---- KEYBINDINGS ----
+---------------------
+
+-- Menu Mod Bindings (macOS Transition - Spotlight & Screenshots)
+hl.bind(menuMod .. " + SPACE",     hl.dsp.exec_cmd(menu))
+hl.bind(menuMod .. " + SHIFT + 1", hl.dsp.exec_cmd("hyprshot -m output"))
+hl.bind(menuMod .. " + SHIFT + 2", hl.dsp.exec_cmd("hyprshot -m window"))
+hl.bind(menuMod .. " + SHIFT + 3", hl.dsp.exec_cmd("hyprshot -m region"))
+hl.bind(menuMod .. " + Q",         hl.dsp.window.close())
+
+-- Primary Bindings
+hl.bind(mainMod .. " + RETURN", hl.dsp.exec_cmd(terminal))
+hl.bind(mainMod .. " + M",      hl.dsp.exec_cmd("uwsm stop"))
+hl.bind(mainMod .. " + V",      hl.dsp.window.float({ action = "toggle" }))
+hl.bind(mainMod .. " + P",      hl.dsp.window.pin())
+hl.bind(mainMod .. " + J",      hl.dsp.layout("togglesplit"))
+hl.bind(mainMod .. " + S",         hl.dsp.workspace.toggle_special("magic"))
+hl.bind(mainMod .. " + SHIFT + S", hl.dsp.window.move({ workspace = "special:magic" }))
+
+-- Window Focus
+hl.bind(mainMod .. " + left",  hl.dsp.focus({ direction = "left" }))
+hl.bind(mainMod .. " + right", hl.dsp.focus({ direction = "right" }))
+hl.bind(mainMod .. " + up",    hl.dsp.focus({ direction = "up" }))
+hl.bind(mainMod .. " + down",  hl.dsp.focus({ direction = "down" }))
+
+-- Workspace switch + move active window to workspace (1-9, 0 -> 10)
+for i = 1, 10 do
+    local key = i % 10
+    hl.bind(mainMod .. " + " .. key,         hl.dsp.focus({ workspace = i }))
+    hl.bind(mainMod .. " + SHIFT + " .. key, hl.dsp.window.move({ workspace = i }))
+end
+
+hl.bind(mainMod .. " + SHIFT + right", hl.dsp.focus({ workspace = "+1" }))
+hl.bind(mainMod .. " + SHIFT + left",  hl.dsp.focus({ workspace = "-1" }))
+
+-- Window move/resize with mouse
+hl.bind(mainMod .. " + mouse:272", hl.dsp.window.drag(),   { mouse = true })
+hl.bind(mainMod .. " + mouse:273", hl.dsp.window.resize(), { mouse = true })
+
+-- Multimedia & Brightness Keys
+hl.bind("XF86AudioRaiseVolume",  hl.dsp.exec_cmd("wpctl set-volume @DEFAULT_AUDIO_SINK@ 5%+"),    { locked = true, repeating = true })
+hl.bind("XF86AudioLowerVolume",  hl.dsp.exec_cmd("wpctl set-volume @DEFAULT_AUDIO_SINK@ 5%-"),    { locked = true, repeating = true })
+hl.bind("XF86AudioMute",         hl.dsp.exec_cmd("wpctl set-mute @DEFAULT_AUDIO_SINK@ toggle"),   { locked = true, repeating = true })
+hl.bind("XF86AudioMicMute",      hl.dsp.exec_cmd("wpctl set-mute @DEFAULT_AUDIO_SOURCE@ toggle"), { locked = true, repeating = true })
+hl.bind("XF86MonBrightnessUp",   hl.dsp.exec_cmd("brightnessctl s 4%+"),                          { locked = true, repeating = true })
+hl.bind("XF86MonBrightnessDown", hl.dsp.exec_cmd("brightnessctl s 5%-"),                          { locked = true, repeating = true })
+
+-- macOS Keyboard Brightness
+hl.bind(menuMod .. " + XF86MonBrightnessUp",   hl.dsp.exec_cmd("brightnessctl -d kbd_backlight s 10%+"), { locked = true, repeating = true })
+hl.bind(menuMod .. " + XF86MonBrightnessDown", hl.dsp.exec_cmd("brightnessctl -d kbd_backlight s 10%-"), { locked = true, repeating = true })
+
+-- Player Controls
+hl.bind("XF86AudioNext",  hl.dsp.exec_cmd("playerctl next"),       { locked = true })
+hl.bind("XF86AudioPause", hl.dsp.exec_cmd("playerctl play-pause"), { locked = true })
+hl.bind("XF86AudioPlay",  hl.dsp.exec_cmd("playerctl play-pause"), { locked = true })
+hl.bind("XF86AudioPrev",  hl.dsp.exec_cmd("playerctl previous"),   { locked = true })
+
+--------------------------------
+---- WINDOWS AND WORKSPACES ----
+--------------------------------
+
+hl.window_rule({
+    name           = "suppress-maximize-events",
+    match          = { class = ".*" },
+    suppress_event = "maximize",
+})
+
+hl.window_rule({
+    name  = "fix-xwayland-drags",
+    match = {
+        class      = "^$",
+        title      = "^$",
+        xwayland   = true,
+        float      = true,
+        fullscreen = false,
+        pin        = false,
+    },
+    no_focus = true,
+})

modules/home/programs/graphical/wms/hyprland/default.nix

@@ -24,90 +24,27 @@ in
 
     wayland.windowManager.hyprland = {
       enable = true;
-      extraConfig = builtins.readFile ./config/hyprland.conf;
-      settings = {
-        "$mainMod" = cfg.mainMod;
-        "$menuMod" = cfg.menuMod;
-        "$terminal" = "ghostty";
-        "$menu" = "wofi --show drun";
+      # Lua Backend - Hyprland 0.55 deprecated hyprlang and home-manager 26.05 defaults configType to "lua".
+      configType = "lua";
+      extraConfig =
+        let
+          # Quote unless the value is numeric, so scale can be `2` or `"auto"`.
+          luaScalar = v: if builtins.match "[0-9]+(\\.[0-9]+)?" v != null then v else ''"${v}"'';
+          mkMonitor =
+            s:
+            let
+              parts = map lib.trim (lib.splitString "," s);
+              field = i: if builtins.length parts > i then builtins.elemAt parts i else "";
+            in
+            ''hl.monitor({ output = "${field 0}", mode = "${field 1}", position = "${field 2}", scale = ${luaScalar (field 3)} })'';
+        in
+        ''
+          local mainMod = "${cfg.mainMod}"
+          local menuMod = "${cfg.menuMod}"
 
-        monitor = cfg.monitors;
-
-        bind = [
-          # Menu Mod Bindings (macOS Transition - Spotlight & Screenshots)
-          "$menuMod, SPACE, exec, $menu"
-          "$menuMod SHIFT, 1, exec, hyprshot -m output"
-          "$menuMod SHIFT, 2, exec, hyprshot -m window"
-          "$menuMod SHIFT, 3, exec, hyprshot -m region"
-          "$menuMod, Q, killactive"
-
-          # Primary Bindings
-          "$mainMod, RETURN, exec, $terminal"
-          "$mainMod, M, exit"
-          "$mainMod, V, togglefloating"
-          "$mainMod, P, pin"
-          "$mainMod, J, togglesplit"
-          "$mainMod, S, togglespecialworkspace, magic"
-          "$mainMod SHIFT, S, movetoworkspace, special:magic"
-
-          # Window Focus
-          "$mainMod, left, movefocus, l"
-          "$mainMod, right, movefocus, r"
-          "$mainMod, up, movefocus, u"
-          "$mainMod, down, movefocus, d"
-
-          # Workspace Switch
-          "$mainMod, 1, workspace, 1"
-          "$mainMod, 2, workspace, 2"
-          "$mainMod, 3, workspace, 3"
-          "$mainMod, 4, workspace, 4"
-          "$mainMod, 5, workspace, 5"
-          "$mainMod, 6, workspace, 6"
-          "$mainMod, 7, workspace, 7"
-          "$mainMod, 8, workspace, 8"
-          "$mainMod, 9, workspace, 9"
-          "$mainMod, 0, workspace, 10"
-
-          # Window Workspace Move
-          "$mainMod SHIFT, 1, movetoworkspace, 1"
-          "$mainMod SHIFT, 2, movetoworkspace, 2"
-          "$mainMod SHIFT, 3, movetoworkspace, 3"
-          "$mainMod SHIFT, 4, movetoworkspace, 4"
-          "$mainMod SHIFT, 5, movetoworkspace, 5"
-          "$mainMod SHIFT, 6, movetoworkspace, 6"
-          "$mainMod SHIFT, 7, movetoworkspace, 7"
-          "$mainMod SHIFT, 8, movetoworkspace, 8"
-          "$mainMod SHIFT, 9, movetoworkspace, 9"
-          "$mainMod SHIFT, 0, movetoworkspace, 10"
-          "$mainMod SHIFT, right, workspace, +1"
-          "$mainMod SHIFT, left, workspace, -1"
-        ];
-        bindm = [
-          # Window Resizing
-          "$mainMod, mouse:272, movewindow"
-          "$mainMod, mouse:273, resizewindow"
-        ];
-        bindel = [
-          # Multimedia & Brightness Keys
-          ",XF86AudioRaiseVolume, exec, wpctl set-volume @DEFAULT_AUDIO_SINK@ 5%+"
-          ",XF86AudioLowerVolume, exec, wpctl set-volume @DEFAULT_AUDIO_SINK@ 5%-"
-          ",XF86AudioMute, exec, wpctl set-mute @DEFAULT_AUDIO_SINK@ toggle"
-          ",XF86AudioMicMute, exec, wpctl set-mute @DEFAULT_AUDIO_SOURCE@ toggle"
-          ",XF86MonBrightnessUp, exec, brightnessctl s 4%+"
-          ",XF86MonBrightnessDown, exec, brightnessctl s 5%-"
-
-          # macOS Keyboard Brightness
-          "$menuMod, XF86MonBrightnessUp, exec, brightnessctl -d kbd_backlight s 10%+"
-          "$menuMod, XF86MonBrightnessDown, exec, brightnessctl -d kbd_backlight s 10%-"
-        ];
-        bindl = [
-          # Player Controls
-          ", XF86AudioNext, exec, playerctl next"
-          ", XF86AudioPause, exec, playerctl play-pause"
-          ", XF86AudioPlay, exec, playerctl play-pause"
-          ", XF86AudioPrev, exec, playerctl previous"
-        ];
-      };
+          ${lib.concatMapStringsSep "\n" mkMonitor cfg.monitors}
+        ''
+        + builtins.readFile ./config/hyprland.lua;
     };
 
     programs.waybar = {
@@ -117,9 +54,9 @@ in
         {
           layer = "top";
           position = "top";
-          mod = "dock";
+          mode = "dock";
           exclusive = true;
-          passtrough = false;
+          passthrough = false;
           gtk-layer-shell = true;
           height = 0;
           modules-left = [

modules/home/programs/terminal/bash/config/hey-intern.sh

@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
 
-MODEL="vllm-qwen3.6-27b-long-text"
+MODEL="qwen3.6-27b-vllm-180k-cuda0"
 SYSTEM_PROMPT="You are a shell command expert. Given a natural language query, generate a single shell command that accomplishes the task."
 
 # Colors

modules/home/programs/terminal/git/config/.gitignore

@@ -2,3 +2,5 @@ _scratch
 .direnv
 .envrc
 .agents
+.pi
+.nvim.lua

modules/home/programs/terminal/nvim/config/lua/diagnostics-config.lua

@@ -2,11 +2,7 @@
 local diagnostics_active = true
 local toggle_diagnostics = function()
 	diagnostics_active = not diagnostics_active
-	if diagnostics_active then
-		vim.diagnostic.enable()
-	else
-		vim.diagnostic.disable()
-	end
+	vim.diagnostic.enable(diagnostics_active)
 end
 
 local diagnostics_loclist_active = false

modules/home/programs/terminal/nvim/config/lua/llm-config.lua

@@ -1,6 +1,6 @@
 local llm_endpoint = "https://llm-api.va.reichard.io"
-local llm_assistant_model = "vllm-qwen3.6-27b-tools-text "
-local llm_infill_model = "qwen3.5-4b-thinking"
+local llm_assistant_model = "qwen3.6-27b-vllm-75k-cuda0"
+local llm_infill_model = "qwen3.5-4b-cuda1"
 local current_fim = "llama"
 
 -- Copilot Configuration

modules/home/programs/terminal/nvim/config/lua/lsp-config.lua

@@ -241,7 +241,12 @@ setup_lsp("gopls", {
 		})
 	end,
 	filetypes = { "go" },
-	cmd = { "gopls", "-remote=auto" },
+	cmd = function(dispatchers, config)
+		return vim.lsp.rpc.start({ "gopls", "-remote=auto" }, dispatchers, {
+			cwd = config.root_dir,
+			env = { GOMEMLIMIT = "6GiB" },
+		})
+	end,
 	settings = {
 		gopls = {
 			buildFlags = { "-tags=e2e" },
@@ -304,3 +309,30 @@ none_ls.setup({
 		end
 	end,
 })
+
+------------------------------------------------------
+---------------------- EXRC LSP ----------------------
+------------------------------------------------------
+
+vim.o.exrc = true -- native path: <cwd>/.nvim.lua on startup
+
+local loaded = {} -- absolute path -> true
+local function load_project_config(buf)
+	local fname = vim.api.nvim_buf_get_name(buf)
+	if fname == "" then return end
+	local found = vim.fs.find(".nvim.lua", {
+		upward = true,
+		path = vim.fs.dirname(fname),
+	})[1]
+	if not found or loaded[found] then return end
+	local content = vim.secure.read(found)
+	if content then
+		loaded[found] = true
+		local chunk, err = loadfile(found)
+		if chunk then chunk() else vim.notify("project config: " .. err, vim.log.levels.ERROR) end
+	end
+end
+
+vim.api.nvim_create_autocmd({ "BufReadPost", "BufNewFile" }, {
+	callback = function(args) load_project_config(args.buf) end,
+})

modules/home/programs/terminal/nvim/config/lua/lualine-config.lua

@@ -49,7 +49,7 @@ local function pr_status()
 end
 
 require("lualine").setup({
-	options = { theme = "catppuccin" },
+	options = { theme = "catppuccin-mocha" },
 	sections = {
 		lualine_c = { { pr_status } },
 		-- lualine_z = { require("opencode").statusline }

modules/home/programs/terminal/nvim/config/lua/ts-config.lua

@@ -1,4 +1,7 @@
-require("nvim-treesitter.configs").setup({
-	highlight = { enable = true, additional_vim_regex_highlighting = false },
-})
 vim.treesitter.language.register("markdown", "octo")
+
+vim.api.nvim_create_autocmd("FileType", {
+	callback = function(args)
+		pcall(vim.treesitter.start, args.buf)
+	end,
+})

modules/home/programs/terminal/nvim/default.nix

@@ -135,18 +135,18 @@ in
         golangci-lint-langserver
         lua-language-server
         nil
-        nodePackages.eslint
-        nodePackages.svelte-language-server
-        nodePackages.typescript
-        nodePackages.typescript-language-server
-        nodePackages.vscode-langservers-extracted
+        eslint
+        svelte-language-server
+        typescript
+        typescript-language-server
+        vscode-langservers-extracted
         pyright
         python312Packages.autopep8
 
         # Formatters
         luaformatter
         nixpkgs-fmt
-        nodePackages.prettier
+        prettier
         stylua
         sql-formatter
 
@@ -177,9 +177,9 @@ in
           clangd = "${pkgs.clang-tools}/bin/clangd",
           golintls = "${pkgs.golangci-lint-langserver}/bin/golangci-lint-langserver",
           luals = "${pkgs.lua-language-server}/bin/lua-language-server",
-          sveltels = "${pkgs.nodePackages.svelte-language-server}/bin/svelteserver",
-          tsls = "${pkgs.nodePackages.typescript-language-server}/bin/typescript-language-server",
-          vscls = "${pkgs.nodePackages.vscode-langservers-extracted}",
+          sveltels = "${pkgs.svelte-language-server}/bin/svelteserver",
+          tsls = "${pkgs.typescript-language-server}/bin/typescript-language-server",
+          vscls = "${pkgs.vscode-langservers-extracted}",
           sqls = "${pkgs.sqls}/bin/sqls",
         }
         return nix_vars

modules/home/programs/terminal/pi/config/AGENTS.md

@@ -1,82 +1,81 @@
 # AI Agent Guidelines
 
-## Important Rules
+Be cognizant of context use; this file is loaded for all LLMs. Keep guidance concise and high-signal.
 
-1. **Timeout for bash tool**: The `bash` tool MUST have a timeout specified. Without a timeout, the tool will hang indefinitely and cause the task to fail.
+## Critical Rules
 
-2. **File writing**: Do NOT use `cat` with heredocs to write files. Use the `write` tool instead (or `edit` for modifications).
+1. **Bash timeouts**: Every `bash` tool call MUST specify a timeout.
+   ```bash
+   bash(command="some command", timeout=30)
+   ```
 
-3. **Ephemeral files**: Put temporary scripts, plans, notes, and other scratch artifacts in `_scratch/`. It is gitignored, and reusable exploration/testing scripts should be iterated there instead of recreated repeatedly.
+2. **File writing**: Do NOT use `cat` with heredocs to write files. Use `write` for new/rewritten files and `edit` for targeted modifications.
 
-## Example of Correct Usage
+3. **Scratch files**: Put temporary scripts, plans, notes, and reusable exploration artifacts in `_scratch/`. It is gitignored.
 
-### Incorrect (will hang):
+4. **Missing commands**: If a tool is not installed, prefer `nix run` instead of installing it.
+   ```bash
+   nix run nixpkgs#python3 -- script.py
+   ```
 
-```bash
-bash(command="some long-running command")
-```
+## Asking Questions
 
-### Correct (with timeout):
+If a task is ambiguous, underspecified, or you foresee a non-obvious tradeoff during implementation, **surface it before coding** rather than guessing and producing rework. Treat this as always-on; an explicit "any questions?" is never required.
 
-```bash
-bash(command="some command", timeout=30)
-```
+## Context Discipline
 
-### Incorrect (file writing):
+Prefer a **search → targeted read** pattern:
 
-```bash
-bash(command="cat > file.txt << 'EOF'\ncontent\nEOF")
-```
+1. Search with `rg -n` / `grep -n` to find relevant line numbers.
+2. Read only the needed range with `read(path, offset, limit)`.
 
-### Correct (file writing):
-
-```bash
-write(path="file.txt", content="content")
-```
-
-## Reading Files
-
-Prefer a **search → targeted read** pattern to minimize context usage:
-
-1. **Search** with `grep -n` / `rg -n` to find relevant line numbers.
-2. **Read** only the needed range using `read(path, offset, limit)` or `sed -n 'X,Yp'`.
-
-```bash
-# Find the relevant lines
-bash(command="rg -n 'functionName' src/", timeout=10)
-
-# Read just that region (e.g. lines 42-70)
-read(path="src/foo.go", offset=42, limit=29)
-```
-
-Full-file reads are fine when genuinely needed (small files, needing full picture), but avoid them as the default reflex.
+Full-file reads are fine when genuinely needed, but avoid them as the default reflex.
 
 ## Principles
 
-1. **KISS / YAGNI** - Keep solutions simple and straightforward. Don't introduce abstractions, generics, or indirection unless there is a concrete, immediate need. Prefer obvious code over clever code.
+1. **Priority order**: When goals conflict, optimize in this order:
+   1. **Correctness** — solve the actual use case, including the realistic failure modes (not just the happy path).
+   2. **Maintainability / readability** — non-negotiable. Code is read far more than it is written; clarity wins over cleverness.
+   3. **Abstraction & polish** — only after the above are solid, and only when a concrete need justifies it.
 
-2. **Maintain AGENTS.md** - If the project has an `AGENTS.md`, keep it up to date as conventions or architecture evolve. However, follow the **BLUF** (Bottom Line Up Front) principle: keep it concise, actionable, and context-size conscious. Don't overload it with information that belongs in code comments or external docs.
+   All three matter, but never sacrifice (2) for (3). Prefer obvious, boring code over slick code that requires a paragraph to explain.
+
+2. **KISS / YAGNI**: Avoid abstractions, generics, or indirection unless there is a concrete, present need. Speculative flexibility is a maintainability tax.
+
+3. **Maintain AGENTS.md**: Keep project guidance up to date, but BLUF: concise, actionable, and context-size conscious.
+
+4. **Rephrase over append**: When extending existing content (docs, comments, prose, code), prefer rephrasing to capture the new intent over tacking on more verbosity.
+
+5. **Positive framing over prohibition**: State what _to_ do, not what _not_ to do. Default to omitting an instruction entirely rather than adding a "don't do X" rule — omission costs less context and avoids the failure mode where deleting a prohibition gets inverted into a mandate. Reserve explicit prohibitions for cases where the wrong behavior is a likely default that positive guidance alone can't redirect.
+
+6. **Knowledge Capture Check**: Before the final response, ask whether the task revealed a non-obvious convention, pitfall, repeatable workflow, or missing helper. If yes, briefly recommend exactly where to capture it: package/project AGENTS.md, global AGENTS.md, a skill, or a helper script. Skip this note when there is nothing meaningful.
 
 ## Style
 
 ### Comment Style
 
-A logical "block" of code (doesn't have to be a scope, but a cohesive group of statements responsible for something) should have a comment above it with a short "title". The title must be in **Title Case**. For example:
+Default to **no comment**. Code should be self-explanatory through naming and structure. Only add a comment when it earns its place by explaining something the code cannot.
+
+Write a comment when, and only when:
+
+1. The _why_ is non-obvious (intent, constraint, workaround, surprising invariant).
+2. A reader familiar with the language/codebase would otherwise stop and ask "why?".
+
+Do not narrate _what_ the code does. Do not add Title Case section headers over logical blocks just to label them.
+
+When a comment _is_ warranted, use a short Title Case label, a dash, and the _why_:
 
 ```go
-// Map Component Results
+// Map Component Results - Downstream consumers expect a name-keyed lookup.
 for _, comp := range components {
     results[comp.Name] = comp.Result
 }
 ```
 
-If the block is more complicated or non-obvious, explain _why_ it does what it does after the title:
+Rules for the explanation after the dash:
 
-```go
-// Map Component Results - This is needed because downstream consumers
-// expect a name-keyed lookup. Without it, the renderer would fall back
-// to O(n) scans on every frame.
-for _, comp := range components {
-    results[comp.Name] = comp.Result
-}
-```
+- Keep it to **2–3 sentences max**. Never a paragraph.
+- State the _why_ directly. Do not restate what the code does, recap prior context, or hedge.
+- Do **not** hard-wrap comments at 80 columns. Up to ~120 is fine.
+
+If a block is complex enough that it needs a heading just to be navigable, that is usually a signal to extract a well-named function instead.

modules/home/programs/terminal/pi/config/extensions/replace-pi-with-claude-code.ts

@@ -8,11 +8,10 @@ export default function replacePiWithClaudeCodeExtension(pi: ExtensionAPI) {
       return undefined;
     }
 
-    // Replace "pi" With "claude code" - Exclude Literal ".pi" (e.g. Paths)
-    // And "pi-coding-agent" (Package Name)
+    // Replace "pi" With "claude code"
     const transformedSystemPrompt = event.systemPrompt.replace(
-      /(?<!\.)pi(?!-coding-agent)/gi,
-      "claude code",
+      /(^|\s)pi(?![\w-])/gi,
+      "$1claude code",
     );
 
     if (transformedSystemPrompt === event.systemPrompt) {

modules/home/programs/terminal/pi/config/skills/create-skill/SKILL.md

@@ -14,6 +14,7 @@ Scaffold a new skill directory with a `SKILL.md` and optional helper scripts und
 ### 1. Gather Requirements
 
 Ask the user:
+
 - **What does the skill do?** (trigger conditions, purpose)
 - **Are there repeatable commands?** (if yes, these become scripts)
 
@@ -26,23 +27,29 @@ Create `skills/<skill-name>/SKILL.md` with this structure:
 ```markdown
 ---
 name: <skill-name>
-description: '<One-liner: what it does and when to trigger. Keep under ~200 chars.>'
+description: "<One-liner: what it does and when to trigger. Keep under ~200 chars.>"
 ---
 
 # <Skill Title>
 
 ## Overview
+
 [1-2 sentences on purpose and scope]
 
 ## Workflow
+
 [Numbered steps the agent follows]
 ```
 
 **Guidelines:**
+
 - **Be concise.** Skills are injected into agent context — every line costs tokens. Aim for the minimum needed to reliably guide the agent.
 - **Use scripts for repeatable logic.** If a step involves a multi-line shell command, `jq` pipeline, or API call that won't change between runs, put it in a `.sh` file next to `SKILL.md` and reference it from the workflow. See `address-gh-review/` for an example.
+- **Needs configurable values (paths, identifiers, etc.; not secrets — values are stored as plaintext files)?** Copy `assets/variable.sh` into the new skill's `scripts/` dir as-is. Callers use `variable.sh --get NAME [--require-exec RELPATH]`; the helper prints self-explaining `--set` instructions on "unset" or "set-but-invalid" and exits non-zero, so callers just propagate. The helper self-ignores its `.vars/` store on first `--set`, so no `.gitignore` setup is needed.
 - **Frontmatter is required.** `name` and `description` fields. The description is what the agent uses to decide whether to load the skill, so make it specific about trigger conditions.
 - **Don't over-specify.** Trust the agent to fill gaps. Document the _what_ and _when_, not every micro-step.
+- **Frame positively; omit rather than prohibit.** Write what the agent _should_ do. Prefer leaving a rule out over adding "don't do X" (see AGENTS.md principle: _Positive framing over prohibition_).
+- **Split workflow from reference when the reference surface grows.** If a skill accumulates lookup tables, mapping rules, or capability references that the workflow consults, move them into a sibling `<skill>/<category>/` directory (e.g. `mappings/`, `references/`) with one sub-doc per category and an index `README.md`. Keep `SKILL.md` focused on the hot path — workflow, hard rules, and a short table pointing at the sub-docs. Include a brief style guide in the index README covering (a) defer to authoritative sources (stubs, schemas, generated docs) whenever possible, (b) row/entry formatting conventions, (c) when to create a new sub-doc vs. extend an existing one.
 
