fixing update-flake-lock.yaml permissions

sunshine is a cool idea but has been causing annoying ui glitches and started preventing the display manning for starting
Its a cool idea in theory but not useful enough for me to want to debug

.github/workflows/build_systems.yml

@@ -23,6 +23,6 @@ jobs:
     steps:
       - uses: actions/checkout@v4
       - name: Build default package
-        run: "nixos-rebuild build --flake ./#${{ matrix.system }}"
+        run: "nixos-rebuild build --accept-flake-config --flake ./#${{ matrix.system }}"
       - name: copy to nix-cache
         run: nix copy --accept-flake-config --to unix:///host-nix/var/nix/daemon-socket/socket .#nixosConfigurations.${{ matrix.system }}.config.system.build.toplevel

.github/workflows/fix_eval_warnings.yml

@@ -1,30 +0,0 @@
-name: fix_eval_warnings
-on:
-  workflow_run:
-    workflows: ["build_systems"]
-    types: [completed]
-
-jobs:
-  check-warnings:
-    if: >-
-      github.event.workflow_run.conclusion != 'cancelled' &&
-      github.event.workflow_run.head_branch == 'main' &&
-      (github.event.workflow_run.event == 'push' || github.event.workflow_run.event == 'schedule')
-    runs-on: self-hosted
-    permissions:
-      contents: write
-      pull-requests: write
-
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Fix eval warnings
-        env:
-          GH_TOKEN: ${{ secrets.GH_TOKEN_FOR_UPDATES }}
-        run: >-
-          nix develop .#devShells.x86_64-linux.default -c
-          python -m python.eval_warnings.main
-          --run-id "${{ github.event.workflow_run.id }}"
-          --repo "${{ github.repository }}"
-          --ollama-url "${{ secrets.OLLAMA_URL }}"
-          --run-url "${{ github.event.workflow_run.html_url }}"

.github/workflows/merge_flake_lock_update.yml

@@ -6,24 +6,18 @@ on:
 
 jobs:
   merge:
-    runs-on: ubuntu-latest
+    runs-on: self-hosted
 
     permissions:
       contents: write
       pull-requests: write
 
     steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
       - name: merge_flake_lock_update
-        run: |
+        run: >-
-          pr_number=$(gh pr list --state open --author RichieCahill --label flake_lock_update --json number --jq '.[0].number')
+          nix develop .#devShells.x86_64-linux.default -c
-          echo "pr_number=$pr_number" >> $GITHUB_ENV
+          python -m python.gitea_flake_lock merge
-          if [ -n "$pr_number" ]; then
+          --repo "${{ github.repository }}"
-            gh pr merge "$pr_number" --rebase
-          else
-            echo "No open PR found with label flake_lock_update"
-          fi
         env:
-          GITHUB_TOKEN: ${{ secrets.GH_TOKEN_FOR_UPDATES }}
+          GITEA_TOKEN: ${{ secrets.GITEA_TOKEN }}
+          GITEA_URL: https://gitea.tmmworkshop.com

.github/workflows/pytest.yml

@@ -1,13 +1,13 @@
 name: pytest
 
 on:
+  workflow_dispatch:
   push:
     branches:
       - main
   pull_request:
     branches:
       - main
-  merge_group:
 
 jobs:
   pytest:

.github/workflows/update-flake-lock.yml

@@ -6,18 +6,21 @@ on:
 
 jobs:
   lockfile:
-    runs-on: ubuntu-latest
+    runs-on: self-hosted
+    permissions:
+      actions: write
+      contents: write
+      pull-requests: write
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
-      - name: Install Nix
-        uses: DeterminateSystems/nix-installer-action@main
       - name: Update flake.lock
-        uses: DeterminateSystems/update-flake-lock@main
+        run: nix flake update
-        with:
+      - name: Create or update flake.lock PR
-          token: ${{ secrets.GH_TOKEN_FOR_UPDATES }}
+        env:
-          pr-title: "Update flake.lock"
+          GITEA_TOKEN: ${{ secrets.GITEA_TOKEN }}
-          pr-labels: |
+          GITEA_URL: https://gitea.tmmworkshop.com
-            dependencies
+        run: >-
-            automated
+          nix develop .#devShells.x86_64-linux.default -c
-            flake_lock_update
+          python -m python.gitea_flake_lock update
+          --repo "${{ github.repository }}"

.gitignore

@@ -170,6 +170,5 @@ test.*
 frontend/dist/
 frontend/node_modules/
 
-# data dir for training, validation, and testing
+# data from testing llms
-data/
+data/*
-config.toml

.vscode/settings.json

@@ -40,7 +40,6 @@
     "cgroupdriver",
     "charliermarsh",
     "Checkpointing",
-    "cloudflared",
     "codellama",
     "codezombiech",
     "compactmode",
@@ -204,6 +203,7 @@
     "peerconnection",
     "PESKYFOX",
     "PGID",
+    "pgvector",
     "pipewire",
     "pkgs",
     "plugdev",

common/global/default.nix

@@ -23,7 +23,10 @@
   boot = {
     tmp.useTmpfs = true;
     kernelPackages = lib.mkDefault pkgs.linuxPackages_6_12;
-    zfs.package = lib.mkDefault pkgs.zfs_2_4;
+    zfs = {
+      package = lib.mkDefault pkgs.zfs_2_4;
+      forceImportRoot = lib.mkDefault false;
+    };
   };
 
   hardware.enableRedistributableFirmware = true;
@@ -37,10 +40,17 @@
 
   nixpkgs = {
     overlays = builtins.attrValues outputs.overlays;
-    config.allowUnfree = true;
+    config = {
+      allowUnfree = true;
+      permittedInsecurePackages = [
+        "openssl-1.1.1w" # This is for discord-canary
+      ];
+    };
   };
 
   services = {
+    