 ### 3. Present for Review
 

modules/home/programs/terminal/pi/config/skills/create-skill/assets/variable.sh

@@ -0,0 +1,60 @@
+#!/usr/bin/env bash
+# Skill-local variable store. Values live in <skill-dir>/.vars/<NAME>.
+#
+# Usage:
+#   variable.sh --get NAME           # prints value to stdout, exits 0
+#                                    # or prints a self-explaining hint to
+#                                    # stderr and exits 2 if unset.
+#   variable.sh --set NAME VALUE     # writes value, exits 0.
+#
+# Callers should treat a non-zero exit as fatal; the stderr message tells
+# the caller (agent or user) exactly how to populate the missing value.
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+SKILL_DIR="$(dirname "$SCRIPT_DIR")"
+STORE="$SKILL_DIR/.vars"
+SELF="$0"
+
+usage() {
+    cat >&2 <<EOF
+Usage:
+  $SELF --get NAME
+  $SELF --set NAME VALUE
+EOF
+    exit 2
+}
+
+case "${1:-}" in
+    --get)
+        [[ $# -eq 2 ]] || usage
+        name="$2"
+        file="$STORE/$name"
+        if [[ ! -f "$file" ]]; then
+            cat >&2 <<EOF
+$SELF: $name is not set.
+Ask the user for the value, then set it:
+  $SELF --set $name <value>
+EOF
+            exit 2
+        fi
+        cat "$file"
+        ;;
+    --set)
+        [[ $# -eq 3 ]] || usage
+        name="$2"; value="$3"
+        [[ "$name" =~ ^[A-Z][A-Z0-9_]*$ ]] || {
+            echo >&2 "$SELF: invalid name '$name' (must match [A-Z][A-Z0-9_]*)"
+            exit 2
+        }
+        mkdir -p "$STORE"
+        # Self-ignore the store so values never get committed, even if the
+        # skill root lacks a .gitignore entry for .vars/.
+        [[ -f "$STORE/.gitignore" ]] || printf '*\n' > "$STORE/.gitignore"
+        printf '%s' "$value" > "$STORE/$name"
+        ;;
+    *)
+        usage
+        ;;
+esac

modules/home/programs/terminal/pi/config/skills/git-commit/SKILL.md

@@ -126,6 +126,8 @@ EOF
 - Imperative mood: "fix bug" not "fixes bug"
 - Reference issues: `Closes #123`, `Refs #456`
 - Keep description under 72 characters
+- Keep commit bodies short: 4 sentences max.
+- If you are unsure whether a body is useful, omit it entirely.
 
 ## Git Safety Protocol
 

modules/home/programs/terminal/pi/config/skills/web-glimpse/SKILL.md

@@ -1,112 +0,0 @@
----
-name: web-glimpse
-description: 'Search the web, read pages, extract content, run JavaScript, or capture screenshots using the `glimpse` headless browser tool. Use when the user asks to search the web, look something up online, read/fetch a page, inspect dynamic content, or capture visual state. Does not replace curl for simple HTTP/API requests.'
----
-
-# Web Browsing With Glimpse
-
-`glimpse` runs headless Firefox via WebDriver. Use it for web search, reading rendered pages, running JavaScript, and screenshots. Prefer `curl` for simple APIs, static files, and direct downloads.
-
-## Commands
-
-| Command | Purpose |
-| ------- | ------- |
-| `reader <url>` | Extract page content as Markdown (Reader View → raw fallback) |
-| `exec <url>` | Run JavaScript on a page, return the result |
-| `screenshot <url>` | Save a PNG screenshot |
-| `search <query>` | Search the web (Kagi) and return results |
-| `serve` | Start a persistent browser for faster repeat commands |
-
-## Persistent Server
-
-For multi-command sessions, start a persistent browser server first. All subsequent commands auto-discover it via Unix socket — no extra flags needed.
-
-```bash
-# Start persistent server (keeps geckodriver + Firefox alive)
-glimpse serve &
-
-# All commands now reuse the running browser (~300ms vs ~2-3s each)
-glimpse reader https://example.com
-glimpse reader https://other.com
-glimpse exec https://example.com --js='return document.title'
-
-# Check server status
-glimpse serve --status
-
-# Stop when done
-glimpse serve --stop
-```
-
-State (cookies, localStorage) persists between commands — this is intentional for sticky sessions. Without a running server, commands work normally with ad-hoc browser startup.
-
-## Quick Reference
-
-```bash
-# Read a page (tries Reader View, falls back to raw Turndown)
-glimpse reader https://example.com --timeout=15
-
-# Read without Reader View (raw HTML → Markdown via Turndown)
-glimpse reader https://example.com --no-reader --timeout=15
-
-# Get structured JSON instead of Markdown (includes method: "reader"|"raw")
-glimpse reader https://example.com --format=json
-
-# Save extracted content to a file
-glimpse reader https://example.com --output=page.md
-
-# Run JavaScript and return a value
-glimpse exec https://example.com --js='return document.title'
-
-# Extract specific data with JavaScript
-glimpse exec https://example.com --wait-until=complete --js='return {
-  title: document.title,
-  text: document.body.innerText.slice(0, 4000)
-}'
-
-# Wait for dynamic content before extracting
-glimpse reader https://example.com \
-  --wait-js='return document.querySelector(".content")?.innerText?.length > 100' \
-  --timeout=30
-
-# Capture a screenshot
-glimpse screenshot https://example.com --output=page.png
-
-# Search the web
-glimpse search "query terms" --timeout=15
-
-# Search and get JSON instead of Markdown
-glimpse search "query terms" --format=json
-```
-
-## Common Options
-
-| Option | Default | Purpose |
-| ------ | ------- | ------- |
-| `--timeout=<s>` | `10` | Max wait time in seconds; increase for slow/JS-heavy pages |
-| `--wait-until=<state>` | `none` | Wait for `none`, `interactive`, or `complete` |
-| `--wait-js=<code>` | — | Poll JS expression until truthy |
-| `--js=<code>` | — | Run inline JS before command logic |
-| `--script=<file>` | — | Run JS file before command logic |
-| `--no-headless` | — | Show the browser window |
-| `--format=<fmt>` | varies | Output format (reader: `markdown`/`html`/`text`/`json`; search: `markdown`/`json`) |
-| `--output=<file>` | — | Write output to file (reader, screenshot) |
-| `--no-reader` | — | Skip Reader View, use raw page extraction |
-
-## Workflow
-
-1. **Search first** when the user asks an open-ended question. Pick authoritative results to read.
-2. **Read pages with `reader`** — it tries Firefox Reader View for clean article extraction, then falls back to converting the raw page HTML to Markdown via Turndown. Most pages work without extra options.
-3. **Add `--wait-until=complete`** for JS-heavy pages, SPAs, or pages that load content dynamically.
-4. **Use `exec`** when you need targeted data extraction via JavaScript rather than full page content.
-5. **Use `screenshot`** when visual layout, charts, or rendering state matters.
-6. **Increase timeouts** — start at `15`, go to `30` for slow sites. The default `10` is often too tight for real-world pages.
-7. **Cite URLs** when summarizing web research. Distinguish search snippets from verified page content.
-
-## Error Handling
-
-| Error | Fix |
-| ----- | --- |
-| `TIMEOUT` | Increase `--timeout` (in seconds), add `--wait-until=complete`, or use `--wait-js` |
-| `USAGE_ERROR` | Check arg order: `glimpse <command> <url>`, search is `glimpse search "query"` |
-| Thin/empty content | Try `--wait-until=complete`, `--no-reader`, or targeted `exec` |
-| Search auth errors | Kagi token is configured via `~/.config/glimpse/config.json` or `KAGI_TOKEN` env |

modules/home/programs/terminal/pi/config/subagents/scout.md

@@ -0,0 +1,10 @@
+---
+name: scout
+description: Fast codebase reconnaissance. Reads and searches files, then returns concise findings with paths.
+approved_tools:
+  - read
+  - bash
+---
+You are a focused codebase scout running as a subagent with isolated context.
+
+Find the information requested by the task and return a concise report. Prefer exact file paths, symbol names, and line numbers. Do not modify files.

modules/home/programs/terminal/pi/default.nix

@@ -1,9 +1,8 @@
-{
-  lib,
-  pkgs,
-  config,
-  namespace,
-  ...
+{ lib
+, pkgs
+, config
+, namespace
+, ...
 }:
 let
   inherit (lib) mkIf;
@@ -18,6 +17,8 @@ let
   # writing other fields (current model, etc.) without us clobbering them.
   piPackages = [
     "https://gitea.va.reichard.io/evan/pi-lsp.git@main"
+    "https://gitea.va.reichard.io/evan/pi-web.git@main"
+    "https://gitea.va.reichard.io/evan/pi-subagents.git@main"
     "https://gitea.va.reichard.io/evan/pi-statusline.git@main"
   ];
 
@@ -35,15 +36,19 @@ let
   ];
 
   piAuthJqRawfiles = lib.concatStringsSep " \\\n          " (
-    map (
+    map
+      (
         auth: ''--rawfile ${auth.jqVar} "${config.sops.secrets.${auth.secretName}.path}"''
-    ) piAuthApiKeys
+      )
+      piAuthApiKeys
   );
 
   piAuthJqFilter = lib.concatStringsSep " | " (
-    map (
+    map
+      (
         auth: ''.["${auth.provider}"] = { type: "api_key", key: ($'' + auth.jqVar + ''| rtrimstr("\n")) }''
-    ) piAuthApiKeys
+      )
+      piAuthApiKeys
   );
 
   piAuthMergeScript = pkgs.writeShellScript "pi-auth-merge" ''
@@ -89,7 +94,6 @@ in
 
     # Define Pi Configuration
     home.file = {
-
       ".pi/agent/AGENTS.md" = {
         source = ./config/AGENTS.md;
       };
@@ -97,6 +101,10 @@ in
         source = ./config/skills;
         recursive = true;
       };
+      ".pi/agent/subagents" = {
+        source = ./config/subagents;
+        recursive = true;
+      };
       ".pi/agent/prompts" = {
         source = ./config/prompts;
         recursive = true;
@@ -116,11 +124,24 @@ in
         };
       }
       // lib.listToAttrs (
-        map (auth: {
+        map
+          (auth: {
             name = auth.secretName;
             value.sopsFile = auth.sopsFile;
-        }) piAuthApiKeys
+          })
+          piAuthApiKeys
       );
+      # Pi Web Config - Sops template so the kagi token (declared by the
+      # glimpse module, which pi enables above) can be embedded alongside
+      # the non-secret searxng base URL.
+      templates."pi-web.json" = {
+        path = "${config.home.homeDirectory}/.pi/pi-web/config.json";
+        content = builtins.toJSON {
+          provider = "searxng";
+          kagi.token = "${config.sops.placeholder.kagi_token}";
+          searxng.baseUrl = "https://search.va.reichard.io";
+        };
+      };
       templates."pi-models.json" = {
         path = "${config.home.homeDirectory}/.pi/agent/models.json";
         content = builtins.toJSON {

modules/home/services/awww/default.nix

@@ -5,23 +5,23 @@
 , ...
 }:
 let
-  cfg = config.${namespace}.services.swww;
+  cfg = config.${namespace}.services.awww;
 in
 {
-  options.${namespace}.services.swww = {
-    enable = lib.mkEnableOption "swww wallpaper service";
+  options.${namespace}.services.awww = {
+    enable = lib.mkEnableOption "awww wallpaper service";
   };
 
   config = lib.mkIf cfg.enable {
     home.packages = with pkgs; [
-      swww
+      awww
     ];
 
     systemd.user = {
       services = {
-        swww-daemon = {
+        awww-daemon = {
           Unit = {
-            Description = "SWWW Wallpaper Daemon";
+            Description = "AWWW Wallpaper Daemon";
             After = [ "graphical-session.target" ];
           };
 
@@ -31,7 +31,7 @@ in
 
           Service = {
             Type = "simple";
-            ExecStart = "${pkgs.swww}/bin/swww-daemon";
+            ExecStart = "${pkgs.awww}/bin/awww-daemon";
             Restart = "on-failure";
             RestartSec = 5;
           };
@@ -39,28 +39,28 @@ in
 
         change-wallpaper = {
           Unit = {
-            Description = "SWWW Wallpaper Changer";
-            After = [ "swww-daemon.service" ];
-            Requires = [ "swww-daemon.service" ];
+            Description = "AWWW Wallpaper Changer";
+            After = [ "awww-daemon.service" ];
+            Requires = [ "awww-daemon.service" ];
           };
 
           Install = {
-            WantedBy = [ "swww-daemon.service" ];
+            WantedBy = [ "awww-daemon.service" ];
           };
 
           Service = {
             Type = "oneshot";
             ExecStart = "${pkgs.writeShellScript "change-wallpaper-script" ''
               WALLPAPER=$(${pkgs.findutils}/bin/find $HOME/Wallpapers -type f | ${pkgs.coreutils}/bin/shuf -n 1)
-              ${pkgs.swww}/bin/swww img "$WALLPAPER" --transition-type random
+              ${pkgs.awww}/bin/awww img "$WALLPAPER" --transition-type random
             ''}";
           };
         };
       };
 
-      timers.swww-schedule = {
+      timers.awww-schedule = {
         Unit = {
-          Description = "SWWW Wallpaper Schedule";
+          Description = "AWWW Wallpaper Schedule";
         };
 
         Install = {

modules/home/services/open-proxy/default.nix

@@ -0,0 +1,76 @@
+{ config
+, lib
+, pkgs
+, namespace
+, ...
+}:
+let
+  inherit (lib) mkIf mkEnableOption;
+  cfg = config.${namespace}.services.open-proxy;
+  package = pkgs.reichard.open-proxy;
+  secretName = "open_proxy_token";
+  tokenPath = config.sops.secrets.${secretName}.path;
+in
+{
+  options.${namespace}.services.open-proxy = {
+    server.enable = mkEnableOption "open-proxy host server (opens forwarded URLs/files on this machine)";
+    client.enable = mkEnableOption "open-proxy client (shadows open/xdg-open to forward to the host)";
+  };
+
+  config = lib.mkMerge [
+    (mkIf (cfg.server.enable || cfg.client.enable) {
+      sops.secrets.${secretName} = {
+        sopsFile = lib.snowfall.fs.get-file "secrets/common/evanreichard.yaml";
+      };
+    })
+
+    (mkIf cfg.server.enable {
+      assertions = [
+        {
+          assertion = pkgs.stdenv.isDarwin;
+          message = "reichard.services.open-proxy.server is only supported on macOS (Darwin).";
+        }
+      ];
+
+      launchd.agents.open-proxy = {
+        enable = true;
+        config = {
+          Label = "io.reichard.open-proxy";
+          ProgramArguments = [ "${package}/bin/open-proxy" "serve" ];
+          RunAtLoad = true;
+          KeepAlive = true;
+          EnvironmentVariables = {
+            OPEN_PROXY_TOKEN_FILE = tokenPath;
+            # open(1) lives in /usr/bin; launchd agents don't inherit a login PATH.
+            PATH = "/usr/bin:/bin:/usr/sbin:/sbin";
+          };
+          StandardOutPath = "${config.home.homeDirectory}/Library/Logs/open-proxy/open-proxy.out.log";
+          StandardErrorPath = "${config.home.homeDirectory}/Library/Logs/open-proxy/open-proxy.err.log";
+        };
+      };
+    })
+
+    (mkIf cfg.client.enable {
+      assertions = [
+        {
+          assertion = pkgs.stdenv.isLinux;
+          message = "reichard.services.open-proxy.client is only supported on Linux.";
+        }
+      ];
+
+      # Shadow the openers via ~/.local/bin (prepended to PATH below). open-proxy
+      # keys off argv[0], so these symlinks run in client mode and fall back to
+      # any real opener further down PATH when the host is unreachable.
+      home.file = {
+        ".local/bin/open".source = "${package}/bin/open-proxy";
+        ".local/bin/xdg-open".source = "${package}/bin/open-proxy";
+      };
+
+      home.sessionPath = [ "$HOME/.local/bin" ];
+      home.sessionVariables = {
+        BROWSER = "open";
+        OPEN_PROXY_TOKEN_FILE = tokenPath;
+      };
+    })
+  ];
+}

modules/nixos/hardware/opengl/default.nix

@@ -16,6 +16,12 @@ in
     enable32Bit = mkBoolOpt false "enable 32-bit";
     enableIntel = mkBoolOpt false "support for intel";
     enableNvidia = mkBoolOpt false "support for nvidia";
+    nvidiaPackage = lib.mkOption {
+      type = lib.types.package;
+      default = config.boot.kernelPackages.nvidiaPackages.stable;
+      defaultText = "config.boot.kernelPackages.nvidiaPackages.stable";
+      description = "nvidia driver package; pin to legacy_580 for Pascal (GTX 10xx) and older";
+    };
   };
 
   config = mkIf cfg.enable {
@@ -32,7 +38,7 @@ in
 
     # Enable Nvidia Hardware
     hardware.nvidia = mkIf cfg.enableNvidia {
-      package = config.boot.kernelPackages.nvidiaPackages.stable;
+      package = cfg.nvidiaPackage;
       modesetting.enable = true;
       powerManagement.enable = true;
       open = false;

modules/nixos/nix/default.nix

@@ -38,6 +38,7 @@ in
           sshUser = "evanreichard";
           protocol = "ssh";
           sshKey = config.sops.secrets.builder_ssh_key.path;
+          publicHostKey = "c3NoLWVkMjU1MTkgQUFBQUMzTnphQzFsWkRJMU5URTVBQUFBSUdscEMwcm9yQVRLeks4bUxNS2dDWXFNNU4yTi9HZ1MydDRNMTNjd25BT1M=";
           supportedFeatures = [
             "benchmark"
             "big-parallel"

modules/nixos/services/llama-swap/AGENTS.md

@@ -1,8 +1,12 @@
 # llama-swap Module — Agent Guide
 
+## Model ID Convention
+
+Use `<family>-<size>[-backend/variant][-context][-vl]-<placement>`. Omit `thinking` from IDs, use `vl` for vision-language models, and keep placement as the final suffix (`cuda0`, `cuda1`, or `dual`). Keep quantization and richer behavior details in the display `name` unless they are needed to distinguish two active configs for the same family/placement.
+
 ## Syncing vLLM Configs from club-3090
 
-The three vLLM model configs in `config.nix` (`vllm-qwen3.6-27b-long-text`, `vllm-qwen3.6-27b-long-vision`, `vllm-qwen3.6-27b-tools-text`) are derived from the club-3090 repo's Docker Compose files. Each config block has a `Synced from:` comment with the commit hash it was last aligned to.
+The three vLLM model configs in `config.nix` (`qwen3.6-27b-vllm-180k-cuda0`, `qwen3.6-27b-vllm-145k-vl-cuda0`, `qwen3.6-27b-vllm-75k-cuda0`) are derived from the club-3090 repo's Docker Compose files. Each config block has a `Synced from:` comment with the commit hash it was last aligned to.
 
 ### Source Files
 
@@ -10,9 +14,9 @@ The upstream compose files live at https://github.com/noonghunna/club-3090 under
 
 | config.nix model ID               | Compose file                        |
 |------------------------------------|-------------------------------------|
-| `vllm-qwen3.6-27b-long-text`      | `docker-compose.long-text.yml`      |
-| `vllm-qwen3.6-27b-long-vision`    | `docker-compose.long-vision.yml`    |
-| `vllm-qwen3.6-27b-tools-text`     | `docker-compose.tools-text.yml`     |
+| `qwen3.6-27b-vllm-180k-cuda0`    | `docker-compose.long-text.yml`      |
+| `qwen3.6-27b-vllm-145k-vl-cuda0` | `docker-compose.long-vision.yml`    |
+| `qwen3.6-27b-vllm-75k-cuda0`     | `docker-compose.tools-text.yml`     |
 
 ### Sync Process
 
@@ -28,7 +32,7 @@ The upstream compose files live at https://github.com/noonghunna/club-3090 under
    - Genesis env vars — the full set grows frequently; add new ones, remove deprecated ones
    - Sidecar patches — old patches get absorbed into Genesis; drop them from entrypoint + volume mounts
    - Docker image tag — update when the compose files move to a new nightly
-4. **Keep `patch_timings_07351e088.py`** — this is our own patch, not from club-3090. Always retain it in the entrypoint and volume mounts.
+4. **Keep `patch_timings_1acd67a.py`** — this is our own patch, not from club-3090. Always retain it in the entrypoint and volume mounts.
 5. **Update the `Synced from:` comment** on each config block with the new commit hash and date.
 6. **Update `setup-qwen36-vllm.sh`** if the upstream `patches/` directory changed (new patches added, old ones removed). The setup script downloads sidecar patches and creates cache directories.
 7. **Verify syntax**: `nix-instantiate --parse config.nix`

modules/nixos/services/llama-swap/config.nix

@@ -1,6 +1,7 @@
 { pkgs }:
 let
   llama-cpp = pkgs.reichard.llama-cpp;
+  ik-llama-cpp = pkgs.reichard.ik-llama-cpp;
   stable-diffusion-cpp = pkgs.reichard.stable-diffusion-cpp.override {
     cudaSupport = true;
   };
@@ -13,8 +14,8 @@ in
     # ---------------------------------------
 
     # https://huggingface.co/mradermacher/gpt-oss-20b-heretic-v2-i1-GGUF/tree/main
-    "gpt-oss-20b-thinking" = {
-      name = "GPT OSS (20B) - Thinking";
+    "gpt-oss-20b-cuda0" = {
+      name = "GPT OSS 20B (CUDA0)";
       macros.ctx = "131072";
       cmd = ''
         ${llama-cpp}/bin/llama-server \
@@ -32,20 +33,24 @@ in
     };
 
     # https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF/tree/main
-    "qwen3.6-35b-thinking" = {
-      name = "Qwen3.6 (35B) - Thinking";
+    "qwen3.6-35b-cuda0" = {
+      name = "Qwen3.6 35B (CUDA0, UD-IQ4)";
       macros.ctx = "262144";
       cmd = ''
         ${llama-cpp}/bin/llama-server \
           --port ''${PORT} \
-          -m /mnt/ssd/Models/Qwen3.6/Qwen3.6-35B-A3B-UD-IQ4_XS.gguf \
+          -m /mnt/ssd/Models/Qwen3.6/Qwen3.6-35B-A3B-UD-IQ4_NL.gguf \
           -c ''${ctx} \
+          -np 2 -kvu \
           --temp 0.6 \
           --top-p 0.95 \
           --top-k 20 \
           --min-p 0.0 \
           --presence-penalty 0.0 \
-          --repeat-penalty 1.0 \
+          -ctk q8_0 \
+          -ctv q8_0 \
+          --spec-type draft-mtp \
+          --spec-draft-n-max 3 \
           -dev CUDA0 \
           -fit off \
           --chat-template-kwargs "{\"preserve_thinking\": true}"
@@ -58,37 +63,126 @@ in
       };
     };
 
-    # https://huggingface.co/unsloth/Qwen3.6-27B-GGUF/tree/main
-    "qwen3.6-27b-thinking" = {
-      name = "Qwen3.6 (27B) - Thinking";
-      macros.ctx = "196608";
+    # https://huggingface.co/ubergarm/Qwen3.6-27B-GGUF/tree/main
+    "qwen3.6-27b-ik-cuda0" = {
+      name = "Qwen3.6 (27B) (CUDA0, IQ4_KS)";
+      macros.ctx = "156000";
+      env = [ "CUDA_VISIBLE_DEVICES=0" ];
+      cmd = ''
+        ${ik-llama-cpp}/bin/llama-server \
+          --port ''${PORT} \
+          -m /mnt/ssd/Models/Qwen3.6/Qwen3.6-27B-MTP-IQ4_KS.gguf \
+          -c ''${ctx} -ctk q8_0 -ctv q8_0 -ngl 99 \
+          -mtp --draft-max 4 --draft-p-min 0.75 \
+          -muge -mqkv -cram 32768 --ctx-checkpoints 32 \
+          --jinja --chat-template-kwargs '{"preserve_thinking":true}'
+      '';
+      metadata = {
+        type = [
+          "text-generation"
+          "coding"
+        ];
+      };
+    };
+
+    # https://huggingface.co/unsloth/Qwen3.6-27B-GGUF-MTP/tree/main
+    "qwen3.6-27b-cuda0" = {
+      name = "Qwen3.6 27B (CUDA0, UD-Q4)";
+      macros.ctx = "110000";
       cmd = ''
         ${llama-cpp}/bin/llama-server \
           --port ''${PORT} \
-          -m /mnt/ssd/Models/Qwen3.6/Qwen3.6-27B-IQ4_XS.gguf \
+          -m /mnt/ssd/Models/Qwen3.6/Qwen3.6-27B-UD-Q4_K_XL.gguf \
           -c ''${ctx} \
-          --parallel 2 \
+          -np 2 -kvu \
           --temp 0.6 \
           --top-p 0.95 \
           --top-k 20 \
           --min-p 0.00 \
-          --presence-penalty 1.5 \
+          --presence-penalty 0.0 \
           -ctk q8_0 \
           -ctv q8_0 \
-          --keep 3000 \
-          --batch-size 4096 \
-          --ubatch-size 1024 \
-          --spec-type ngram-mod \
-          --spec-ngram-mod-n-match 24 \
-          --spec-draft-n-min 16 \
-          --spec-draft-n-max 64 \
+          --spec-type draft-mtp \
+          --spec-draft-n-max 3 \
           -dev CUDA0 \
           -fit off \
           --chat-template-kwargs "{\"preserve_thinking\": true}"
       '';
-      # --chat-template-kwargs "{\"enable_thinking\": false}"
-      # --spec-draft-n-min 16 \
-      # --spec-draft-n-max 32 \
+      metadata = {
+        type = [
+          "text-generation"
+          "coding"
+        ];
+      };
+    };
+
+    # https://huggingface.co/unsloth/gemma-4-26B-A4B-it-GGUF/tree/main
+    "gemma-4-26b-vl-cuda0" = {
+      name = "Gemma 4 26B (VL, CUDA0)";
+      macros.ctx = "196608";
+      cmd = ''
+        ${llama-cpp}/bin/llama-server \
+          --port ''${PORT} \
+          -m /mnt/ssd/Models/Gemma/gemma-4-26B-A4B-it-UD-Q4_K_XL.gguf \
+          --mmproj /mnt/ssd/Models/Gemma/mmproj-BF16_gemma-4-26B-A4B-it-UD-Q4_K_XL.gguf \
+          -c ''${ctx} \
+          --parallel 1 \
+          --spec-type ngram-mod \
+          --spec-ngram-mod-n-match 24 \
+          --spec-ngram-mod-n-min 48 \
+          --spec-ngram-mod-n-max 64 \
+          --temp 1.0 \
+          --top-k 64 \
+          --top-p 0.95 \
+          --no-warmup \
+          --jinja \
+          -fit off \
+          -dev CUDA0
+      '';
+      metadata = {
+        type = [
+          "text-generation"
+          "vision"
+        ];
+      };
+    };
+
+    # https://huggingface.co/Lorbus/Qwen3.6-27B-int4-AutoRound
+    "qwen3.6-27b-vllm-50k-cuda0" = {
+      name = "Qwen3.6 27B (vLLM, 50K, CUDA0)";
+      checkEndpoint = "/v1/models";
+      macros.ctx = "50000";
+      proxy = "http://127.0.0.1:\${PORT}";
+      cmd = ''
+        ${pkgs.docker}/bin/docker run --rm --device=nvidia.com/gpu=all \
+          --name ''${MODEL_ID} \
+          -e CUDA_DEVICE_ORDER=PCI_BUS_ID \
+          -e CUDA_VISIBLE_DEVICES=0 \
+          -e VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0 \
+          -v /mnt/ssd/vLLM/Models:/root/.cache/huggingface \
+          -p ''${PORT}:8000 \
+          vllm/vllm-openai:latest \
+          /root/.cache/huggingface/qwen3.6-27b-autoround-int4 \
+          --served-model-name ''${MODEL_ID} \
+          --quantization auto_round \
+          --dtype float16 \
+          --tensor-parallel-size 1 \
+          --gpu-memory-utilization 0.97 \
+          --max-model-len ''${ctx} \
+          --max-num-seqs 1 \
+          --max-num-batched-tokens 4128 \
+          --kv-cache-dtype fp8_e5m2 \
+          --enable-chunked-prefill \
+          --enable-prefix-caching \
+          --speculative-config '{"method":"mtp","num_speculative_tokens":3}' \
+          --enable-auto-tool-choice \
+          --tool-call-parser qwen3_coder \
+          --trust-remote-code \
+          --default-chat-template-kwargs '{"enable_thinking": false}' \
+          --host 0.0.0.0 \
+          --port 8000
+      '';
+      cmdStop = "${pkgs.docker}/bin/docker stop \${MODEL_ID}";
       metadata = {
         type = [
           "text-generation"
@@ -98,11 +192,240 @@ in
     };
 
     # https://github.com/noonghunna/club-3090/tree/master/models/qwen3.6-27b/vllm
-    # Synced from: club-3090 f6613c8 (2026-05-02) — docker-compose.long-text.yml
-    # Long-text variant - 180K context, text-only (no vision)
-    # TurboQuant 3-bit KV + MTP n=3 + Genesis v7.69 + Cliff 2 closure recipe
-    "vllm-qwen3.6-27b-long-text" = {
-      name = "vLLM Qwen3.6 (27B) - Long Text";
+    "qwen3.6-27b-vllm-75k-cuda0" = {
+      name = "Qwen3.6 27B (vLLM, 75K, CUDA0)";
+      checkEndpoint = "/v1/models";
+      macros.ctx = "75000";
+      proxy = "http://127.0.0.1:\${PORT}";
+      cmd =
+        let
+          vllmCmd = ''
+            set -e; pip install xxhash pandas scipy -q;
+            python3 -m vllm._genesis.patches.apply_all;
+            python3 /patches/patch_timings_1acd67a.py;
+            exec vllm serve ''${VLLM_ENFORCE_EAGER:+--enforce-eager}
+            --served-model-name ''${MODEL_ID}
+            --model /root/.cache/huggingface/qwen3.6-27b-autoround-int4
+            --quantization auto_round
+            --dtype float16
+            --tensor-parallel-size 1
+            --max-model-len ''${ctx}
+            --gpu-memory-utilization 0.97
+            --max-num-seqs 1
+            --max-num-batched-tokens 2048
+            --kv-cache-dtype fp8_e5m2
+            --language-model-only
+            --trust-remote-code
+            --reasoning-parser qwen3
+            --enable-auto-tool-choice
+            --tool-call-parser qwen3_coder
+            --chat-template /templates/chat_template.jinja
+            --enable-prefix-caching
+            --enable-chunked-prefill
+            --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":3}'
+            --host 0.0.0.0
+            --port 8000
+          '';
+          vllmCmdFlat = builtins.replaceStrings [ "\n" ] [ " " ] vllmCmd;
+        in
+        ''
+          ${pkgs.docker}/bin/docker run --rm --device=nvidia.com/gpu=all \
+            --name ''${MODEL_ID} \
+            --ipc=host \
+            -e CUDA_DEVICE_MAX_CONNECTIONS=8 \
+            -e CUDA_DEVICE_ORDER=PCI_BUS_ID \
+            -e CUDA_VISIBLE_DEVICES=0 \
+            -e GENESIS_ENABLE_P58_ASYNC_PLACEHOLDER_FIX=1 \
+            -e GENESIS_ENABLE_P64_QWEN3CODER_MTP_STREAMING=1 \
+            -e GENESIS_ENABLE_P66_CUDAGRAPH_SIZE_FILTER=1 \
+            -e GENESIS_ENABLE_P68_AUTO_FORCE_TOOL=1 \
+            -e GENESIS_ENABLE_P69_LONG_CTX_TOOL_REMINDER=1 \
+            -e GENESIS_ENABLE_P72_PROFILE_RUN_CAP=1 \
+            -e GENESIS_ENABLE_P74_CHUNK_CLAMP=1 \
+            -e GENESIS_ENABLE_P94=1 \
+            -e GENESIS_ENABLE_PN13_CUDA_GRAPH_LAMBDA_ARITY=1 \
+            -e GENESIS_ENABLE_PN14_TQ_DECODE_OOB_CLAMP=1 \
+            -e GENESIS_ENABLE_PN17_FA2_LSE_CLAMP=1 \
+            -e GENESIS_ENABLE_PN19_SCOPED_MAX_SPLIT=1 \
+            -e GENESIS_ENABLE_PN59_STREAMING_GDN=1 \
+            -e GENESIS_ENABLE_PN8_MTP_DRAFT_ONLINE_QUANT=1 \
+            -e GENESIS_P68_P69_LONG_CTX_THRESHOLD_CHARS=50000 \
+            -e GENESIS_PROFILE_RUN_CAP_M=4128 \
+            -e NCCL_CUMEM_ENABLE=0 \
+            -e NCCL_P2P_DISABLE=1 \
+            -e OMP_NUM_THREADS=1 \
+            -e PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True,max_split_size_mb:512 \
+            -e VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 \
+            -e VLLM_FLOAT32_MATMUL_PRECISION=high \
+            -e VLLM_MARLIN_USE_ATOMIC_ADD=1 \
+            -e VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0 \
+            -e VLLM_NO_USAGE_STATS=1 \
+            -e VLLM_USE_FLASHINFER_SAMPLER=1 \
+            -e VLLM_WORKER_MULTIPROC_METHOD=spawn \
+            -e VLLM_ENFORCE_EAGER \
+            -v /mnt/ssd/vLLM/Models:/root/.cache/huggingface \
+            -v /mnt/ssd/vLLM/Patches/genesis/vllm/_genesis:/usr/local/lib/python3.12/dist-packages/vllm/_genesis:ro \
+            -v /mnt/ssd/vLLM/Patches/patch_timings_1acd67a.py:/patches/patch_timings_1acd67a.py:ro \
+            -v /mnt/ssd/vLLM/Templates/chat_template-v11.jinja:/templates/chat_template.jinja \
+            -p ''${PORT}:8000 \
+            --entrypoint /bin/bash \
+            vllm/vllm-openai:nightly-1acd67a795ebccdf9b9db7697ae9082058301657 \
+            -c "${vllmCmdFlat}"
+        '';
+
+      # Cache Bug - On resume from cache, VRAM usage is higher than just generating in real time.
+
+      # -e TRITON_CACHE_DIR=/root/.triton/cache \
+      # -v /mnt/ssd/vLLM/Cache/torch_compile:/root/.cache/vllm/torch_compile_cache \
+      # -v /mnt/ssd/vLLM/Cache/triton:/root/.triton/cache \
+
+      cmdStop = "${pkgs.docker}/bin/docker stop \${MODEL_ID}";
+
+      metadata = {
+        type = [
+          "text-generation"
+          "coding"
+        ];
+      };
+    };
+
+    # https://github.com/noonghunna/club-3090/tree/master/models/qwen3.6-27b/vllm
+    "qwen3.6-27b-vllm-145k-vl-cuda0" = {
+      name = "Qwen3.6 27B (vLLM, 145K, VL, CUDA0)";
+      checkEndpoint = "/v1/models";
+      macros.ctx = "145000";
+      proxy = "http://127.0.0.1:\${PORT}";
+      cmd =
+        let
+          vllmCmd = ''
+            set -e; pip install xxhash pandas scipy -q;
+            python3 -m vllm._genesis.patches.apply_all;
+            python3 /patches/patch_timings_1acd67a.py;
+            exec vllm serve ''${VLLM_ENFORCE_EAGER:+--enforce-eager}
+            --served-model-name ''${MODEL_ID}
+            --model /root/.cache/huggingface/qwen3.6-27b-autoround-int4
+            --quantization auto_round
+            --dtype float16
+            --tensor-parallel-size 1
+            --max-model-len ''${ctx}
+            --gpu-memory-utilization 0.95
+            --max-num-seqs 1
+            --max-num-batched-tokens 4128
+            --kv-cache-dtype turboquant_3bit_nc
+            --trust-remote-code
+            --reasoning-parser qwen3
+            --enable-auto-tool-choice
+            --tool-call-parser qwen3_coder
+            --enable-prefix-caching
+            --enable-chunked-prefill
+            --no-scheduler-reserve-full-isl
+            --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":3}'
+            --host 0.0.0.0
+            --port 8000
+          '';
+          vllmCmdFlat = builtins.replaceStrings [ "\n" ] [ " " ] vllmCmd;
+        in
+        ''
+          ${pkgs.docker}/bin/docker run --rm --device=nvidia.com/gpu=all \
+            --name ''${MODEL_ID} \
+            --ipc=host \
+            -e CUDA_DEVICE_MAX_CONNECTIONS=8 \
+            -e CUDA_DEVICE_ORDER=PCI_BUS_ID \
+            -e CUDA_VISIBLE_DEVICES=0 \
+            -e GENESIS_BUFFER_MODE=shared \
+            -e GENESIS_ENABLE_P100=1 \
+            -e GENESIS_ENABLE_P101=1 \
+            -e GENESIS_ENABLE_P103=1 \
+            -e GENESIS_ENABLE_P15B_FA_VARLEN_CLAMP=1 \
+            -e GENESIS_ENABLE_P38B_COMPILE_SAFE=1 \
+            -e GENESIS_ENABLE_P4=1 \
+            -e GENESIS_ENABLE_P58_ASYNC_PLACEHOLDER_FIX=1 \
+            -e GENESIS_ENABLE_P60B_TRITON_KERNEL=1 \
+            -e GENESIS_ENABLE_P60_GDN_NGRAM_FIX=1 \
+            -e GENESIS_ENABLE_P61B_STREAMING_OVERLAP=1 \
+            -e GENESIS_ENABLE_P61_QWEN3_MULTI_TOOL=1 \
+            -e GENESIS_ENABLE_P62_STRUCT_OUT_SPEC_TIMING=1 \
+            -e GENESIS_ENABLE_P64_QWEN3CODER_MTP_STREAMING=1 \
+            -e GENESIS_ENABLE_P66_CUDAGRAPH_SIZE_FILTER=1 \
+            -e GENESIS_ENABLE_P67_TQ_MULTI_QUERY_KERNEL=1 \
+            -e GENESIS_ENABLE_P68_AUTO_FORCE_TOOL=1 \
+            -e GENESIS_ENABLE_P69_LONG_CTX_TOOL_REMINDER=1 \
+            -e GENESIS_ENABLE_P72_PROFILE_RUN_CAP=1 \
+            -e GENESIS_ENABLE_P74_CHUNK_CLAMP=1 \
+            -e GENESIS_ENABLE_P78_TOLIST_CAPTURE_GUARD=0 \
+            -e GENESIS_ENABLE_P81_FP8_BLOCK_SCALED_M_LE_8=0 \
+            -e GENESIS_ENABLE_P82=0 \
+            -e GENESIS_ENABLE_P83=1 \
+            -e GENESIS_ENABLE_P87=1 \
+            -e GENESIS_ENABLE_P91=1 \
+            -e GENESIS_ENABLE_P94=1 \
+            -e GENESIS_ENABLE_P98=1 \
+            -e GENESIS_ENABLE_P99=1 \
+            -e GENESIS_ENABLE_PN11_GDN_AB_CONTIGUOUS=1 \
+            -e GENESIS_ENABLE_PN12_FFN_INTERMEDIATE_POOL=1 \
+            -e GENESIS_ENABLE_PN13_CUDA_GRAPH_LAMBDA_ARITY=1 \
+            -e GENESIS_ENABLE_PN14_TQ_DECODE_OOB_CLAMP=1 \
+            -e GENESIS_ENABLE_PN17_FA2_LSE_CLAMP=1 \
+            -e GENESIS_ENABLE_PN19_SCOPED_MAX_SPLIT=1 \
+            -e GENESIS_ENABLE_PN22_LOCAL_ARGMAX_TP=1 \
+            -e GENESIS_ENABLE_PN25_SILU_INDUCTOR_SAFE=1 \
+            -e GENESIS_ENABLE_PN26_SPARSE_V=1 \
+            -e GENESIS_ENABLE_PN30_DS_LAYOUT_SPEC_DECODE=1 \
+            -e GENESIS_ENABLE_PN34_WORKSPACE_LOCK_RELAX=1 \
+            -e GENESIS_ENABLE_PN59_STREAMING_GDN=1 \
+            -e GENESIS_ENABLE_PN8_MTP_DRAFT_ONLINE_QUANT=1 \
+            -e GENESIS_ENABLE_PN9_INDEPENDENT_DRAFTER_ATTN=1 \
+            -e GENESIS_P68_P69_LONG_CTX_THRESHOLD_CHARS=50000 \
+            -e GENESIS_P82_THRESHOLD_SINGLE=0.3 \
+            -e GENESIS_PN26_SPARSE_V_BLOCK_KV=8 \
+            -e GENESIS_PN26_SPARSE_V_NUM_WARPS=4 \
+            -e GENESIS_PN26_SPARSE_V_THRESHOLD=0.01 \
+            -e GENESIS_PREALLOC_TOKEN_BUDGET=4128 \
+            -e GENESIS_PROFILE_RUN_CAP_M=4128 \
+            -e NCCL_CUMEM_ENABLE=0 \
+            -e NCCL_P2P_DISABLE=1 \
+            -e OMP_NUM_THREADS=1 \
+            -e PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True,max_split_size_mb:512 \
+            -e VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 \
+            -e VLLM_FLOAT32_MATMUL_PRECISION=high \
+            -e VLLM_MARLIN_USE_ATOMIC_ADD=1 \
+            -e VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0 \
+            -e VLLM_NO_USAGE_STATS=1 \
+            -e VLLM_SSM_CONV_STATE_LAYOUT=DS \
+            -e VLLM_USE_FLASHINFER_SAMPLER=1 \
+            -e VLLM_USE_FUSED_MOE_GROUPED_TOPK=1 \
+            -e VLLM_WORKER_MULTIPROC_METHOD=spawn \
+            -e VLLM_ENFORCE_EAGER \
+            -v /mnt/ssd/vLLM/Models:/root/.cache/huggingface \
+            -v /mnt/ssd/vLLM/Patches/genesis/vllm/_genesis:/usr/local/lib/python3.12/dist-packages/vllm/_genesis:ro \
+            -v /mnt/ssd/vLLM/Patches/patch_timings_1acd67a.py:/patches/patch_timings_1acd67a.py:ro \
+            -p ''${PORT}:8000 \
+            --entrypoint /bin/bash \
+            vllm/vllm-openai:nightly-1acd67a795ebccdf9b9db7697ae9082058301657 \
+            -c "${vllmCmdFlat}"
+        '';
+
+      # Cache Bug - On resume from cache, VRAM usage is higher than just generating in real time.
+
+      # -e TRITON_CACHE_DIR=/root/.triton/cache \
+      # -v /mnt/ssd/vLLM/Cache/torch_compile:/root/.cache/vllm/torch_compile_cache \
+      # -v /mnt/ssd/vLLM/Cache/triton:/root/.triton/cache \
+
+      cmdStop = "${pkgs.docker}/bin/docker stop \${MODEL_ID}";
+
+      metadata = {
+        type = [
+          "text-generation"
+          "coding"
+          "vision"
+        ];
+      };
+    };
+
+    # https://github.com/noonghunna/club-3090/tree/master/models/qwen3.6-27b/vllm
+    "qwen3.6-27b-vllm-180k-cuda0" = {
+      name = "Qwen3.6 27B (vLLM, 180K, CUDA0)";
+      checkEndpoint = "/v1/models";
       macros.ctx = "180000";
       proxy = "http://127.0.0.1:\${PORT}";
       cmd =
@@ -110,8 +433,8 @@ in
           vllmCmd = ''
             set -e; pip install xxhash pandas scipy -q;
             python3 -m vllm._genesis.patches.apply_all;
-            python3 /patches/patch_timings_07351e088.py;
-            exec vllm serve
+            python3 /patches/patch_timings_1acd67a.py;
+            exec vllm serve ''${VLLM_ENFORCE_EAGER:+--enforce-eager}
             --served-model-name ''${MODEL_ID}
             --model /root/.cache/huggingface/qwen3.6-27b-autoround-int4
             --quantization auto_round
@@ -157,7 +480,6 @@ in
             -e GENESIS_ENABLE_P61_QWEN3_MULTI_TOOL=1 \
             -e GENESIS_ENABLE_P62_STRUCT_OUT_SPEC_TIMING=1 \
             -e GENESIS_ENABLE_P64_QWEN3CODER_MTP_STREAMING=1 \
-            -e GENESIS_ENABLE_P65_TURBOQUANT_SPEC_CG_DOWNGRADE=1 \
             -e GENESIS_ENABLE_P66_CUDAGRAPH_SIZE_FILTER=1 \
             -e GENESIS_ENABLE_P67_TQ_MULTI_QUERY_KERNEL=1 \
             -e GENESIS_ENABLE_P68_AUTO_FORCE_TOOL=1 \
@@ -183,6 +505,7 @@ in
             -e GENESIS_ENABLE_PN25_SILU_INDUCTOR_SAFE=1 \
             -e GENESIS_ENABLE_PN26_SPARSE_V=1 \
             -e GENESIS_ENABLE_PN30_DS_LAYOUT_SPEC_DECODE=1 \
+            -e GENESIS_ENABLE_PN31_FA_VARLEN_PERSISTENT_OUT=1 \
             -e GENESIS_ENABLE_PN32_GDN_CHUNKED_PREFILL=1 \
             -e GENESIS_ENABLE_PN34_WORKSPACE_LOCK_RELAX=1 \
             -e GENESIS_ENABLE_PN59_STREAMING_GDN=1 \
@@ -203,7 +526,6 @@ in
             -e NCCL_P2P_DISABLE=1 \
             -e OMP_NUM_THREADS=1 \
             -e PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True,max_split_size_mb:512 \
-            -e TRITON_CACHE_DIR=/root/.triton/cache \
             -e VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 \
             -e VLLM_FLOAT32_MATMUL_PRECISION=high \
             -e VLLM_MARLIN_USE_ATOMIC_ADD=1 \
@@ -213,244 +535,22 @@ in
             -e VLLM_USE_FLASHINFER_SAMPLER=1 \
             -e VLLM_USE_FUSED_MOE_GROUPED_TOPK=1 \
             -e VLLM_WORKER_MULTIPROC_METHOD=spawn \
+            -e VLLM_ENFORCE_EAGER \
             -v /mnt/ssd/vLLM/Models:/root/.cache/huggingface \
-            -v /mnt/ssd/vLLM/Cache/torch_compile:/root/.cache/vllm/torch_compile_cache \
-            -v /mnt/ssd/vLLM/Cache/triton:/root/.triton/cache \
             -v /mnt/ssd/vLLM/Patches/genesis/vllm/_genesis:/usr/local/lib/python3.12/dist-packages/vllm/_genesis:ro \
-            -v /mnt/ssd/vLLM/Patches/patch_timings_07351e088.py:/patches/patch_timings_07351e088.py:ro \
+            -v /mnt/ssd/vLLM/Patches/patch_timings_1acd67a.py:/patches/patch_timings_1acd67a.py:ro \
             -p ''${PORT}:8000 \
             --entrypoint /bin/bash \
-            vllm/vllm-openai:nightly-01d4d1ad375dc5854779c593eee093bcebb0cada \
+            vllm/vllm-openai:nightly-1acd67a795ebccdf9b9db7697ae9082058301657 \
             -c "${vllmCmdFlat}"
         '';
-      cmdStop = "${pkgs.docker}/bin/docker stop \${MODEL_ID}";
 
-      metadata = {
-        type = [
-          "text-generation"
-          "coding"
-        ];
-      };
-    };
+      # Cache Bug - On resume from cache, VRAM usage is higher than just generating in real time.
 
-    # https://github.com/noonghunna/club-3090/tree/master/models/qwen3.6-27b/vllm
-    # Synced from: club-3090 f6613c8 (2026-05-02) — docker-compose.long-vision.yml
-    # Long-vision variant - 145K context with vision tower active
-    # TurboQuant 3-bit KV + MTP n=3 + Genesis v7.69 + Cliff 2 env vars (mem-util kept at 0.95)
-    "vllm-qwen3.6-27b-long-vision" = {
-      name = "vLLM Qwen3.6 (27B) - Long Vision";
-      macros.ctx = "145000";
-      proxy = "http://127.0.0.1:\${PORT}";
-      cmd =
-        let
-          vllmCmd = ''
-            set -e; pip install xxhash pandas scipy -q;
-            python3 -m vllm._genesis.patches.apply_all;
-            python3 /patches/patch_timings_07351e088.py;
-            exec vllm serve
-            --served-model-name ''${MODEL_ID}
-            --model /root/.cache/huggingface/qwen3.6-27b-autoround-int4
-            --quantization auto_round
-            --dtype float16
-            --tensor-parallel-size 1
-            --max-model-len ''${ctx}
-            --gpu-memory-utilization 0.95
-            --max-num-seqs 1
-            --max-num-batched-tokens 4128
-            --kv-cache-dtype turboquant_3bit_nc
-            --trust-remote-code
-            --reasoning-parser qwen3
-            --enable-auto-tool-choice
-            --tool-call-parser qwen3_coder
-            --enable-prefix-caching
-            --enable-chunked-prefill
-            --no-scheduler-reserve-full-isl
-            --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":3}'
-            --host 0.0.0.0
-            --port 8000
-          '';
-          vllmCmdFlat = builtins.replaceStrings [ "\n" ] [ " " ] vllmCmd;
-        in
-        ''
-          ${pkgs.docker}/bin/docker run --rm --device=nvidia.com/gpu=all \
-            --name ''${MODEL_ID} \
-            --ipc=host \
-            -e CUDA_DEVICE_MAX_CONNECTIONS=8 \
-            -e CUDA_DEVICE_ORDER=PCI_BUS_ID \
-            -e CUDA_VISIBLE_DEVICES=0 \
-            -e GENESIS_BUFFER_MODE=shared \
-            -e GENESIS_ENABLE_P100=1 \
-            -e GENESIS_ENABLE_P101=1 \
-            -e GENESIS_ENABLE_P103=1 \
-            -e GENESIS_ENABLE_P15B_FA_VARLEN_CLAMP=1 \
-            -e GENESIS_ENABLE_P38B_COMPILE_SAFE=1 \
-            -e GENESIS_ENABLE_P4=1 \
-            -e GENESIS_ENABLE_P58_ASYNC_PLACEHOLDER_FIX=1 \
-            -e GENESIS_ENABLE_P60B_TRITON_KERNEL=1 \
-            -e GENESIS_ENABLE_P60_GDN_NGRAM_FIX=1 \
-            -e GENESIS_ENABLE_P61B_STREAMING_OVERLAP=1 \
-            -e GENESIS_ENABLE_P61_QWEN3_MULTI_TOOL=1 \
-            -e GENESIS_ENABLE_P62_STRUCT_OUT_SPEC_TIMING=1 \
-            -e GENESIS_ENABLE_P64_QWEN3CODER_MTP_STREAMING=1 \
-            -e GENESIS_ENABLE_P65_TURBOQUANT_SPEC_CG_DOWNGRADE=1 \
-            -e GENESIS_ENABLE_P66_CUDAGRAPH_SIZE_FILTER=1 \
-            -e GENESIS_ENABLE_P67_TQ_MULTI_QUERY_KERNEL=1 \
-            -e GENESIS_ENABLE_P68_AUTO_FORCE_TOOL=1 \
-            -e GENESIS_ENABLE_P69_LONG_CTX_TOOL_REMINDER=1 \
-            -e GENESIS_ENABLE_P72_PROFILE_RUN_CAP=1 \
-            -e GENESIS_ENABLE_P74_CHUNK_CLAMP=1 \
-            -e GENESIS_ENABLE_P78_TOLIST_CAPTURE_GUARD=0 \
-            -e GENESIS_ENABLE_P81_FP8_BLOCK_SCALED_M_LE_8=0 \
-            -e GENESIS_ENABLE_P82=0 \
-            -e GENESIS_ENABLE_P83=1 \
-            -e GENESIS_ENABLE_P87=1 \
-            -e GENESIS_ENABLE_P91=1 \
-            -e GENESIS_ENABLE_P94=1 \
-            -e GENESIS_ENABLE_P98=1 \
-            -e GENESIS_ENABLE_P99=1 \
-            -e GENESIS_ENABLE_PN11_GDN_AB_CONTIGUOUS=1 \
-            -e GENESIS_ENABLE_PN12_FFN_INTERMEDIATE_POOL=1 \
-            -e GENESIS_ENABLE_PN13_CUDA_GRAPH_LAMBDA_ARITY=1 \
-            -e GENESIS_ENABLE_PN14_TQ_DECODE_OOB_CLAMP=1 \
-            -e GENESIS_ENABLE_PN17_FA2_LSE_CLAMP=1 \
-            -e GENESIS_ENABLE_PN19_SCOPED_MAX_SPLIT=1 \
-            -e GENESIS_ENABLE_PN22_LOCAL_ARGMAX_TP=1 \
-            -e GENESIS_ENABLE_PN25_SILU_INDUCTOR_SAFE=1 \
-            -e GENESIS_ENABLE_PN26_SPARSE_V=1 \
-            -e GENESIS_ENABLE_PN30_DS_LAYOUT_SPEC_DECODE=1 \
-            -e GENESIS_ENABLE_PN32_GDN_CHUNKED_PREFILL=1 \
-            -e GENESIS_ENABLE_PN34_WORKSPACE_LOCK_RELAX=1 \
-            -e GENESIS_ENABLE_PN59_STREAMING_GDN=1 \
-            -e GENESIS_ENABLE_PN8_MTP_DRAFT_ONLINE_QUANT=1 \
-            -e GENESIS_ENABLE_PN9_INDEPENDENT_DRAFTER_ATTN=1 \
-            -e GENESIS_FLA_FWD_H_MAX_T=16384 \
-            -e GENESIS_P68_P69_LONG_CTX_THRESHOLD_CHARS=50000 \
-            -e GENESIS_P82_THRESHOLD_SINGLE=0.3 \
-            -e GENESIS_PN26_SPARSE_V_BLOCK_KV=8 \
-            -e GENESIS_PN26_SPARSE_V_NUM_WARPS=4 \
-            -e GENESIS_PN26_SPARSE_V_THRESHOLD=0.01 \
-            -e GENESIS_PN32_GDN_CHUNK_SIZE=8192 \
-            -e GENESIS_PN32_GDN_CHUNK_THRESHOLD=16384 \
-            -e GENESIS_PREALLOC_TOKEN_BUDGET=4128 \
-            -e GENESIS_PROFILE_RUN_CAP_M=4128 \
-            -e NCCL_CUMEM_ENABLE=0 \
-            -e NCCL_P2P_DISABLE=1 \
-            -e OMP_NUM_THREADS=1 \
-            -e PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True,max_split_size_mb:512 \
-            -e TRITON_CACHE_DIR=/root/.triton/cache \
-            -e VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 \
-            -e VLLM_FLOAT32_MATMUL_PRECISION=high \
-            -e VLLM_MARLIN_USE_ATOMIC_ADD=1 \
-            -e VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0 \
-            -e VLLM_NO_USAGE_STATS=1 \
-            -e VLLM_SSM_CONV_STATE_LAYOUT=DS \
-            -e VLLM_USE_FLASHINFER_SAMPLER=1 \
-            -e VLLM_USE_FUSED_MOE_GROUPED_TOPK=1 \
-            -e VLLM_WORKER_MULTIPROC_METHOD=spawn \
-            -v /mnt/ssd/vLLM/Models:/root/.cache/huggingface \
-            -v /mnt/ssd/vLLM/Cache/torch_compile:/root/.cache/vllm/torch_compile_cache \
-            -v /mnt/ssd/vLLM/Cache/triton:/root/.triton/cache \
-            -v /mnt/ssd/vLLM/Patches/genesis/vllm/_genesis:/usr/local/lib/python3.12/dist-packages/vllm/_genesis:ro \
-            -v /mnt/ssd/vLLM/Patches/patch_timings_07351e088.py:/patches/patch_timings_07351e088.py:ro \
-            -p ''${PORT}:8000 \
-            --entrypoint /bin/bash \
-            vllm/vllm-openai:nightly-01d4d1ad375dc5854779c593eee093bcebb0cada \
-            -c "${vllmCmdFlat}"
-        '';
-      cmdStop = "${pkgs.docker}/bin/docker stop \${MODEL_ID}";
+      # -e TRITON_CACHE_DIR=/root/.triton/cache \
+      # -v /mnt/ssd/vLLM/Cache/torch_compile:/root/.cache/vllm/torch_compile_cache \
+      # -v /mnt/ssd/vLLM/Cache/triton:/root/.triton/cache \
 
-      metadata = {
-        type = [
-          "text-generation"
-          "coding"
-          "vision"
-        ];
-      };
-    };
-
-    # https://github.com/noonghunna/club-3090/tree/master/models/qwen3.6-27b/vllm
-    # Synced from: club-3090 ae4846f (2026-05-02) — docker-compose.tools-text.yml
-    # Tools-text variant - 75K context, text-only (no vision)
-    # fp8_e5m2 KV + MTP n=3. IDE agents (Cline, Cursor, OpenCode, etc.)
-    "vllm-qwen3.6-27b-tools-text" = {
-      name = "vLLM Qwen3.6 (27B) - Tools Text";
-      macros.ctx = "75000";
-      proxy = "http://127.0.0.1:\${PORT}";
-      cmd =
-        let
-          vllmCmd = ''
-            set -e; pip install xxhash pandas scipy -q;
-            python3 -m vllm._genesis.patches.apply_all;
-            python3 /patches/patch_timings_07351e088.py;
-            exec vllm serve
-            --served-model-name ''${MODEL_ID}
-            --model /root/.cache/huggingface/qwen3.6-27b-autoround-int4
-            --quantization auto_round
-            --dtype float16
-            --tensor-parallel-size 1
-            --max-model-len ''${ctx}
-            --gpu-memory-utilization 0.97
-            --max-num-seqs 1
-            --max-num-batched-tokens 2048
-            --kv-cache-dtype fp8_e5m2
-            --language-model-only
-            --trust-remote-code
-            --reasoning-parser qwen3
-            --enable-auto-tool-choice
-            --tool-call-parser qwen3_coder
-            --enable-prefix-caching
-            --enable-chunked-prefill
-            --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":3}'
-            --host 0.0.0.0
-            --port 8000
-          '';
-          vllmCmdFlat = builtins.replaceStrings [ "\n" ] [ " " ] vllmCmd;
-        in
-        ''
-          ${pkgs.docker}/bin/docker run --rm --device=nvidia.com/gpu=all \
-            --name ''${MODEL_ID} \
-            --ipc=host \
-            -e CUDA_DEVICE_MAX_CONNECTIONS=8 \
-            -e CUDA_DEVICE_ORDER=PCI_BUS_ID \
-            -e CUDA_VISIBLE_DEVICES=0 \
-            -e GENESIS_ENABLE_P58_ASYNC_PLACEHOLDER_FIX=1 \
-            -e GENESIS_ENABLE_P64_QWEN3CODER_MTP_STREAMING=1 \
-            -e GENESIS_ENABLE_P66_CUDAGRAPH_SIZE_FILTER=1 \
-            -e GENESIS_ENABLE_P68_AUTO_FORCE_TOOL=1 \
-            -e GENESIS_ENABLE_P69_LONG_CTX_TOOL_REMINDER=1 \
-            -e GENESIS_ENABLE_P72_PROFILE_RUN_CAP=1 \
-            -e GENESIS_ENABLE_P74_CHUNK_CLAMP=1 \
-            -e GENESIS_ENABLE_P94=1 \
-            -e GENESIS_ENABLE_PN13_CUDA_GRAPH_LAMBDA_ARITY=1 \
-            -e GENESIS_ENABLE_PN14_TQ_DECODE_OOB_CLAMP=1 \
-            -e GENESIS_ENABLE_PN17_FA2_LSE_CLAMP=1 \
-            -e GENESIS_ENABLE_PN19_SCOPED_MAX_SPLIT=1 \
-            -e GENESIS_ENABLE_PN59_STREAMING_GDN=1 \
-            -e GENESIS_ENABLE_PN8_MTP_DRAFT_ONLINE_QUANT=1 \
-            -e GENESIS_P68_P69_LONG_CTX_THRESHOLD_CHARS=50000 \
-            -e GENESIS_PROFILE_RUN_CAP_M=4128 \
-            -e NCCL_CUMEM_ENABLE=0 \
-            -e NCCL_P2P_DISABLE=1 \
-            -e OMP_NUM_THREADS=1 \
-            -e PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True,max_split_size_mb:512 \
-            -e TRITON_CACHE_DIR=/root/.triton/cache \
-            -e VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 \
-            -e VLLM_FLOAT32_MATMUL_PRECISION=high \
-            -e VLLM_MARLIN_USE_ATOMIC_ADD=1 \
-            -e VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0 \
-            -e VLLM_NO_USAGE_STATS=1 \
-            -e VLLM_USE_FLASHINFER_SAMPLER=1 \
-            -e VLLM_WORKER_MULTIPROC_METHOD=spawn \
-            -v /mnt/ssd/vLLM/Models:/root/.cache/huggingface \
-            -v /mnt/ssd/vLLM/Cache/torch_compile:/root/.cache/vllm/torch_compile_cache \
-            -v /mnt/ssd/vLLM/Cache/triton:/root/.triton/cache \
-            -v /mnt/ssd/vLLM/Patches/genesis/vllm/_genesis:/usr/local/lib/python3.12/dist-packages/vllm/_genesis:ro \
-            -v /mnt/ssd/vLLM/Patches/patch_timings_07351e088.py:/patches/patch_timings_07351e088.py:ro \
-            -p ''${PORT}:8000 \
-            --entrypoint /bin/bash \
-            vllm/vllm-openai:nightly-01d4d1ad375dc5854779c593eee093bcebb0cada \
-            -c "${vllmCmdFlat}"
-        '';
       cmdStop = "${pkgs.docker}/bin/docker stop \${MODEL_ID}";
 
       metadata = {
@@ -465,37 +565,9 @@ in
     # ------------- GTX 1080 Ti -------------
     # ---------------------------------------
 
-    # https://huggingface.co/unsloth/Qwen3-VL-8B-Instruct-GGUF/tree/main
-    "qwen3-8b-vision" = {
-      name = "Qwen3 Vision (8B) - Thinking";
-      macros.ctx = "60000";
-      env = [ "CUDA_VISIBLE_DEVICES=1" ];
-      cmd = ''
-        ${llama-cpp}/bin/llama-server \
-          --port ''${PORT} \
-          -m /mnt/ssd/Models/Qwen3/Qwen3-VL-8B-Instruct-UD-Q4_K_XL.gguf \
-          --mmproj /mnt/ssd/Models/Qwen3/Qwen3-VL-8B-Instruct-UD-Q4_K_XL_mmproj-F16.gguf \
-          -c ''${ctx} \
-          --temp 0.7 \
-          --min-p 0.0 \
-          --top-p 0.8 \
-          --top-k 20 \
-          -ctk q8_0 \
-          -ctv q8_0 \
-          -fit off \
-          -dev CUDA0
-      '';
-      metadata = {
-        type = [
-          "text-generation"
-          "vision"
-        ];
-      };
-    };
-
     # https://huggingface.co/unsloth/Qwen3.5-9B-GGUF/tree/main
-    "qwen3.5-9b-thinking" = {
-      name = "Qwen3.5 (9B) - Thinking";
+    "qwen3.5-9b-vl-cuda1" = {
+      name = "Qwen3.5 9B (VL, CUDA1)";
       macros.ctx = "131072";
       env = [ "CUDA_VISIBLE_DEVICES=1" ];
       cmd = ''
@@ -521,8 +593,8 @@ in
     };
 
     # https://huggingface.co/unsloth/Qwen3.5-4B-GGUF/tree/main
-    "qwen3.5-4b-thinking" = {
-      name = "Qwen3.5 (4B) - Thinking";
+    "qwen3.5-4b-cuda1" = {
+      name = "Qwen3.5 4B (CUDA1)";
       macros.ctx = "131072";
       env = [ "CUDA_VISIBLE_DEVICES=1" ];
       cmd = ''
@@ -540,6 +612,7 @@ in
       metadata = {
         type = [
           "text-generation"
+          "coding"
         ];
       };
     };
@@ -549,8 +622,8 @@ in
     # ---------------------------------------
 
     # https://huggingface.co/unsloth/Qwen3-Coder-Next-GGUF/tree/main
-    "qwen3-coder-next-80b-instruct" = {
-      name = "Qwen3 Coder Next (80B) - Instruct";
+    "qwen3-coder-next-80b-dual" = {
+      name = "Qwen3 Coder Next 80B (Dual GPU)";
       macros.ctx = "131072";
       cmd = ''
         ${llama-cpp}/bin/llama-server \
@@ -574,13 +647,79 @@ in
       };
     };
 
+    # https://huggingface.co/unsloth/Qwen3.6-27B-GGUF-MTP/tree/main
+    "qwen3.6-27b-dual" = {
+      name = "Qwen3.6 27B (Dual GPU, UD-Q6)";
+      macros.ctx = "120000";
+      cmd = ''
+        ${llama-cpp}/bin/llama-server \
+          --port ''${PORT} \
+          -m /mnt/ssd/Models/Qwen3.6/Qwen3.6-27B-UD-Q6_K_XL.gguf \
+          -c ''${ctx} \
+          -np 4 -kvu \
+          --temp 0.6 \
+          --top-p 0.95 \
+          --top-k 20 \
+          --min-p 0.00 \
+          --presence-penalty 0.0 \
+          -ctk q8_0 \
+          -ctv q8_0 \
+          --spec-type draft-mtp \
+          --spec-draft-n-max 3 \
+          -dev CUDA0,CUDA1 \
+          -ts 73,27 \
+          -fit off \
+          --chat-template-kwargs "{\"preserve_thinking\": true}"
+      '';
+      metadata = {
+        type = [
+          "text-generation"
+          "coding"
+        ];
+      };
+    };
+
+    # https://huggingface.co/unsloth/Qwen3.6-35B-A3B-MTP-GGUF/tree/main
+    "qwen3.6-35b-dual" = {
+      name = "Qwen3.6 35B (Dual GPU, UD-Q6)";
+      # macros.ctx = "215000";
+      # -ctk q8_0 \
+      # -ctv q8_0 \
+      macros.ctx = "131072";
+      cmd = ''
+        ${llama-cpp}/bin/llama-server \
+          --port ''${PORT} \
+          -m /mnt/ssd/Models/Qwen3.6/Qwen3.6-35B-A3B-UD-Q6_K.gguf \
+          -c ''${ctx} \
+          -np 4 -kvu \
+          --temp 0.6 \
+          --top-p 0.95 \
+          --top-k 20 \
+          --min-p 0.00 \
+          --presence-penalty 0.0 \
+          --spec-type draft-mtp \
+          --spec-draft-n-max 3 \
+          -dev CUDA0,CUDA1 \
+          -fit off \
+          -ts 72,28 \
+          --chat-template-kwargs "{\"preserve_thinking\": true}"
+      '';
+      metadata = {
+        type = [
+          "text-generation"
+          "coding"
+        ];
+      };
+    };
+
     # ---------------------------------------
-    # ---------- Stable Diffussion ----------
+    # ---------- Stable Diffusion ----------
     # ---------------------------------------
 
-    "z-image-turbo" = {
+    "z-image-turbo-cuda0" = {
       name = "Z-Image-Turbo";
       checkEndpoint = "/";
+      env = [ "CUDA_VISIBLE_DEVICES=0" ];
       cmd = ''
         ${stable-diffusion-cpp}/bin/sd-server \
           --listen-port ''${PORT} \
@@ -597,9 +736,10 @@ in
       };
     };
 
-    "qwen-image-edit-2511" = {
+    "qwen-image-edit-2511-cuda0" = {
       name = "Qwen Image Edit 2511";
       checkEndpoint = "/";
+      env = [ "CUDA_VISIBLE_DEVICES=0" ];
       cmd = ''
         ${stable-diffusion-cpp}/bin/sd-server \
           --listen-port ''${PORT} \
@@ -623,9 +763,10 @@ in
       };
     };
 
-    "qwen-image-2512" = {
+    "qwen-image-2512-cuda0" = {
       name = "Qwen Image 2512";
       checkEndpoint = "/";
+      env = [ "CUDA_VISIBLE_DEVICES=0" ];
       cmd = ''
         ${stable-diffusion-cpp}/bin/sd-server \
           --listen-port ''${PORT} \
@@ -645,9 +786,10 @@ in
       };
     };
 
-    "chroma-radiance" = {
+    "chroma-radiance-cuda0" = {
       name = "Chroma Radiance";
       checkEndpoint = "/";
+      env = [ "CUDA_VISIBLE_DEVICES=0" ];
       cmd = ''
         ${stable-diffusion-cpp}/bin/sd-server \
           --listen-port ''${PORT} \
@@ -667,36 +809,31 @@ in
   # Concurrent Model Matrix
   #
   # CUDA0 models can run alongside CUDA1 models (one each). Models not
-  # listed in any set (dual-GPU models using -ts) run alone and evict
-  # everything.
+  # listed in any set (dual-GPU models) run alone and evict everything.
   matrix = {
     vars = {
       # --- RTX 3090 Models ---
-      vlt = "vllm-qwen3.6-27b-long-text";
-      vtt = "vllm-qwen3.6-27b-tools-text";
-      vlv = "vllm-qwen3.6-27b-long-vision";
-      go = "gpt-oss-20b-thinking";
-      q36a = "qwen3.6-35b-thinking";
-      q36b = "qwen3.6-27b-thinking";
-      zi = "z-image-turbo";
-      qie = "qwen-image-edit-2511";
-      qi = "qwen-image-2512";
-      cr = "chroma-radiance";
+      v180 = "qwen3.6-27b-vllm-180k-cuda0";
+      v145 = "qwen3.6-27b-vllm-145k-vl-cuda0";
+      v75 = "qwen3.6-27b-vllm-75k-cuda0";
+      v50 = "qwen3.6-27b-vllm-50k-cuda0";
+      go = "gpt-oss-20b-cuda0";
+      g4 = "gemma-4-26b-vl-cuda0";
+      q36a = "qwen3.6-35b-cuda0";
+      q36b = "qwen3.6-27b-cuda0";
+      q36ik = "qwen3.6-27b-ik-cuda0";
+      zi = "z-image-turbo-cuda0";
+      qie = "qwen-image-edit-2511-cuda0";
+      qi = "qwen-image-2512-cuda0";
+      cr = "chroma-radiance-cuda0";
 
       # --- GTX 1080 Ti Models ---
-      qv = "qwen3-8b-vision";
-      q4 = "qwen3.5-4b-thinking";
-      q9 = "qwen3.5-9b-thinking";
-    };
-
-    evict_costs = {
-      vlt = 50;
-      vtt = 50;
-      vlv = 50;
+      q4 = "qwen3.5-4b-cuda1";
+      q9 = "qwen3.5-9b-vl-cuda1";
     };
 
     sets = {
-      concurrent = "(go | q36a | q36b | vlt | vtt | vlv | zi | qie | qi | cr) & (qv | q4 | q9)";
+      concurrent = "(go | g4 | q36a | q36b | q36ik | v180 | v145 | v75 | v50 | zi | qie | qi | cr) & (q4 | q9)";
     };
   };
 }

modules/nixos/services/llama-swap/default.nix

@@ -11,6 +11,31 @@ let
   cfg = config.${namespace}.services.llama-swap;
 
   llama-swap = pkgs.reichard.llama-swap;
+  llamaCppPresets =
+    let
+      models = (import ./config.nix { inherit pkgs; }).models;
+      llamaCppModels = lib.filterAttrs (_: model: lib.hasInfix "/bin/llama-server" (model.cmd or "")) models;
+    in
+    builtins.mapAttrs (_: model: {
+      inherit (model) cmd;
+      name = model.name or "";
+      env = model.env or [ ];
+    }) llamaCppModels;
+  llamaCppPresetFile = pkgs.writeText "llama-cpp-presets.json" (builtins.toJSON llamaCppPresets);
+  llama-cpp-bisect-context = pkgs.writeShellApplication {
+    name = "llama-cpp-bisect-context";
+    runtimeInputs = with pkgs; [
+      coreutils
+      curl
+      gnused
+      python3
+      util-linux
+    ];
+    text = builtins.replaceStrings
+      [ "__LLAMA_CPP_PRESETS__" ]
+      [ "${llamaCppPresetFile}" ]
+      (builtins.readFile ./scripts/llama-cpp-bisect-context);
+  };
 in
 {
   options.${namespace}.services.llama-swap = {
@@ -37,7 +62,6 @@ in
       description = "Model swapping for LLaMA C++ Server (or any local OpenAPI compatible server)";
       after = [ "network.target" ];
       wantedBy = [ "multi-user.target" ];
-
       serviceConfig = {
         Type = "exec";
         ExecStart = "${lib.getExe llama-swap} --listen :8080 --config ${
@@ -97,6 +121,7 @@ in
         };
       }) apiKeys);
       templates."llama-swap.json" = {
+        restartUnits = [ "llama-swap.service" ];
         owner = "llama-swap";
         group = "llama-swap";
         mode = "0400";
@@ -108,6 +133,8 @@ in
       };
     };
 
+    environment.systemPackages = [ llama-cpp-bisect-context ];
+
     networking.firewall.allowedTCPPorts = [ 8080 ];
   };
 }

modules/nixos/services/llama-swap/patches/README.md

@@ -1,22 +1,54 @@
 # vLLM Timings Patch
 
-This scratch directory contains two ways to patch vLLM so its OpenAI-compatible responses include llama.cpp-compatible `timings` data. llama-swap already parses this `timings` object to populate cached tokens, prompt processing speed, and generation speed.
+This directory contains the custom timings patch for the current vLLM Docker image used by the llama-swap module:
+
+```text
+vllm/vllm-openai:nightly-1acd67a795ebccdf9b9db7697ae9082058301657
+```
+
+The patch adds a top-level llama.cpp-compatible `timings` object to OpenAI-compatible responses so llama-swap can populate cached tokens, prompt processing speed, and generation speed.
 
 ## Files
 
-- `patch_timings_07351e088.py` — disk-edit patch script for running inside the vLLM Docker container before `vllm serve`.
-- `vllm-timings-07351e088.patch` — standard unified git patch against `vllm/vllm-openai:nightly-07351e0883470724dd5a7e9730ed10e01fc99d08`.
+- `patch_timings_1acd67a.py` — idempotent boot-time disk-edit patch script for the vLLM Docker container.
+- `vllm-timings-1acd67a.patch` — equivalent standard unified git patch against the current image's vLLM source.
 
-## What The Patch Adds
+## Runtime Script
 
-The patch adds a top-level `timings` object to:
+Deploy the script under `/mnt/ssd/vLLM/Patches/` and mount it into the container:
 
-- `/v1/chat/completions` non-streaming responses
-- `/v1/chat/completions` streaming final usage chunk
-- `/v1/completions` non-streaming responses
-- `/v1/completions` streaming final usage chunk
+```nix
+-v /mnt/ssd/vLLM/Patches/patch_timings_1acd67a.py:/patches/patch_timings_1acd67a.py:ro \
+```
 
-The object matches llama.cpp's fields:
+Run it before `exec vllm serve`:
+
+```bash
+python3 /patches/patch_timings_1acd67a.py;
+exec vllm serve ...
+```
+
+The script is idempotent. Re-running it skips files that already contain `# [patch_timings]`.
+
+## Standard Patch
+
+For a source checkout at commit `1acd67a795ebccdf9b9db7697ae9082058301657`:
+
+```bash
+git apply --check /path/to/vllm-timings-1acd67a.patch
+git apply /path/to/vllm-timings-1acd67a.patch
+```
+
+At container runtime, applying the `.patch` directly is possible if the image has `patch` or `git` installed:
+
+```bash
+cd /usr/local/lib/python3.12/dist-packages
+patch -p1 < /patches/vllm-timings-1acd67a.patch
+```
+
+The Python script remains the safer boot-time option because it is idempotent and does not depend on external patch tools being present in the Docker image.
+
+## Timings Fields
 
 ```json
 {
@@ -35,78 +67,3 @@ Data comes from vLLM's existing internal `RequestStateStats` and `RequestOutput.
 - prompt/prefill time: `first_token_ts - scheduled_ts`
 - generation/decode time: `last_token_ts - first_token_ts`
 - cached tokens: `num_cached_tokens`
-
-## Option 1: Runtime Docker Patch Script
-
-Copy the script into the deployed patch directory:
-
-```bash
-cp _scratch/patch_timings_07351e088.py /mnt/ssd/vLLM/Patches/patch_timings_07351e088.py
-```
-
-Add the Docker mount in `/etc/nixos/modules/nixos/services/llama-swap/config.nix`:
-
-```nix
--v /mnt/ssd/vLLM/Patches/patch_timings_07351e088.py:/patches/patch_timings_07351e088.py:ro \
-```
-
-Run it before `exec vllm serve` in `vllmCmd`:
-
-```bash
-python3 /patches/patch_timings_07351e088.py;
-exec vllm serve ...
-```
-
-The script is idempotent. Re-running it skips files that already contain `# [patch_timings]`.
-
-## Option 2: Standard Patch File
-
-Use this for a source checkout or future vLLM updates where conflicts can be resolved normally.
-
-From a vLLM checkout at commit `07351e0883470724dd5a7e9730ed10e01fc99d08`:
-
-```bash
-git apply /path/to/_scratch/vllm-timings-07351e088.patch
-```
-
-Or with `patch`:
-
-```bash
-patch -p1 < /path/to/_scratch/vllm-timings-07351e088.patch
-```
-
-For future vLLM versions, try:
-
-```bash
-git apply --check /path/to/_scratch/vllm-timings-07351e088.patch
-```
-
-If it fails, apply manually or with rejects and resolve conflicts around the changed response-construction code.
-
-## Verification Performed
-
-The patch was checked against the Docker tag's pinned commit:
-
-```text
-vllm/vllm-openai:nightly-07351e0883470724dd5a7e9730ed10e01fc99d08
-```
-
-Validation done locally:
-
-```bash
-git apply --check _scratch/vllm-timings-07351e088.patch
-git apply _scratch/vllm-timings-07351e088.patch
-nix run nixpkgs#python3 -- -m py_compile \
-  vllm/entrypoints/openai/chat_completion/protocol.py \
-  vllm/entrypoints/openai/chat_completion/serving.py \
-  vllm/entrypoints/openai/completion/protocol.py \
-  vllm/entrypoints/openai/completion/serving.py
-```
-
-The runtime `patch_timings_07351e088.py` script was also tested against files extracted from the pinned commit and confirmed idempotent.
-
-## Caveats
-
-- Normal chat completion usage should be correct.
-- `/v1/completions` with multiple prompts returns aggregate token counts, but the timing values come from the last completed request. Single-prompt completions are the expected use case.
-- Streaming timings are attached only to the final usage chunk, so clients must request/include usage for streaming if they want timings in the stream.

modules/nixos/services/llama-swap/patches/patch_timings_1acd67a.py

@@ -1,5 +1,5 @@
 """
-Disk-edit patch for vLLM nightly-07351e0883470724dd5a7e9730ed10e01fc99d08:
+Disk-edit patch for vLLM nightly-1acd67a795ebccdf9b9db7697ae9082058301657:
 inject llama.cpp-compatible `timings` into chat/completion API responses.
 
 Adds `timings` to:
@@ -13,7 +13,7 @@ The `timings` object matches llama.cpp fields consumed by llama-swap:
   predicted_n, predicted_ms, predicted_per_second, cache_n
 
 Usage, before `exec vllm serve`:
-  python3 /patches/patch_timings.py
+  python3 /patches/patch_timings_1acd67a.py
 """
 
 import logging
@@ -85,70 +85,8 @@ def _write(path, content):
 
 def _replace_once(content, old, new, label):
     count = content.count(old)
-    if count == 1:
-        return content.replace(old, new, 1)
-
-    # vLLM v0.20 added system_fingerprint to response constructors. Preserve
-    # compatibility with the original dev205 anchors by retrying with that
-    # field inserted when the old anchor is not present.
-    variants = [
-        (
-            old.replace(
-                "                    usage=final_usage,\n                )",
-                "                    usage=final_usage,\n                    system_fingerprint=self.system_fingerprint,\n                )",
-            ),
-            new.replace(
-                "                    usage=final_usage,\n                )",
-                "                    usage=final_usage,\n                    system_fingerprint=self.system_fingerprint,\n                )",
-            ),
-        ),
-        (
-            old.replace(
-                "            usage=usage,\n            prompt_logprobs=",
-                "            usage=usage,\n            system_fingerprint=self.system_fingerprint,\n            prompt_logprobs=",
-            ),
-            new.replace(
-                "            usage=usage,\n            prompt_logprobs=",
-                "            usage=usage,\n            system_fingerprint=self.system_fingerprint,\n            prompt_logprobs=",
-            ),
-        ),
-        (
-            old.replace(
-                "                    usage=final_usage_info,\n                )",
-                "                    usage=final_usage_info,\n                    system_fingerprint=self.system_fingerprint,\n                )",
-            ),
-            new.replace(
-                "                    usage=final_usage_info,\n                )",
-                "                    usage=final_usage_info,\n                    system_fingerprint=self.system_fingerprint,\n                )",
-            ),
-        ),
-        (
-            old.replace(
-                "            usage=usage,\n            kv_transfer_params=kv_transfer_params,",
-                "            usage=usage,\n            system_fingerprint=self.system_fingerprint,\n            kv_transfer_params=kv_transfer_params,",
-            ),
-            new.replace(
-                "            usage=usage,\n            kv_transfer_params=kv_transfer_params,",
-                "            usage=usage,\n            system_fingerprint=self.system_fingerprint,\n            kv_transfer_params=kv_transfer_params,",
-            ),
-        ),
-    ]
-    matches = [(variant_old, variant_new) for variant_old, variant_new in variants if content.count(variant_old) == 1]
-    if len(matches) == 1:
-        variant_old, variant_new = matches[0]
-        return content.replace(variant_old, variant_new, 1)
-
-    variant_counts = [content.count(variant_old) for variant_old, _ in variants]
-    raise RuntimeError(f"{label}: anchor matched {count} times; v0.20 variants matched {variant_counts}")
-
-
-def _replace_once_any(content, replacements, label):
-    """Replace exactly one of several version-specific anchors."""
-    matches = [(old, new) for old, new in replacements if content.count(old) == 1]
-    if len(matches) != 1:
-        counts = [content.count(old) for old, _ in replacements]
-        raise RuntimeError(f"{label}: versioned anchors matched {counts}")
-    old, new = matches[0]
+    if count != 1:
+        raise RuntimeError(f"{label}: anchor matched {count} times")
     return content.replace(old, new, 1)
 
 
@@ -231,19 +169,19 @@ def _patch_chat_serving(vllm_dir):
             label,
         )
 
-        # Streaming Final Usage Chunk - pinned image has no system_fingerprint arg.
+        # Streaming Final Usage Chunk
         content = _replace_once(
             content,
-            '''                final_usage_chunk = ChatCompletionStreamResponse(\n                    id=request_id,\n                    object=chunk_object_type,\n                    created=created_time,\n                    choices=[],\n                    model=model_name,\n                    usage=final_usage,\n                )\n''',
-            f'''                final_usage_chunk = ChatCompletionStreamResponse(\n                    id=request_id,\n                    object=chunk_object_type,\n                    created=created_time,\n                    choices=[],\n                    model=model_name,\n                    usage=final_usage,\n                )\n                # Inject Timings  {PATCH_TAG}\n                try:\n                    _s_cached = _last_stream_res.num_cached_tokens\n                    final_usage_chunk.timings = _compute_timings(\n                        _last_stream_res.metrics,\n                        num_prompt_tokens, completion_tokens, _s_cached,\n                    )\n                except NameError:\n                    pass\n''',
+            '''                final_usage_chunk = ChatCompletionStreamResponse(\n                    id=request_id,\n                    object=chunk_object_type,\n                    created=created_time,\n                    choices=[],\n                    model=model_name,\n                    usage=final_usage,\n                    system_fingerprint=self.system_fingerprint,\n                )\n''',
+            f'''                final_usage_chunk = ChatCompletionStreamResponse(\n                    id=request_id,\n                    object=chunk_object_type,\n                    created=created_time,\n                    choices=[],\n                    model=model_name,\n                    usage=final_usage,\n                    system_fingerprint=self.system_fingerprint,\n                )\n                # Inject Timings  {PATCH_TAG}\n                try:\n                    _s_cached = _last_stream_res.num_cached_tokens\n                    final_usage_chunk.timings = _compute_timings(\n                        _last_stream_res.metrics,\n                        num_prompt_tokens, completion_tokens, _s_cached,\n                    )\n                except NameError:\n                    pass\n''',
             label,
         )
 
-        # Non-Streaming Response - pinned image has no system_fingerprint arg.
+        # Non-Streaming Response
         content = _replace_once(
             content,
-            '''        response = ChatCompletionResponse(\n            id=request_id,\n            created=created_time,\n            model=model_name,\n            choices=choices,\n            usage=usage,\n            prompt_logprobs=clamp_prompt_logprobs(final_res.prompt_logprobs),\n            prompt_token_ids=(\n                final_res.prompt_token_ids if request.return_token_ids else None\n            ),\n            kv_transfer_params=final_res.kv_transfer_params,\n        )\n''',
-            f'''        response = ChatCompletionResponse(\n            id=request_id,\n            created=created_time,\n            model=model_name,\n            choices=choices,\n            usage=usage,\n            prompt_logprobs=clamp_prompt_logprobs(final_res.prompt_logprobs),\n            prompt_token_ids=(\n                final_res.prompt_token_ids if request.return_token_ids else None\n            ),\n            kv_transfer_params=final_res.kv_transfer_params,\n        )\n\n        # Inject Timings  {PATCH_TAG}\n        _cached = final_res.num_cached_tokens\n        response.timings = _compute_timings(\n            final_res.metrics, num_prompt_tokens, num_generated_tokens,\n            _cached,\n        )\n''',
+            '''        response = ChatCompletionResponse(\n            id=request_id,\n            created=created_time,\n            model=model_name,\n            choices=choices,\n            usage=usage,\n            system_fingerprint=self.system_fingerprint,\n            prompt_logprobs=clamp_prompt_logprobs(final_res.prompt_logprobs),\n            prompt_token_ids=(\n                final_res.prompt_token_ids if request.return_token_ids else None\n            ),\n            kv_transfer_params=final_res.kv_transfer_params,\n            prompt_routed_experts=prompt_routed_experts,\n        )\n''',
+            f'''        response = ChatCompletionResponse(\n            id=request_id,\n            created=created_time,\n            model=model_name,\n            choices=choices,\n            usage=usage,\n            system_fingerprint=self.system_fingerprint,\n            prompt_logprobs=clamp_prompt_logprobs(final_res.prompt_logprobs),\n            prompt_token_ids=(\n                final_res.prompt_token_ids if request.return_token_ids else None\n            ),\n            kv_transfer_params=final_res.kv_transfer_params,\n            prompt_routed_experts=prompt_routed_experts,\n        )\n\n        # Inject Timings  {PATCH_TAG}\n        _cached = final_res.num_cached_tokens\n        response.timings = _compute_timings(\n            final_res.metrics, num_prompt_tokens, num_generated_tokens,\n            _cached,\n        )\n''',
             label,
         )
     except RuntimeError as e:
@@ -284,19 +222,19 @@ def _patch_completion_serving(vllm_dir):
             label,
         )
 
-        # Streaming Final Usage Chunk - pinned image has no system_fingerprint arg.
+        # Streaming Final Usage Chunk
         content = _replace_once(
             content,
-            '''                final_usage_chunk = CompletionStreamResponse(\n                    id=request_id,\n                    created=created_time,\n                    model=model_name,\n                    choices=[],\n                    usage=final_usage_info,\n                )\n''',
-            f'''                final_usage_chunk = CompletionStreamResponse(\n                    id=request_id,\n                    created=created_time,\n                    model=model_name,\n                    choices=[],\n                    usage=final_usage_info,\n                )\n                # Inject Timings  {PATCH_TAG}\n                try:\n                    _sc_cached = _last_comp_res.num_cached_tokens\n                    final_usage_chunk.timings = _compute_timings(\n                        _last_comp_res.metrics,\n                        total_prompt_tokens, total_completion_tokens,\n                        _sc_cached,\n                    )\n                except NameError:\n                    pass\n''',
+            '''                final_usage_chunk = CompletionStreamResponse(\n                    id=request_id,\n                    created=created_time,\n                    model=model_name,\n                    choices=[],\n                    usage=final_usage_info,\n                    system_fingerprint=self.system_fingerprint,\n                )\n''',
+            f'''                final_usage_chunk = CompletionStreamResponse(\n                    id=request_id,\n                    created=created_time,\n                    model=model_name,\n                    choices=[],\n                    usage=final_usage_info,\n                    system_fingerprint=self.system_fingerprint,\n                )\n                # Inject Timings  {PATCH_TAG}\n                try:\n                    _sc_cached = _last_comp_res.num_cached_tokens\n                    final_usage_chunk.timings = _compute_timings(\n                        _last_comp_res.metrics,\n                        total_prompt_tokens, total_completion_tokens,\n                        _sc_cached,\n                    )\n                except NameError:\n                    pass\n''',
             label,
         )
 
-        # Non-Streaming Response - pinned image has no system_fingerprint arg.
+        # Non-Streaming Response
         content = _replace_once(
             content,
-            '''        return CompletionResponse(\n            id=request_id,\n            created=created_time,\n            model=model_name,\n            choices=choices,\n            usage=usage,\n            kv_transfer_params=kv_transfer_params,\n        )\n''',
-            f'''        _comp_response = CompletionResponse(  {PATCH_TAG}\n            id=request_id,\n            created=created_time,\n            model=model_name,\n            choices=choices,\n            usage=usage,\n            kv_transfer_params=kv_transfer_params,\n        )\n        # Inject Timings  {PATCH_TAG}\n        if last_final_res is not None:\n            _comp_cached = last_final_res.num_cached_tokens\n            _comp_response.timings = _compute_timings(\n                last_final_res.metrics, num_prompt_tokens,\n                num_generated_tokens, _comp_cached,\n            )\n        return _comp_response\n''',
+            '''        return CompletionResponse(\n            id=request_id,\n            created=created_time,\n            model=model_name,\n            choices=choices,\n            usage=usage,\n            system_fingerprint=self.system_fingerprint,\n            kv_transfer_params=kv_transfer_params,\n            prompt_routed_experts=prompt_routed_experts,\n        )\n''',
+            f'''        _comp_response = CompletionResponse(  {PATCH_TAG}\n            id=request_id,\n            created=created_time,\n            model=model_name,\n            choices=choices,\n            usage=usage,\n            system_fingerprint=self.system_fingerprint,\n            kv_transfer_params=kv_transfer_params,\n            prompt_routed_experts=prompt_routed_experts,\n        )\n        # Inject Timings  {PATCH_TAG}\n        if last_final_res is not None:\n            _comp_cached = last_final_res.num_cached_tokens\n            _comp_response.timings = _compute_timings(\n                last_final_res.metrics, num_prompt_tokens,\n                num_generated_tokens, _comp_cached,\n            )\n        return _comp_response\n''',
             label,
         )
     except RuntimeError as e:

modules/nixos/services/llama-swap/patches/vllm-timings-1acd67a.patch

@@ -1,8 +1,8 @@
 diff --git a/vllm/entrypoints/openai/chat_completion/protocol.py b/vllm/entrypoints/openai/chat_completion/protocol.py
-index aacac38..074ca45 100644
+index 742f9cc..ade939f 100644
 --- a/vllm/entrypoints/openai/chat_completion/protocol.py
 +++ b/vllm/entrypoints/openai/chat_completion/protocol.py
-@@ -111,6 +111,9 @@ class ChatCompletionResponse(OpenAIBaseModel):
+@@ -115,6 +115,9 @@ class ChatCompletionResponse(OpenAIBaseModel):
          default=None, description="KVTransfer parameters."
      )
  
@@ -12,7 +12,7 @@ index aacac38..074ca45 100644
  
  class ChatCompletionResponseStreamChoice(OpenAIBaseModel):
      index: int
-@@ -132,6 +135,9 @@ class ChatCompletionStreamResponse(OpenAIBaseModel):
+@@ -139,6 +142,9 @@ class ChatCompletionStreamResponse(OpenAIBaseModel):
      # not part of the OpenAI spec but for tracing the tokens
      prompt_token_ids: list[int] | None = None
  
@@ -23,10 +23,10 @@ index aacac38..074ca45 100644
  class ChatCompletionToolsParam(OpenAIBaseModel):
      type: Literal["function"] = "function"
 diff --git a/vllm/entrypoints/openai/chat_completion/serving.py b/vllm/entrypoints/openai/chat_completion/serving.py
-index 12dc2cd..c15fb6d 100644
+index 1026e0a..a9c5708 100644
 --- a/vllm/entrypoints/openai/chat_completion/serving.py
 +++ b/vllm/entrypoints/openai/chat_completion/serving.py
-@@ -83,6 +83,34 @@ if TYPE_CHECKING:
+@@ -79,6 +79,34 @@ if TYPE_CHECKING:
  logger = init_logger(__name__)
  
  
@@ -61,7 +61,7 @@ index 12dc2cd..c15fb6d 100644
  class OpenAIServingChat(OpenAIServing):
      def __init__(
          self,
-@@ -633,6 +661,7 @@ class OpenAIServingChat(OpenAIServing):
+@@ -485,6 +513,7 @@ class OpenAIServingChat(OpenAIServing):
  
          try:
              async for res in result_generator:
@@ -69,9 +69,9 @@ index 12dc2cd..c15fb6d 100644
                  if res.prompt_token_ids is not None:
                      num_prompt_tokens = len(res.prompt_token_ids)
                      if res.encoder_prompt_token_ids is not None:
-@@ -1230,6 +1259,15 @@ class OpenAIServingChat(OpenAIServing):
-                     model=model_name,
+@@ -947,6 +976,15 @@ class OpenAIServingChat(OpenAIServing):
                      usage=final_usage,
+                     system_fingerprint=self.system_fingerprint,
                  )
 +                # Inject Timings  # [patch_timings]
 +                try:
@@ -85,8 +85,8 @@ index 12dc2cd..c15fb6d 100644
                  final_usage_data = final_usage_chunk.model_dump_json(
                      exclude_unset=True, exclude_none=True
                  )
-@@ -1644,6 +1682,13 @@ class OpenAIServingChat(OpenAIServing):
-             kv_transfer_params=final_res.kv_transfer_params,
+@@ -1377,6 +1415,13 @@ class OpenAIServingChat(OpenAIServing):
+             prompt_routed_experts=prompt_routed_experts,
          )
  
 +        # Inject Timings  # [patch_timings]
@@ -100,10 +100,10 @@ index 12dc2cd..c15fb6d 100644
          if self.enable_log_outputs and self.request_logger:
              for choice in choices:
 diff --git a/vllm/entrypoints/openai/completion/protocol.py b/vllm/entrypoints/openai/completion/protocol.py
-index c785d25..85928f4 100644
+index 7bb3c8d..8487e93 100644
 --- a/vllm/entrypoints/openai/completion/protocol.py
 +++ b/vllm/entrypoints/openai/completion/protocol.py
-@@ -485,6 +485,9 @@ class CompletionResponse(OpenAIBaseModel):
+@@ -489,6 +489,9 @@ class CompletionResponse(OpenAIBaseModel):
          default=None, description="KVTransfer parameters."
      )
  
@@ -113,15 +113,18 @@ index c785d25..85928f4 100644
  
  class CompletionResponseStreamChoice(OpenAIBaseModel):
      index: int
-@@ -512,3 +515,6 @@ class CompletionStreamResponse(OpenAIBaseModel):
+@@ -516,6 +519,9 @@ class CompletionStreamResponse(OpenAIBaseModel):
      model: str
      choices: list[CompletionResponseStreamChoice]
      usage: UsageInfo | None = Field(default=None)
 +
 +    # llama.cpp-compatible per-request timings  # [patch_timings]
 +    timings: dict[str, Any] | None = None
+     # Set only on the final chunk of a stream to mirror non-streaming responses
+     # without the per-chunk serialization overhead.
+     system_fingerprint: str | None = None
 diff --git a/vllm/entrypoints/openai/completion/serving.py b/vllm/entrypoints/openai/completion/serving.py
-index fb7f253..11a5350 100644
+index ee4ca9f..8b27011 100644
 --- a/vllm/entrypoints/openai/completion/serving.py
 +++ b/vllm/entrypoints/openai/completion/serving.py
 @@ -48,6 +48,34 @@ if TYPE_CHECKING:
@@ -159,7 +162,7 @@ index fb7f253..11a5350 100644
  class OpenAIServingCompletion(OpenAIServing):
      def __init__(
          self,
-@@ -290,6 +318,7 @@ class OpenAIServingCompletion(OpenAIServing):
+@@ -291,6 +319,7 @@ class OpenAIServingCompletion(OpenAIServing):
  
          try:
              async for prompt_idx, res in result_generator:
@@ -167,9 +170,9 @@ index fb7f253..11a5350 100644
                  prompt_token_ids = res.prompt_token_ids
                  prompt_logprobs = res.prompt_logprobs
  
-@@ -434,6 +463,16 @@ class OpenAIServingCompletion(OpenAIServing):
-                     choices=[],
+@@ -445,6 +474,16 @@ class OpenAIServingCompletion(OpenAIServing):
                      usage=final_usage_info,
+                     system_fingerprint=self.system_fingerprint,
                  )
 +                # Inject Timings  # [patch_timings]
 +                try:
@@ -184,18 +187,18 @@ index fb7f253..11a5350 100644
                  final_usage_data = final_usage_chunk.model_dump_json(
                      exclude_unset=False, exclude_none=True
                  )
-@@ -556,7 +595,7 @@ class OpenAIServingCompletion(OpenAIServing):
-         request_metadata.final_usage_info = usage
-         if final_res_batch:
-             kv_transfer_params = final_res_batch[0].kv_transfer_params
+@@ -577,7 +616,7 @@ class OpenAIServingCompletion(OpenAIServing):
+             if pre is not None:
+                 prompt_routed_experts = pre.tolist()
+ 
 -        return CompletionResponse(
 +        _comp_response = CompletionResponse(  # [patch_timings]
              id=request_id,
              created=created_time,
              model=model_name,
-@@ -564,6 +603,14 @@ class OpenAIServingCompletion(OpenAIServing):
-             usage=usage,
+@@ -587,6 +626,14 @@ class OpenAIServingCompletion(OpenAIServing):
              kv_transfer_params=kv_transfer_params,
+             prompt_routed_experts=prompt_routed_experts,
          )
 +        # Inject Timings  # [patch_timings]
 +        if last_final_res is not None:

modules/nixos/services/llama-swap/scripts/llama-cpp-bisect-context

@@ -0,0 +1,464 @@
+#!/usr/bin/env bash
+set -Eeuo pipefail
+
+usage() {
+  cat <<'EOF'
+Usage:
+  llama-cpp-bisect-context MODEL --low N --high N [options]
+  llama-cpp-bisect-context --cmd-template CMD --low N --high N [options]
+  llama-cpp-bisect-context --cmd-file FILE --low N --high N [options]
+
+Bisect the largest llama.cpp llama-server context that can start and complete
+a near-context prompt without OOMing. Startup-only mode is available for isolating the first cliff.
+
+MODEL is a llama.cpp preset generated from the Nix llama-swap config.
+
+Command templates are evaluated with these environment variables:
+  PORT  random listen port for this trial
+  CTX   candidate context size
+
+Options:
+  --cmd-template CMD        llama-server command, e.g. 'llama-server --port ${PORT} -c ${CTX} ...'
+  --cmd-file FILE           executable or shell snippet using $PORT and $CTX
+  --preset-file FILE        preset JSON file (default: Nix-generated presets)
+  --list-presets            list available Nix-generated presets and exit
+  --low N                   known/assumed lower context bound
+  --high N                  upper context bound to test
+  --step N                  stop when high-low <= N (default: 1024)
+  --prompt-ratio PCT        prompt fill target as percent of CTX (default: 90)
+  --chars-per-token N       rough prompt sizing ratio (default: 4)
+  --prompt-turns N          split the prompt across N user/assistant turns (default: 4)
+  --max-tokens N            generated tokens for prompt test (default: 32)
+  --startup-timeout SEC     seconds to wait for /health readiness (default: 300)
+  --request-timeout SEC     seconds to wait for prompt response (default: 600)
+  --cooldown SEC            seconds to sleep after stopping server (default: 5)
+  --startup-only            only test server startup, not prompt/runtime OOM
+  --verbose                 print llama-server logs for each failed trial
+  --keep-logs               keep trial logs after a successful run too
+  -h, --help                show this help
+
+Examples:
+  llama-cpp-bisect-context \
+    --cmd-template 'llama-server --port ${PORT} -m model.gguf -c ${CTX} -ngl 99' \
+    --low 32768 --high 196608
+
+  llama-cpp-bisect-context qwen3.6-27b-ik-cuda0 --low 32768 --high 180000
+  llama-cpp-bisect-context --cmd-file ./server-command.sh --low 32768 --high 196608
+EOF
+}
+
+preset_model=""
+preset_file="__LLAMA_CPP_PRESETS__"
+list_presets=0
+cmd_template=""
+cmd_file=""
+low=""
+high=""
+step=1024
+prompt_ratio=90
+chars_per_token=4
+prompt_turns=4
+max_tokens=32
+startup_timeout=300
+request_timeout=600
+cooldown=5
+startup_only=0
+verbose=0
+keep_logs=0
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --cmd-template) cmd_template="${2:-}"; shift 2 ;;
+    --cmd-file) cmd_file="${2:-}"; shift 2 ;;
+    --preset-file) preset_file="${2:-}"; shift 2 ;;
+    --list-presets) list_presets=1; shift ;;
+    --low) low="${2:-}"; shift 2 ;;
+    --high) high="${2:-}"; shift 2 ;;
+    --step) step="${2:-}"; shift 2 ;;
+    --prompt-ratio) prompt_ratio="${2:-}"; shift 2 ;;
+    --chars-per-token) chars_per_token="${2:-}"; shift 2 ;;
+    --prompt-turns) prompt_turns="${2:-}"; shift 2 ;;
+    --max-tokens) max_tokens="${2:-}"; shift 2 ;;
+    --startup-timeout) startup_timeout="${2:-}"; shift 2 ;;
+    --request-timeout) request_timeout="${2:-}"; shift 2 ;;
+    --cooldown) cooldown="${2:-}"; shift 2 ;;
+    --startup-only) startup_only=1; shift ;;
+    --verbose) verbose=1; shift ;;
+    --keep-logs) keep_logs=1; shift ;;
+    -h|--help) usage; exit 0 ;;
+    --*) echo "unknown argument: $1" >&2; usage >&2; exit 2 ;;
+    *)
+      if [[ -n "$preset_model" ]]; then
+        echo "unexpected positional argument: $1" >&2
+        usage >&2
+        exit 2
+      fi
+      preset_model="$1"
+      shift
+      ;;
+  esac
+done
+
+list_presets_json() {
+  python3 - "$preset_file" <<'PY'
+import json
+import sys
+with open(sys.argv[1]) as f:
+    presets = json.load(f)
+for key in sorted(presets):
+    name = presets[key].get("name", "")
+    print(f"{key}\t{name}" if name else key)
+PY
+  exit 0
+}
+
+if (( list_presets )); then
+  list_presets_json
+fi
+
+load_preset() {
+  local command_file="$tmpdir/preset-command.sh"
+  python3 - "$preset_file" "$preset_model" "$command_file" <<'PY'
+import json
+import shlex
+import sys
+
+preset_file, model_id, command_file = sys.argv[1:]
+with open(preset_file) as f:
+    presets = json.load(f)
+try:
+    preset = presets[model_id]
+except KeyError:
+    print(f"unknown preset: {model_id}", file=sys.stderr)
+    print("available presets:", file=sys.stderr)
+    for key in sorted(presets):
+        print(f"  {key}", file=sys.stderr)
+    sys.exit(2)
+
+cmd = preset["cmd"].replace("${ctx}", "${CTX}").replace("$ctx", "${CTX}")
+env = preset.get("env", [])
+with open(command_file, "w") as f:
+    f.write("set -e\n")
+    for item in env:
+        key, sep, value = item.partition("=")
+        if not sep or not key:
+            continue
+        f.write(f"export {key}={shlex.quote(value)}\n")
+    f.write(cmd)
+    if not cmd.endswith("\n"):
+        f.write("\n")
+PY
+  cmd_file="$command_file"
+}
+
+require_int() {
+  local name="$1" value="$2"
+  if [[ ! "$value" =~ ^[0-9]+$ ]]; then
+    echo "$name must be a positive integer" >&2
+    exit 2
+  fi
+}
+
+mode_count=0
+[[ -n "$preset_model" ]] && mode_count=$((mode_count + 1))
+[[ -n "$cmd_template" ]] && mode_count=$((mode_count + 1))
+[[ -n "$cmd_file" ]] && mode_count=$((mode_count + 1))
+if (( mode_count != 1 )); then
+  echo "use exactly one of MODEL, --cmd-template, or --cmd-file" >&2
+  exit 2
+fi
+if [[ -z "$low" || -z "$high" ]]; then
+  echo "missing --low or --high" >&2
+  exit 2
+fi
+
+for pair in \
+  "--low:$low" \
+  "--high:$high" \
+  "--step:$step" \
+  "--prompt-ratio:$prompt_ratio" \
+  "--chars-per-token:$chars_per_token" \
+  "--prompt-turns:$prompt_turns" \
+  "--max-tokens:$max_tokens" \
+  "--startup-timeout:$startup_timeout" \
+  "--request-timeout:$request_timeout" \
+  "--cooldown:$cooldown"; do
+  require_int "${pair%%:*}" "${pair#*:}"
+done
+
+if (( low <= 0 || high <= low || step <= 0 || prompt_ratio <= 0 || chars_per_token <= 0 || prompt_turns <= 0 )); then
+  echo "invalid numeric bounds/options" >&2
+  exit 2
+fi
+
+if [[ -n "$cmd_file" && ! -f "$cmd_file" ]]; then
+  echo "cmd file not found: $cmd_file" >&2
+  exit 2
+fi
+
+for dep in curl python3; do
+  if ! command -v "$dep" >/dev/null 2>&1; then
+    echo "missing required command: $dep" >&2
+    exit 2
+  fi
+done
+
+tmpdir="$(mktemp -d)"
+server_pid=""
+log_file=""
+
+terminate_server() {
+  if [[ -z "${server_pid:-}" ]]; then
+    return 0
+  fi
+
+  kill -- "-${server_pid}" >/dev/null 2>&1 || true
+  kill "$server_pid" >/dev/null 2>&1 || true
+
+  local waited=0
+  while kill -0 "$server_pid" >/dev/null 2>&1 && (( waited < 30 )); do
+    sleep 1
+    waited=$((waited + 1))
+  done
+
+  if kill -0 "$server_pid" >/dev/null 2>&1; then
+    kill -9 -- "-${server_pid}" >/dev/null 2>&1 || true
+    kill -9 "$server_pid" >/dev/null 2>&1 || true
+  fi
+
+  wait "$server_pid" >/dev/null 2>&1 || true
+  server_pid=""
+}
+
+cleanup() {
+  local status=$?
+  trap - EXIT INT TERM HUP
+  terminate_server
+  if (( keep_logs || status != 0 )); then
+    echo "logs kept in: $tmpdir" >&2
+  else
+    rm -rf "$tmpdir"
+  fi
+}
+
+interrupt() {
+  echo "interrupted; stopping llama-server" >&2
+  exit 130
+}
+
+trap cleanup EXIT
+trap interrupt INT TERM HUP
+
+if [[ -n "$preset_model" ]]; then
+  load_preset
+fi
+
+free_port() {
+  python3 - <<'PY'
+import socket
+with socket.socket() as s:
+    s.bind(("127.0.0.1", 0))
+    print(s.getsockname()[1])
+PY
+}
+
+start_server() {
+  local ctx="$1"
+  PORT="$(free_port)"
+  CTX="$ctx"
+  export PORT CTX
+  log_file="$tmpdir/llama-server-${ctx}.log"
+  {
+    printf 'CTX=%s\n' "$CTX"
+    printf 'PORT=%s\n' "$PORT"
+    if [[ -n "$cmd_file" ]]; then
+      printf 'CMD_FILE=%s\n' "$cmd_file"
+    else
+      printf 'CMD_TEMPLATE=%s\n' "$cmd_template"
+    fi
+    printf -- '--- llama-server output ---\n'
+  } >"$log_file"
+
+  if [[ -n "$cmd_file" ]]; then
+    setsid bash "$cmd_file" >>"$log_file" 2>&1 &
+  else
+    setsid bash -c "$cmd_template" >>"$log_file" 2>&1 &
+  fi
+  server_pid="$!"
+}
+
+stop_server() {
+  terminate_server
+  sleep "$cooldown"
+}
+
+print_failure_log() {
+  local label="$1" ctx="$2"
+  echo "[$label] ctx=$ctx failed; log: $log_file" >&2
+  if (( verbose )) && [[ -f "$log_file" ]]; then
+    sed -n '1,220p' "$log_file" >&2 || true
+  fi
+}
+
+wait_ready() {
+  local deadline=$((SECONDS + startup_timeout))
+  while (( SECONDS < deadline )); do
+    if [[ -n "${server_pid:-}" ]] && ! kill -0 "$server_pid" >/dev/null 2>&1; then
+      return 1
+    fi
+    if curl -fsS --max-time 5 "http://127.0.0.1:${PORT}/health" >/dev/null 2>&1; then
+      return 0
+    fi
+    if curl -fsS --max-time 5 "http://127.0.0.1:${PORT}/v1/models" >/dev/null 2>&1; then
+      return 0
+    fi
+    sleep 2
+  done
+  return 1
+}
+
+make_prompt_json() {
+  local ctx="$1"
+  local approx_tokens=$(( ctx * prompt_ratio / 100 ))
+  local chars=$(( approx_tokens * chars_per_token ))
+  python3 - "$chars" "$max_tokens" "$prompt_turns" <<'PY'
+import json
+import sys
+
+chars = int(sys.argv[1])
+max_tokens = int(sys.argv[2])
+prompt_turns = int(sys.argv[3])
+seed = (
+    "This is deterministic context filler for memory testing. "
+    "It uses normal words so token estimates are closer to real prompts. "
+)
+messages = []
+remaining = chars
+for turn in range(prompt_turns):
+    turns_left = prompt_turns - turn
+    chunk_chars = max(1, remaining // turns_left)
+    content = (seed * ((chunk_chars // len(seed)) + 1))[:chunk_chars]
+    messages.append({"role": "user", "content": content})
+    remaining -= chunk_chars
+    if turn != prompt_turns - 1:
+        messages.append({"role": "assistant", "content": "Acknowledged."})
+
+print(json.dumps({
+    "messages": messages,
+    "max_tokens": max_tokens,
+    "temperature": 0,
+    "stream": False,
+}))
+PY
+}
+
+run_prompt() {
+  local ctx="$1"
+  local payload="$tmpdir/prompt-${ctx}.json"
+  make_prompt_json "$ctx" >"$payload"
+  curl -fsS \
+    --max-time "$request_timeout" \
+    -H 'Content-Type: application/json' \
+    -d "@$payload" \
+    "http://127.0.0.1:${PORT}/v1/chat/completions" \
+    >/dev/null
+}
+
+test_startup() {
+  local ctx="$1"
+  echo "[startup] testing ctx=$ctx" >&2
+  start_server "$ctx"
+  if wait_ready; then
+    stop_server
+    echo "[startup] ctx=$ctx PASS" >&2
+    return 0
+  fi
+  print_failure_log startup "$ctx"
+  stop_server
+  return 1
+}
+
+test_qualified_context() {
+  local ctx="$1"
+  echo "[ctx] testing ctx=$ctx with prompt_ratio=${prompt_ratio}% prompt_turns=${prompt_turns}" >&2
+  start_server "$ctx"
+  if ! wait_ready; then
+    print_failure_log ctx-startup "$ctx"
+    stop_server
+    return 1
+  fi
+  if run_prompt "$ctx"; then
+    stop_server
+    echo "[ctx] ctx=$ctx PASS" >&2
+    return 0
+  fi
+  print_failure_log ctx-prompt "$ctx"
+  stop_server
+  return 1
+}
+
+bisect_max() {
+  local label="$1" pass="$2" fail="$3" fn="$4"
+  while (( fail - pass > step )); do
+    local mid=$(( (pass + fail) / 2 ))
+    if "$fn" "$mid"; then
+      pass="$mid"
+    else
+      fail="$mid"
+    fi
+  done
+  printf '%s:%s:%s\n' "$label" "$pass" "$fail"
+}
+
+if (( startup_only )); then
+  if ! test_startup "$low"; then
+    echo "low bound does not pass startup: $low" >&2
+    exit 1
+  fi
+  result="$(bisect_max startup "$low" "$high" test_startup)"
+  pass="$(cut -d: -f2 <<<"$result")"
+  fail="$(cut -d: -f3 <<<"$result")"
+
+  printf '\nResult:\n'
+  printf '  startup max passing ctx: %s\n' "$pass"
+  printf '  startup min failing ctx: %s\n' "$fail"
+  python3 - "$pass" "$fail" <<'PY'
+import json
+import sys
+max_passing, min_failing = map(int, sys.argv[1:])
+print(json.dumps({"startup": {"maxPassingCtx": max_passing, "minFailingCtx": min_failing}}, indent=2))
+PY
+  exit 0
+fi
+
+if ! test_qualified_context "$low"; then
+  echo "low bound does not pass qualified context test: $low" >&2
+  exit 1
+fi
+
+result="$(bisect_max context "$low" "$high" test_qualified_context)"
+pass="$(cut -d: -f2 <<<"$result")"
+fail="$(cut -d: -f3 <<<"$result")"
+
+printf '\nResult:\n'
+printf '  context max passing ctx: %s\n' "$pass"
+printf '  context min failing ctx: %s\n' "$fail"
+printf '  prompt ratio:            %s%%\n' "$prompt_ratio"
+printf '  prompt turns:            %s\n' "$prompt_turns"
+
+python3 - "$pass" "$fail" "$prompt_ratio" "$prompt_turns" <<'PY'
+import json
+import sys
+max_passing = int(sys.argv[1])
+min_failing = int(sys.argv[2])
+prompt_ratio = int(sys.argv[3])
+prompt_turns = int(sys.argv[4])
+print(json.dumps({
+    "context": {
+        "maxPassingCtx": max_passing,
+        "minFailingCtx": min_failing,
+        "promptRatio": prompt_ratio,
+        "promptTurns": prompt_turns,
+    }
+}, indent=2))
+PY

modules/nixos/services/llama-swap/setup-qwen36-vllm.sh

@@ -1,10 +1,6 @@
 #!/usr/bin/env bash
 # Setup script for vLLM Qwen3.6-27B on a single 3090.
 #
-# Downloads the model, clones Genesis patches (pinned), applies setup-time
-# source patches to the Genesis tree, and fetches all boot-time sidecar
-# patches into place under /mnt/ssd/vLLM/.
-#
 # Idempotent - safe to re-run; skips steps already completed.
 #
 # Prerequisites: git (with git-lfs), docker
@@ -15,17 +11,18 @@ set -euo pipefail
 MODEL_DIR="/mnt/ssd/vLLM/Models"
 MODEL_SUBDIR="qwen3.6-27b-autoround-int4"
 PATCHES_DIR="/mnt/ssd/vLLM/Patches"
+TEMPLATES_DIR="/mnt/ssd/vLLM/Templates"
 CACHE_DIR="/mnt/ssd/vLLM/Cache"
 GENESIS_DIR="${PATCHES_DIR}/genesis"
 GENESIS_PIN="${GENESIS_PIN:-7b9fd319}"
 
-# 3090 Patches
-BASE_3090_PATCH_URL="https://raw.githubusercontent.com/noonghunna/club-3090/v7.69-cliff2-test/models/qwen3.6-27b/vllm/patches"
-INPUTS_EMBEDS_PATCH="${PATCHES_DIR}/patch_inputs_embeds_optional.py"
-
 # Timings Patch
-TIMINGS_PATCH="${PATCHES_DIR}/patch_timings_07351e088.py"
-TIMINGS_PATCH_URL="${TIMINGS_PATCH_URL:-https://gitea.va.reichard.io/evan/nix/raw/branch/master/modules/nixos/services/llama-swap/patches/patch_timings_07351e088.py}"
+TIMINGS_PATCH="${PATCHES_DIR}/patch_timings_1acd67a.py"
+TIMINGS_PATCH_URL="${TIMINGS_PATCH_URL:-https://gitea.va.reichard.io/evan/nix/raw/branch/master/modules/nixos/services/llama-swap/patches/patch_timings_1acd67a.py}"
+
+# Template
+TEMPLATE="${TEMPLATES_DIR}/chat_template-v11.jinja"
+TEMPLATE_URL="https://huggingface.co/froggeric/Qwen-Fixed-Chat-Templates/resolve/main/qwen3.6/chat_template-v11.jinja"
 
 # ---------- Preflight Checks ----------
 for cmd in git git-lfs curl; do
@@ -37,7 +34,7 @@ done
 
 # ---------- Create Directories ----------
 echo "Creating directories..."
-mkdir -p "${MODEL_DIR}" "${PATCHES_DIR}" "${CACHE_DIR}/torch_compile" "${CACHE_DIR}/triton"
+mkdir -p "${TEMPLATES_DIR}" "${MODEL_DIR}" "${PATCHES_DIR}" "${CACHE_DIR}/torch_compile" "${CACHE_DIR}/triton"
 
 # ---------- Download Model ----------
 if [ -d "${MODEL_DIR}/${MODEL_SUBDIR}/.git" ]; then
@@ -67,36 +64,31 @@ if [[ ! -d "${GENESIS_DIR}/vllm/_genesis" ]]; then
 fi
 echo "Genesis pinned to ${GENESIS_PIN} ($(cd "${GENESIS_DIR}" && git rev-parse --short HEAD))"
 
-# ---------- Download Sidecar Patches ----------
-download_patch() {
-  local dest="$1"
-  local filename
-  filename="$(basename "$dest")"
-  if [ -f "${dest}" ]; then
-    echo "Patch ${filename} already present, skipping."
+# ---------- Download URL Patch ----------
+install_via_url() {
+  local name="$1"
+  local url="$2"
+  local dest="$3"
+  local tmp_patch
+  tmp_patch="$(mktemp)"
+
+  echo "Downloading ${name}..."
+  curl -fsSL "${url}" -o "${tmp_patch}"
+
+  if [ -f "${dest}" ] && cmp -s "${tmp_patch}" "${dest}"; then
+    echo "${name} already current at ${dest}, skipping."
   else
-    echo "Downloading ${filename}..."
-    curl -fsSL "${BASE_3090_PATCH_URL}/${filename}" -o "${dest}"
-    echo "Patch ${filename} written."
+    echo "Installing ${name} to ${dest}..."
+    install -m 0644 "${tmp_patch}" "${dest}"
+    echo "${name} installed."
   fi
+
+  rm -f "${tmp_patch}"
 }
 
-download_patch "${INPUTS_EMBEDS_PATCH}"
-
-# ---------- Download Timing Patch ----------
-tmp_timings_patch="$(mktemp)"
-trap 'rm -f "${tmp_timings_patch}"' EXIT
-
-echo "Downloading patch_timings_07351e088.py from this repo..."
-curl -fsSL "${TIMINGS_PATCH_URL}" -o "${tmp_timings_patch}"
-
-if [ -f "${TIMINGS_PATCH}" ] && cmp -s "${tmp_timings_patch}" "${TIMINGS_PATCH}"; then
-  echo "Timing patch already current at ${TIMINGS_PATCH}, skipping."
-else
-  echo "Installing timing patch to ${TIMINGS_PATCH}..."
-  install -m 0644 "${tmp_timings_patch}" "${TIMINGS_PATCH}"
-  echo "Timing patch installed."
-fi
+# ---------- Download Assets ----------
+install_via_url "patch_timings_1acd67a.py" "${TIMINGS_PATCH_URL}" "${TIMINGS_PATCH}"
+install_via_url "chat_template-v11.jinja" "${TEMPLATE_URL}" "${TEMPLATE}"
 
 # ---------- Summary ----------
 echo ""
@@ -109,11 +101,12 @@ echo "Expected layout:"
 echo "  /mnt/ssd/vLLM/"
 echo "  ├── Models/"
 echo "  │   └── qwen3.6-27b-autoround-int4/           (model weights)"
+echo "  ├── Templates/"
+echo "  │   └── chat_template-v11.jinja               (chat template)"
 echo "  ├── Cache/"
 echo "  │   ├── torch_compile/                        (torch.compile cache)"
 echo "  │   └── triton/                               (Triton kernel cache)"
 echo "  └── Patches/"
 echo "      ├── genesis/                              (Genesis @ ${GENESIS_PIN})"
 echo "      │   └── vllm/_genesis/                    (mounted into container)"
-echo "      ├── patch_inputs_embeds_optional.py        (boot-time: vllm#35975 backport, text-only models)"
-echo "      └── patch_timings_07351e088.py             (boot-time: llama.cpp-compatible timings)"
+echo "      └── patch_timings_1acd67a.py              (boot-time: llama.cpp-compatible timings)"

overlays/firefox-addons/default.nix

@@ -1,4 +1,4 @@
 { inputs, ... }:
 final: _prev: {
-  firefox-addons = inputs.firefox-addons.packages.${final.system};
+  firefox-addons = inputs.firefox-addons.packages.${final.stdenv.hostPlatform.system};
 }

overlays/fish/default.nix

@@ -1,12 +0,0 @@
-# Workaround for aarch64-darwin codesigning bug (nixpkgs#208951 / #507531):
-# fish binaries from the binary cache occasionally have invalid ad-hoc
-# signatures on Apple Silicon. Forcing a local rebuild ensures codesigning
-# is applied on this machine with a valid signature.
-{ inputs, ... }:
-final: prev: {
-  fish = prev.fish.overrideAttrs (_old: {
-    # Bust the cache key so fish is always built locally rather than
-    # substituted from the binary cache where the signature may be stale.
-    NIX_FORCE_LOCAL_REBUILD = "darwin-codesign-fix";
-  });
-}

packages/conduit/default.nix

@@ -5,12 +5,12 @@
 
 buildGoModule rec {
   pname = "conduit";
-  version = "unstable-2026-05-03";
+  version = "unstable-2026-05-15";
 
   src = fetchgit {
     url = "https://gitea.va.reichard.io/evan/conduit.git";
-    rev = "9edea27148670b208c935c070ff3f58a416241b1";
-    hash = "sha256-s8/ghyoAyFOvAMhE7vzckEZ8OxIF116OyJ4Uj30s65A=";
+    rev = "8dfb14f1e7f952bee92cad29703dba55fb156f0c";
+    hash = "sha256-Fc0FHLCNBbEpOFFD0bHSDo1E5AsOzL2fJzHufleKBIo=";
   };
 
   vendorHash = "sha256-LOFT8eCNRm5Q2tVl7ifu4dB5cr828B/E2NJW5WiW0LI=";

packages/ik-llama-cpp/default.nix

@@ -0,0 +1,43 @@
+{ pkgs }:
+let
+  rev = "f9a93c37e2fc021760c3c1aa99cf74c73b7591a7";
+  src = pkgs.fetchFromGitHub {
+    owner = "ikawrakow";
+    repo = "ik_llama.cpp";
+    inherit rev;
+    hash = "sha256-vBVosqBi8FyrllWGJOYsOYaNYAKoTTq6bn+i0Y32pu4=";
+    leaveDotGit = true;
+    postFetch = ''
+      git -C "$out" rev-parse --short HEAD > $out/COMMIT
+      find "$out" -name .git -print0 | xargs -0 rm -rf
+    '';
+  };
+in
+(pkgs.callPackage "${src}/.devops/nix/package.nix" {
+  useCuda = true;
+  useVulkan = true;
+  useBlas = true;
+  useRocm = false;
+  useMetalKit = false;
+}).overrideAttrs
+  (oldAttrs: {
+    inherit src;
+
+    # Add SPIR-V Headers for Vulkan Backend
+    # Newer ggml requires spirv/unified1/spirv.hpp which isn't pulled in by
+    # vulkan-headers alone.
+    buildInputs = (oldAttrs.buildInputs or [ ]) ++ [ pkgs.spirv-headers ];
+
+    # Auto CPU Optimizations + CUDA Arches
+    # Appended after upstream's flags so CMAKE_CUDA_ARCHITECTURES wins.
+    cmakeFlags = (oldAttrs.cmakeFlags or [ ]) ++ [
+      "-DGGML_CUDA_ENABLE_UNIFIED_MEMORY=1"
+      "-DCMAKE_CUDA_ARCHITECTURES=61;86" # GTX 1070 / GTX 1080ti / RTX 3090
+    ];
+
+    # Disable Nix's march=native Stripping
+    preConfigure = ''
+      export NIX_ENFORCE_NO_NATIVE=0
+      ${oldAttrs.preConfigure or ""}
+    '';
+  })

packages/llama-cpp/AGENTS.md

@@ -0,0 +1,35 @@
+# llama-cpp — Agent Notes
+
+Override of `pkgs.llama-cpp` with CUDA + Vulkan + BLAS, custom CMake flags, and an optional fork pin.
+
+## Pitfalls
+
+### `version` must be numeric
+
+Upstream `pkgs/by-name/ll/llama-cpp/package.nix` passes `version` straight through as a C integer via:
+
+```nix
+(cmakeFeature "LLAMA_BUILD_NUMBER" finalAttrs.version)
+```
+
+`build-info.cpp` then emits `int LLAMA_BUILD_NUMBER = <version>;`. A non-numeric `version` (e.g. `"mtp-clean-08b1474"`) breaks the build with:
+
+```
+error: '<value>' was not declared in this scope
+    int LLAMA_BUILD_NUMBER = <value>;
+```
+
+**Convention:**
+- Upstream tag pins: use the bare build number, e.g. `version = "9048";` with `tag = "b${version}";`.
+- Fork / arbitrary commit pins: use a `YYYYMMDD` date derived from the commit's author/commit date (`gh api repos/<owner>/<repo>/commits/<sha>` → `.commit.committer.date`).
+
+### `leaveDotGit` + `postFetch`
+
+We keep `.git` only long enough to record the short SHA into `$out/COMMIT`, then strip it. Preserve this pattern when changing `src` so downstream tooling that reads `COMMIT` keeps working.
+
+## Refreshing the pinned commit (fork)
+
+1. `git ls-remote https://github.com/<owner>/llama.cpp refs/heads/<branch>` → get the full SHA.
+2. `nix run nixpkgs#nix-prefetch-github -- <owner> llama.cpp --rev <sha> --leave-dot-git` → get the hash.
+3. Look up the commit date: `curl -s https://api.github.com/repos/<owner>/llama.cpp/commits/<sha> | jq -r '.commit.committer.date'`.
+4. Update `src.{owner,rev,hash}` and set `version = "YYYYMMDD"`.

packages/llama-cpp/default.nix

@@ -1,4 +1,22 @@
 { pkgs }:
+let
+  # Version MUST be an integer string.
+  # For tagged releases use the tag number (e.g. "9222").
+  # For HEAD builds use YYYYMMDD (e.g. "20260519").
+  version = "9496";
+
+  src = pkgs.fetchFromGitHub {
+    owner = "ggml-org";
+    repo = "llama.cpp";
+    rev = "94a220cd6745e6e3f8de62870b66fd5b9bc92700";
+    hash = "sha256-1jAowfGVzrrHDwWWzKESY7aV82whnuIg1N37fmtcgyw=";
+    leaveDotGit = true;
+    postFetch = ''
+      git -C "$out" rev-parse --short HEAD > $out/COMMIT
+      find "$out" -name .git -print0 | xargs -0 rm -rf
+    '';
+  };
+in
 (pkgs.llama-cpp.override {
   cudaSupport = true;
   blasSupport = true;
@@ -6,19 +24,13 @@
   metalSupport = false;
   vulkanSupport = true;
 }).overrideAttrs
-  (oldAttrs: rec {
-    version = "9009";
-    src = pkgs.fetchFromGitHub {
-      owner = "ggml-org";
-      repo = "llama.cpp";
-      tag = "b${version}";
-      hash = "sha256-rDnO+NIjXYtMqxOeYvecQsB7KgohVUuMV4jG6L0b4S4=";
-      leaveDotGit = true;
-      postFetch = ''
-        git -C "$out" rev-parse --short HEAD > $out/COMMIT
-        find "$out" -name .git -print0 | xargs -0 rm -rf
-      '';
-    };
+  (oldAttrs: {
+    inherit version src;
+
+    # WebUI npm deps hash for our pinned src. Upstream nixpkgs builds the WebUI
+    # from tools/ui via `npm run build` in preConfigure (offline, using these
+    # deps), so no custom webui derivation / HF-bucket workaround is needed.
+    npmDepsHash = "sha256-1iM0LGeI9e+gZEHk46lkBe51DxIhiimfAm9o3Z3m9Ik=";
 
     # Add SPIR-V Headers for Vulkan Backend
     # Newer llama.cpp requires spirv/unified1/spirv.hpp which isn't

packages/llama-swap/default.nix

@@ -13,13 +13,13 @@ let
 in
 buildGo126Module (finalAttrs: {
   pname = "llama-swap";
-  version = "208";
+  version = "216";
 
   src = fetchFromGitHub {
     owner = "mostlygeek";
     repo = "llama-swap";
     tag = "v${finalAttrs.version}";
-    hash = "sha256-E+BqqQcCLlW/DWvjwC66ClV6yuQ5x7cAMkLPJkS3x5M=";
+    hash = "sha256-PHSY4z2h406xL+EcIYyrzr4s28txO7SCsWm8hrXf+2U=";
     # populate values that require us to use git. By doing this in postFetch we
     # can delete .git afterwards and maintain better reproducibility of the src.
     leaveDotGit = true;
@@ -32,10 +32,10 @@ buildGo126Module (finalAttrs: {
     '';
   };
 
-  vendorHash = "sha256-tOOZgugiVcICYg9HyeTolyAg+YZWtxSJTvAuwfMazHQ=";
+  vendorHash = "sha256-QysQ7YdwJcLTziwL25j73n3tQVvzVQIFxN4GkTU8JZg=";
 
   passthru.ui = callPackage ./ui.nix { llama-swap = finalAttrs.finalPackage; };
-  passthru.npmDepsHash = "sha256-6D4F58sSBkr7FKKO34gDhnZ9uN/SfsyYn1xJjYsMeq4=";
+  passthru.npmDepsHash = "sha256-NJqEJ+XTdpPFtJJxP4CGu+JDUW7lKDcFgsixQJ3SXtQ=";
 
   nativeBuildInputs = [
     versionCheckHook

packages/open-proxy/default.nix

@@ -0,0 +1,25 @@
+{ lib
+, buildGoModule
+, fetchgit
+}:
+
+buildGoModule rec {
+  pname = "open-proxy";
+  version = "unstable-2026-06-16";
+
+  src = fetchgit {
+    url = "https://gitea.va.reichard.io/evan/open-proxy.git";
+    rev = "a589341214a1e035b6ce2b2d79870e591a25ccca";
+    hash = "sha256-onfvxOl4TdeRrVLD1oJWcnhEDzKFYU/V0qxV1+NpQrg=";
+  };
+
+  vendorHash = null;
+
+  meta = {
+    description = "Forward `open`/`xdg-open` from a remote VM to the host machine";
+    homepage = "https://gitea.va.reichard.io/evan/open-proxy";
+    license = lib.licenses.mit;
+    maintainers = with lib.maintainers; [ evanreichard ];
+    mainProgram = "open-proxy";
+  };
+}

packages/pi-coding-agent/AGENTS.md

@@ -0,0 +1,23 @@
+# pi-coding-agent Packaging Notes
+
+`pi-coding-agent` is built from the `earendil-works/pi-mono` monorepo with `buildNpmPackage`.
+
+## Lockfile Metadata
+
+Upstream `package-lock.json` may omit `resolved` / `integrity` metadata that npm can recover online, but Nix needs for its offline npm cache. Keep a package-local enriched lockfile at `packages/pi-coding-agent/package-lock.json` and copy it in during `prePatch` before `npmConfigHook` validates/generates `npmDeps`.
+
+After bumping `version` in `default.nix`, regenerate it with:
+
+```bash
+node packages/pi-coding-agent/update-lockfile.mjs
+# or explicitly:
+node packages/pi-coding-agent/update-lockfile.mjs 0.74.0
+```
+
+Then refresh `npmDepsHash` from the FOD mismatch:
+
+```bash
+nix build .#packages.aarch64-linux.pi-coding-agent.npmDeps --no-link
+```
+
+Remember: new files must be `git add`ed before the flake can see them.

packages/pi-coding-agent/default.nix

@@ -3,6 +3,8 @@
 , fetchFromGitHub
 , nodejs
 , nodejs_22
+, firefox
+, geckodriver
 , makeWrapper
 , pkg-config
 , pixman
@@ -16,19 +18,24 @@
 
 buildNpmPackage rec {
   pname = "pi-coding-agent";
-  version = "0.72.1";
+  version = "0.78.1";
 
   src = fetchFromGitHub {
-    owner = "badlogic";
+    owner = "earendil-works";
     repo = "pi-mono";
     rev = "v${version}";
-    hash = "sha256-SqUxghc60P3HfmaFJGB/m23mvzw0cD7cDEUrNFOqo0Y=";
+    hash = "sha256-K5+reVdi9LPwUHxFgM1iFWojuj6M/m25ymhkDOQdBE4=";
   };
 
-  npmDepsHash = "sha256-KUC1xQK6oJXtg962YeLOnO76uTdR10/VNa9iiCdT3VM=";
+  npmDepsHash = "sha256-PknwCOAr61Fq2Mhl6jd79Rdsje1OXFts2MDLM/gIEYE=";
 
   nativeBuildInputs = [ pkg-config makeWrapper ];
 
+  # Restore NPM Metadata - upstream lockfile omits resolved/integrity entries needed by buildNpmPackage.
+  prePatch = ''
+    cp ${./package-lock.json} package-lock.json
+  '';
+
   buildInputs = [
     pixman
     cairo
@@ -41,7 +48,7 @@ buildNpmPackage rec {
   # Skip generate-models in ai package (models.generated.ts already in repo)
   preBuild = ''
     substituteInPlace packages/ai/package.json \
-      --replace-fail '"build": "npm run generate-models && tsgo -p tsconfig.build.json"' \
+      --replace-fail '"build": "npm run generate-models && npm run generate-image-models && tsgo -p tsconfig.build.json"' \
                      '"build": "tsgo -p tsconfig.build.json"'
   '';
 
@@ -73,16 +80,21 @@ buildNpmPackage rec {
       chmod +x $out/bin/pi
 
       wrapProgram $out/bin/pi \
-        --prefix PATH : ${lib.makeBinPath [ nodejs_22 ]}
+        --prefix PATH : ${lib.makeBinPath [
+          nodejs_22
+          # evan/pi-web - Browser automation tools are needed for web-fetch support.
+          firefox
+          geckodriver
+        ]}
 
       runHook postInstall
   '';
 
   meta = {
     description = "Coding agent CLI with read, bash, edit, write tools and session management";
-    homepage = "https://github.com/badlogic/pi-mono";
-    downloadPage = "https://www.npmjs.com/package/@mariozechner/pi-coding-agent";
-    changelog = "https://github.com/badlogic/pi-mono/blob/main/packages/coding-agent/CHANGELOG.md";
+    homepage = "https://github.com/earendil-works/pi-mono";
+    downloadPage = "https://www.npmjs.com/package/@earendil-works/pi-coding-agent";
+    changelog = "https://github.com/earendil-works/pi-mono/blob/main/packages/coding-agent/CHANGELOG.md";
     license = lib.licenses.mit;
     maintainers = with lib.maintainers; [ evanreichard ];
     mainProgram = "pi";

packages/pi-coding-agent/update-lockfile.mjs

@@ -0,0 +1,92 @@
+#!/usr/bin/env node
+import fs from 'node:fs/promises';
+import path from 'node:path';
+import process from 'node:process';
+
+const repoRoot = new URL('../..', import.meta.url);
+const packageDir = new URL('.', import.meta.url);
+const defaultNixPath = new URL('default.nix', packageDir);
+const lockfilePath = new URL('package-lock.json', packageDir);
+const registryCache = new Map();
+
+// Version Selection
+async function getVersion() {
+  const argVersion = process.argv[2];
+  if (argVersion) return argVersion.replace(/^v/, '');
+
+  const defaultNix = await fs.readFile(defaultNixPath, 'utf8');
+  const match = defaultNix.match(/version\s*=\s*"([^"]+)";/);
+  if (!match) throw new Error(`Unable to find version in ${defaultNixPath.pathname}`);
+  return match[1];
+}
+
+// Package Name Extraction
+function packageNameFromLockPath(lockPath) {
+  const parts = lockPath.split('/');
+  const idx = parts.lastIndexOf('node_modules');
+  if (idx < 0 || idx + 1 >= parts.length) return null;
+
+  const first = parts[idx + 1];
+  if (first.startsWith('@')) return `${first}/${parts[idx + 2]}`;
+  return first;
+}
+
+// Registry Fetching
+function registryUrl(name, version) {
+  const encodedName = name.startsWith('@') ? name.replace('/', '%2f') : name;
+  return `https://registry.npmjs.org/${encodedName}/${version}`;
+}
+
+async function fetchPackageMetadata(name, version) {
+  const key = `${name}@${version}`;
+  if (registryCache.has(key)) return registryCache.get(key);
+
+  const response = await fetch(registryUrl(name, version), {
+    headers: { accept: 'application/json' },
+  });
+  if (!response.ok) {
+    throw new Error(`Failed to fetch ${key}: HTTP ${response.status}`);
+  }
+
+  const metadata = await response.json();
+  const dist = metadata.dist;
+  if (!dist?.tarball || !dist?.integrity) {
+    throw new Error(`Missing dist.tarball/dist.integrity for ${key}`);
+  }
+
+  registryCache.set(key, dist);
+  return dist;
+}
+
+// Lockfile Enrichment
+async function enrichLockfile(lock) {
+  let updated = 0;
+
+  for (const [lockPath, entry] of Object.entries(lock.packages ?? {})) {
+    if (!lockPath || entry.link || !entry.version || entry.resolved) continue;
+    if (!lockPath.includes('node_modules')) continue;
+
+    const name = packageNameFromLockPath(lockPath);
+    if (!name) continue;
+
+    const dist = await fetchPackageMetadata(name, entry.version);
+    entry.resolved = dist.tarball;
+    entry.integrity = dist.integrity;
+    updated += 1;
+  }
+
+  return updated;
+}
+
+// Main
+const version = await getVersion();
+const lockUrl = `https://raw.githubusercontent.com/earendil-works/pi-mono/v${version}/package-lock.json`;
+const response = await fetch(lockUrl, { headers: { accept: 'application/json' } });
+if (!response.ok) throw new Error(`Failed to fetch ${lockUrl}: HTTP ${response.status}`);
+
+const lock = await response.json();
+const updated = await enrichLockfile(lock);
+await fs.writeFile(lockfilePath, JSON.stringify(lock, null, 2) + '\n');
+
+const displayPath = path.relative(repoRoot.pathname, lockfilePath.pathname);
+console.error(`Wrote ${displayPath} for v${version}; restored metadata for ${updated} entries.`);

packages/qwen-code/default.nix

@@ -11,16 +11,18 @@
 }:
 buildNpmPackage (finalAttrs: {
   pname = "qwen-code";
-  version = "0.4.0-nightly.20251209.a6a57233";
+  version = "0.16.0-preview.0";
 
   src = fetchFromGitHub {
     owner = "QwenLM";
     repo = "qwen-code";
     tag = "v${finalAttrs.version}";
-    hash = "sha256-s9m1IN6jDDbNPr/vI/UcrauYPiyQTDODarLP3EvnG3Y=";
+    hash = "sha256-UAJNw1RjHRoZqtgIWJ1dOTWnE9LoBpfJCAM0Jay+VPI=";
   };
 
-  npmDepsHash = "sha256-ngAjCCoHLPZ+GgBRmAKbRYaF7l+RK3YGf1kEkwFbyQg=";
+  npmDepsHash = "sha256-uJtOeNnhbGE7EzTwkNbg2EHLonjHCbdPH5rcV2bgQUw=";
+  makeCacheWritable = true;
+  npmFlags = [ "--legacy-peer-deps" ];
 
   nativeBuildInputs = [
     jq
@@ -34,8 +36,13 @@ buildNpmPackage (finalAttrs: {
     libsecret
   ];
 
-  postPatch = ''
+  prePatch = ''
+    ${jq}/bin/jq '.dependencies."iconv-lite" = "^0.7.0"' \
+      packages/core/package.json > packages/core/package.json.tmp
+    mv packages/core/package.json.tmp packages/core/package.json
+
     ${jq}/bin/jq '
+      .packages."packages/core".dependencies."iconv-lite" = "^0.7.0" |
       del(.packages."node_modules/node-pty") |
       del(.packages."node_modules/@lydell/node-pty") |
       del(.packages."node_modules/@lydell/node-pty-darwin-arm64") |
@@ -62,9 +69,26 @@ buildNpmPackage (finalAttrs: {
     ' package-lock.json > package-lock.json.tmp && mv package-lock.json.tmp package-lock.json
   '';
 
+  preBuild = ''
+    mkdir -p node_modules/@lydell/node-pty
+    printf '%s\n' \
+      'export interface IPty {' \
+      '  pid: number;' \
+      '  onData(callback: (data: string) => void): void;' \
+      '  onExit(callback: (event: { exitCode: number; signal?: number }) => void): void;' \
+      '  kill(signal?: string): void;' \
+      '  write(data: string): void;' \
+      '  resize(columns: number, rows: number): void;' \
+      '  removeListener(event: string, listener: (...args: unknown[]) => void): void;' \
+      '  exitCode?: number;' \
+      '}' \
+      > node_modules/@lydell/node-pty/node-pty.d.ts
+  '';
+
   buildPhase = ''
     runHook preBuild
     npm run generate
+    npm run build
     npm run bundle
     runHook postBuild
   '';
@@ -75,6 +99,14 @@ buildNpmPackage (finalAttrs: {
     cp -r dist/* $out/share/qwen-code/
     npm prune --production
     cp -r node_modules $out/share/qwen-code/
+    if [ -d $out/share/qwen-code/vendor/ripgrep ]; then
+      find $out/share/qwen-code/vendor/ripgrep -type f -name rg -exec sh -c '
+        for rg; do
+          rm "$rg"
+          ln -s ${ripgrep}/bin/rg "$rg"
+        done
+      ' sh {} +
+    fi
     find $out/share/qwen-code/node_modules -type l -delete || true
     patchShebangs $out/share/qwen-code
     ln -s $out/share/qwen-code/cli.js $out/bin/qwen

packages/tuxguitar/default.nix

@@ -0,0 +1,114 @@
+{ lib
+, stdenv
+, maven
+, fetchFromGitHub
+, jdk17
+, jre
+, swt
+, makeWrapper
+, wrapGAppsHook3
+, pkg-config
+, alsa-lib
+, jack2
+, fluidsynth
+, libpulseaudio
+, lilv
+, suil
+, qt5
+, which
+}:
+
+maven.buildMavenPackage rec {
+  pname = "tuxguitar";
+  version = "2.0.1";
+
+  src = fetchFromGitHub {
+    owner = "helge17";
+    repo = "tuxguitar";
+    rev = version;
+    hash = "sha256-USdYj8ebosXkiZpDqyN5J+g1kjyWm225iQlx/szXmLA=";
+  };
+
+  mvnHash = "sha256-XTODH8SG7iwhACJT4AbIokORUe00r6theV18TEXbrIs=";
+
+  doCheck = false;
+
+  mvnJdk = jdk17;
+
+  nativeBuildInputs = [
+    makeWrapper
+    pkg-config
+    wrapGAppsHook3
+  ];
+
+  buildInputs = [
+    alsa-lib
+    fluidsynth
+    jack2
+    lilv
+    qt5.qtbase
+    suil
+  ];
+
+  mvnFetchExtraArgs = {
+    inherit buildInputs;
+    dontWrapQtApps = true;
+  };
+
+  postPatch = ''
+    substituteInPlace desktop/build-scripts/native-modules/tuxguitar-synth-lv2-linux/pom.xml \
+      --replace-fail /usr/include/lilv-0/lilv ${lib.getDev lilv}/include/lilv-0/lilv \
+      --replace-fail /usr/include/suil-0/suil ${lib.getDev suil}/include/suil-0/suil
+
+    if [[ "$name" == maven-deps-* ]]; then
+      mvn install:install-file \
+        -Dfile=${swt}/jars/swt.jar \
+        -DgroupId=org.eclipse.swt \
+        -DartifactId=org.eclipse.swt.gtk.linux \
+        -Dpackaging=jar \
+        -Dversion=4.36 \
+        -Dmaven.repo.local=$out/.m2
+    fi
+  '';
+
+  mvnParameters = "-f desktop/build-scripts/tuxguitar-linux-swt/pom.xml verify -P native-modules";
+
+  dontWrapGApps = true;
+  dontWrapQtApps = true;
+
+  installPhase = ''
+    runHook preInstall
+
+    mkdir -p $out/bin
+    cp -r desktop/build-scripts/tuxguitar-linux-swt/target/tuxguitar-*-linux-swt/{dist,lib,share,tuxguitar.sh} $out/
+    ln -sf ${swt}/jars/swt.jar $out/lib/swt.jar
+    ln -s ../tuxguitar.sh $out/bin/tuxguitar
+
+    runHook postInstall
+  '';
+
+  postFixup = ''
+    wrapProgram $out/tuxguitar.sh \
+      "''${gappsWrapperArgs[@]}" \
+      --prefix PATH : ${lib.makeBinPath [ jre which ]} \
+      --prefix LD_LIBRARY_PATH : "$out/lib:${lib.makeLibraryPath [
+        swt
+        alsa-lib
+        fluidsynth
+        jack2
+        libpulseaudio
+        lilv
+        qt5.qtbase
+        suil
+      ]}"
+  '';
+
+  meta = {
+    description = "Multitrack guitar tablature editor";
+    homepage = "https://github.com/helge17/tuxguitar";
+    license = lib.licenses.lgpl2;
+    maintainers = with lib.maintainers; [ evanreichard ];
+    mainProgram = "tuxguitar";
+    platforms = lib.platforms.linux;
+  };
+}

secrets/common/evanreichard.yaml

@@ -1,3 +1,4 @@
+open_proxy_token: ENC[AES256_GCM,data:LxC0dR2EQ8XPmw4fwKnKJD3usqImMKH+81I9RSTNsjg=,iv:LQmf+kxtwgAMxcHvIe6y3Qw/oxXvdWGbyV/kdwPpKw4=,tag:TUQbM8sIK6KB7eOsYfHuqw==,type:str]
 conduit_apikey: ENC[AES256_GCM,data:4mjvEI00V7nAhPkDa97eOfLCqItxoRALFe8OdxzUiUc=,iv:2mtSu1LDUvaneTnqs3Z4GVAj+HuAY2+VXrpIITSg/64=,tag:trkgxyX0ssCjyKMB42bFGg==,type:str]
 context7_apikey: ENC[AES256_GCM,data:K8/OoJMWBhN3ufmTa/tAiD3iMergDZQ1OBucUtLsrg+L26DXDPAko9D41w==,iv:/IVpaaPivUTn2rbIAPIwyN5nb7TmtDh05YlMdOlBkhE=,tag:0XJfoNlDelBwMXMAAqKjtQ==,type:str]
 zai_apikey: ENC[AES256_GCM,data:eNgIfEqs8JGM7Qo6D5KMMqRF8fd1qLakYQ9F5oEDUvLqPJ+TAktz8GMVuSndwW5BxA==,iv:eR8IR/MDmhk2JUoT2chCwRYOJGfxEBFGARf1CI7EG8Q=,tag:3fmRWA5eof304WSWKntDFg==,type:str]
@@ -5,8 +6,7 @@ kagi_token: ENC[AES256_GCM,data:6pxxMMQ3RCy6sdUFiuAy8rUzsIMMiBgPzphpgTVMfiHC98ej
 rke2_kubeconfig: ENC[AES256_GCM,data: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,iv:CMrKYb+2QZVKEJMjW51rbiYW/cN6ATDzgwfBdSi9B10=,tag:qVqO5byXdj7DZdaHNx7S9A==,type:str]
 sops:
     age:
-        - recipient: age1sac93wpnjcv62s7583jv6a4yspndh6k0r25g3qx3k7gq748uvafst6nz4w
-          enc: |
+        - enc: |
             -----BEGIN AGE ENCRYPTED FILE-----
             YWdlLWVuY3J5cHRpb24ub3JnL3YxCi0+IFgyNTUxOSBrY0pseTB5R290aXAzdnB3
             aTVMS2tBdnEvSjR5K241dElEa1Z1Tmt2a0ZJCnRSbVBsZ3lNVlNzdnFlY3VvaWpB
@@ -14,8 +14,8 @@ sops:
             S3hOa3p1MHVLek5zaUxhWXFiV2Z3R0UKxSrnYSoN6KcuFdg5K6qwcwh9/j9lI0HB
             HqujumuIfWkcctNk38AMn4beeesmXsbJQcUPHUVOZQw6Ov4jXaGz/Q==
             -----END AGE ENCRYPTED FILE-----
-        - recipient: age17ayje4uv2mhwehhp9jr3u9l0ds07396kt7ef40sufx89vm7cgfjq6d5d4y
-          enc: |
+          recipient: age1sac93wpnjcv62s7583jv6a4yspndh6k0r25g3qx3k7gq748uvafst6nz4w
+        - enc: |
             -----BEGIN AGE ENCRYPTED FILE-----
             YWdlLWVuY3J5cHRpb24ub3JnL3YxCi0+IFgyNTUxOSBMazJ5VmNXQ2FRYzFuejBU
             bWNPWnI2a2R3NVBpY0Y2N0lXd1RXbUdmTkc0ClZDTSs4bklpeVEyMDNrVXJXRHRF
@@ -23,8 +23,8 @@ sops:
             a3dPbEZtL0tsOWpWSEtPblZSYk9NVTgKM6TfK7VX6v059FXpRjpAlgX+ab4f6vq2
             jH8jyO33YxQYI1kSgXJ5AR8evCoV3FzbZ5rzIy2PRmCOwFV8Im2bRg==
             -----END AGE ENCRYPTED FILE-----
-        - recipient: age1mar507c9mxmwalg486chs5kfh0mya38rv5w64ypfwnwlawewrpnswerpg8
-          enc: |
+          recipient: age17ayje4uv2mhwehhp9jr3u9l0ds07396kt7ef40sufx89vm7cgfjq6d5d4y
+        - enc: |
             -----BEGIN AGE ENCRYPTED FILE-----
             YWdlLWVuY3J5cHRpb24ub3JnL3YxCi0+IFgyNTUxOSB4S0NrbGF4WXN5KzY4S0lt
             ZW1PMWE5NUF6NEIyWmp6RWY4MzNYbE0wQnpFCkdOaGtIWmtJT08zNmZNNzcvQldp
@@ -32,8 +32,8 @@ sops:
             VFJrQWNnNTJ4azVpMElHbmsrZnJPM2MKx/7XxkZfd1tPMck9FmoM6g28dp5JeXQ5
             OdiOLlKc2If1f6dLKkjDmmscMui6aLMQ8RJ8dLK7FKlYy+95VsHVrw==
             -----END AGE ENCRYPTED FILE-----
-        - recipient: age1w6avj7gd4f5frk90lsyh4e2k5am6z92hzlr0vpgrm767muyj59qsnuah62
-          enc: |
+          recipient: age1mar507c9mxmwalg486chs5kfh0mya38rv5w64ypfwnwlawewrpnswerpg8
+        - enc: |
             -----BEGIN AGE ENCRYPTED FILE-----
             YWdlLWVuY3J5cHRpb24ub3JnL3YxCi0+IFgyNTUxOSBnZFpKSVpWMCt6Z3M5UGUz
             YVI4TG85UVh1eXYwVStQWk5xdEhhSjdNTzFJCk1mNnBtSVV0VFB0TnFIeVdocWFP
@@ -41,8 +41,8 @@ sops:
             MzNuSHlWWnNMYkJMMHZkcnh0cXZaREkKE+j0yWV/zK8lz8vRa0cywpLL2DiAFsgi
             fgCdeysSacrQLxB8iBWbusJ31ktyJMYLrsWFAdPkl0WN6HjaR3k3CQ==
             -----END AGE ENCRYPTED FILE-----
-        - recipient: age1avlhszrryt4gf4ya536jhzm7qwt9xfttm8x4sns6h9w2tahzqp8sspz9y5
-          enc: |
+          recipient: age1w6avj7gd4f5frk90lsyh4e2k5am6z92hzlr0vpgrm767muyj59qsnuah62
+        - enc: |
             -----BEGIN AGE ENCRYPTED FILE-----
             YWdlLWVuY3J5cHRpb24ub3JnL3YxCi0+IFgyNTUxOSBDUFYyTzdpQlJ4VVJJbDF5
             cEhIQ1NSRVcwRUFBUzdaeXdkOWhPVjFIdGtBCk5qVmx1L09kVGhiNnlaQ0xoeTVz
@@ -50,8 +50,8 @@ sops:
             VmFINkwvZEFuWnkyb2JpTCtmRUVBbUkKKzbifH6Ue84MkpaLHrwDvJu8uvjY7yOY
             +qYg0rOqFuZAx9YiOjDR7JVeGpfHM+7pO9ZjSNTPH0f1NC3XwsNp1A==
             -----END AGE ENCRYPTED FILE-----
-        - recipient: age1dccte7xtwswgef089nd80dutp96xnezx5lrqnneh9cusegsnda8sj3dj6c
-          enc: |
+          recipient: age1avlhszrryt4gf4ya536jhzm7qwt9xfttm8x4sns6h9w2tahzqp8sspz9y5
+        - enc: |
             -----BEGIN AGE ENCRYPTED FILE-----
             YWdlLWVuY3J5cHRpb24ub3JnL3YxCi0+IFgyNTUxOSBBSS9CQlE0azMwbFo4V0Ez
             V3pKL2QvY0NNdmoxZXhQRUU1cG9Temk5M3pZCjBZUHJFRmZ1bTZzQ2RGU2F3WXF5
@@ -59,8 +59,8 @@ sops:
             Zm1ReDlMWW8wak5Jb0VncFV2bFJROWsKWi9DTgveMgGG8eK4qNeAGGG/gfiJS96G
             232Tgf94Pb8eAU2zbF77pLWMaqTBbYPz1tggcMTfrAeDohq+/0sU8g==
             -----END AGE ENCRYPTED FILE-----
-        - recipient: age1ped3hpugq06908ex8kgama33qckqe03rmac5pa6th87vks5d249qhshvqu
-          enc: |
+          recipient: age1dccte7xtwswgef089nd80dutp96xnezx5lrqnneh9cusegsnda8sj3dj6c
+        - enc: |
             -----BEGIN AGE ENCRYPTED FILE-----
             YWdlLWVuY3J5cHRpb24ub3JnL3YxCi0+IFgyNTUxOSBqS0p2TVhWd2VMQTNNR0or
             amhNdTh5bmY3WWh1QkgveWZEMEtxSCt3Tng4CnVEKzhDMFVzRktndmI5OFhSei94
@@ -68,7 +68,8 @@ sops:
             TC91cmtZWm03dzYwS2E3dkorNkdFY2sKj5OZHOtKx1NGPSGKsWjC/8+seUAhvmxb
             wQ0iuPAq6yDLhYV69n7Jx4G9fKoidLIQxq+Ia+tLcYt58UDX7aixJQ==
             -----END AGE ENCRYPTED FILE-----
-    lastmodified: "2026-05-04T03:17:35Z"
-    mac: ENC[AES256_GCM,data:hFnRRENIaWa+Yidk7JqY+XMZ4xyekQLY9MMSgo5j4w7UI+Fb82RpNjmFrOS13ck7kKS+CVLWrFSEDxWJYcjHyLE5btoi06SFiZYfDF0JlOfq6e+loTj+TqhyWyJoQ3ZYaj9weak+qCroN4q88EWrle3Rlw0b/67lnkkzYJYYOlk=,iv:i9DVooO/9Gjr+yWaEkdXEWABAyrp8qNPt48EzqoEfiA=,tag:8hYmoJ9p3fse9eckhYqkvg==,type:str]
+          recipient: age1ped3hpugq06908ex8kgama33qckqe03rmac5pa6th87vks5d249qhshvqu
+    lastmodified: "2026-06-16T18:49:15Z"
+    mac: ENC[AES256_GCM,data:Q51p1A317BYzKvXSpkx2HBduGLGvxdQFi/BCfUtKWV9uAJGlQDp//eGJ7kJEG6DIO6sWUddf75fBLYAQcqm3iogIetTLUuQl3OhHSpAPvGpUDC6Hh87sAZy/ebaN2cy7BhJy5cjPJ9JAkfHqLCFRP+cVIwS/eb87GzwvWdSGZbc=,iv:yTTwhHCB09Qb6a437VENFiWQPp8CHwd0TelFj4ugO3U=,tag:JXN58pSy2I04O44Hg4pQcw==,type:str]
     unencrypted_suffix: _unencrypted
-    version: 3.12.1
+    version: 3.13.1

systems/aarch64-linux/lin-o1-headscale/default.nix

@@ -15,7 +15,7 @@ in
     (modulesPath + "/profiles/qemu-guest.nix")
   ];
 
-  system.stateVersion = "25.11";
+  system.stateVersion = "26.05";
   time.timeZone = "UTC";
 
   boot.loader.grub = {

systems/aarch64-linux/lin-o1-node/default.nix

@@ -12,7 +12,7 @@ in
     (modulesPath + "/profiles/qemu-guest.nix")
   ];
 
-  system.stateVersion = "25.11";
+  system.stateVersion = "26.05";
   time.timeZone = "UTC";
 
   boot.loader.grub = {

systems/aarch64-linux/lin-va-mbp-personal/default.nix

@@ -11,12 +11,20 @@ in
     ./hardware-configuration.nix
   ];
 
-  system.stateVersion = "25.11";
+  system.stateVersion = "26.05";
   time.timeZone = "America/New_York";
 
   programs.firejail.enable = true;
   programs.nix-ld.enable = true;
 
+  # Asahi Wi-Fi Resume Bug - The Broadcom driver can fail to reconnect after suspend on this MacBook.
+  powerManagement.resumeCommands = ''
+    ${pkgs.kmod}/bin/modprobe -r brcmfmac_wcc 2>/dev/null || true
+    ${pkgs.kmod}/bin/modprobe -r brcmfmac 2>/dev/null || true
+    ${pkgs.kmod}/bin/modprobe brcmfmac
+    ${pkgs.systemd}/bin/systemctl restart NetworkManager.service
+  '';
+
   # System Config
   reichard = {
     nix = enabled;

systems/aarch64-linux/lin-va-mbp-work-vm/default.nix

@@ -11,7 +11,7 @@ in
     (modulesPath + "/profiles/qemu-guest.nix")
   ];
 
-  system.stateVersion = "25.11";
+  system.stateVersion = "26.05";
   time.timeZone = "America/New_York";
 
   programs.nix-ld.enable = true;

systems/aarch64-raw-efi/lin-terminal-image/default.nix

@@ -3,7 +3,7 @@ let
   inherit (lib.${namespace}) enabled;
 in
 {
-  system.stateVersion = "25.11";
+  system.stateVersion = "26.05";
   time.timeZone = "America/New_York";
 
   # Config Boot

systems/x86_64-linux/lin-o1-x86-node/default.nix

@@ -15,7 +15,7 @@ in
     (modulesPath + "/profiles/qemu-guest.nix")
   ];
 
-  system.stateVersion = "25.11";
+  system.stateVersion = "26.05";
   time.timeZone = "UTC";
 
   boot.loader.grub = {

systems/x86_64-linux/lin-ovh-kube1/default.nix

@@ -15,7 +15,7 @@ in
     (modulesPath + "/profiles/qemu-guest.nix")
   ];
 
-  system.stateVersion = "25.11";
+  system.stateVersion = "26.05";
   time.timeZone = "UTC";
 
   networking.firewall.allowedTCPPorts = [ 443 ];

systems/x86_64-linux/lin-ssd-kube1/default.nix

@@ -12,7 +12,7 @@ in
     (modulesPath + "/profiles/qemu-guest.nix")
   ];
 
-  system.stateVersion = "25.11";
+  system.stateVersion = "26.05";
   time.timeZone = "UTC";
 
   networking.firewall.allowedTCPPorts = [ 443 ];

systems/x86_64-linux/lin-va-desktop/default.nix

@@ -9,12 +9,13 @@ let
 
   nvidia-smi = "${config.hardware.nvidia.package.bin}/bin/nvidia-smi";
   llama-cpp = pkgs.reichard.llama-cpp;
+  ik-llama-cpp = pkgs.reichard.ik-llama-cpp;
   stable-diffusion-cpp = pkgs.reichard.stable-diffusion-cpp.override {
     cudaSupport = true;
   };
 in
 {
-  system.stateVersion = "25.11";
+  system.stateVersion = "26.05";
   time.timeZone = "America/New_York";
   nixpkgs.config.allowUnfree = true;
   hardware.nvidia-container-toolkit.enable = true;
@@ -65,7 +66,7 @@ in
       wantedBy = [ "multi-user.target" ];
       serviceConfig.Type = "oneshot";
       serviceConfig.RemainAfterExit = true;
-      script = "${nvidia-smi} -i 0 -pl 250";
+      script = "${nvidia-smi} -i 0 -pl 290";
     };
   };
 
@@ -99,6 +100,8 @@ in
       opengl = {
         enable = true;
         enableNvidia = true;
+        # GTX 1080 Ti is Pascal; NVIDIA 590+ (nixpkgs stable = 595) dropped Pascal support.
+        nvidiaPackage = config.boot.kernelPackages.nvidiaPackages.legacy_580;
       };
     };
 
@@ -129,6 +132,7 @@ in
 
     # Local Packages
     llama-cpp
+    ik-llama-cpp
     stable-diffusion-cpp
   ];
 }

systems/x86_64-linux/lin-va-nix-builder/default.nix

@@ -3,7 +3,7 @@
 }:
 {
   time.timeZone = "America/New_York";
-  system.stateVersion = "25.11";
+  system.stateVersion = "26.05";
 
   reichard = {
     system = {

systems/x86_64-linux/lin-va-office/default.nix

@@ -9,7 +9,7 @@ let
   cfg = config.${namespace}.user;
 in
 {
-  system.stateVersion = "25.11";
+  system.stateVersion = "26.05";
   time.timeZone = "America/New_York";
 
   nixpkgs.config.allowUnfree = true;

systems/x86_64-linux/lin-va-terminal/default.nix

@@ -3,9 +3,11 @@ let
   inherit (lib.${namespace}) enabled;
 in
 {
-  system.stateVersion = "25.11";
+  system.stateVersion = "26.05";
   time.timeZone = "America/New_York";
 
+  programs.nix-ld.enable = true;
+
   reichard = {
     nix = enabled;
 

systems/x86_64-linux/lin-va-thinkpad/default.nix

@@ -7,15 +7,10 @@ let
   inherit (lib.${namespace}) enabled;
 in
 {
-  system.stateVersion = "25.11";
+  system.stateVersion = "26.05";
   time.timeZone = "America/New_York";
 
-  boot = {
-    kernelParams = [
-      # Mask GPE03 (EC wakeup events) to allow hibernation without spurious CPU wakeups
-      "acpi_mask_gpe=0x03"
-    ];
-  };
+  programs.nix-ld.enable = true;
 
   hardware = {
     enableRedistributableFirmware = true;
@@ -83,14 +78,4 @@ in
       sops = enabled;
     };
   };
-
-  # Additional System Packages
-  environment.systemPackages = with pkgs; [
-    dool
-    jq
-    mosh
-    rclone
-    sqlite-interactive
-    unzip
-  ];
 }

systems/x86_64-vmware/lin-va-rke2/default.nix

@@ -13,7 +13,7 @@ in
 
   config = {
     # Basic System
-    system.stateVersion = "25.11";
+    system.stateVersion = "26.05";
     time.timeZone = "UTC";
 
     reichard = {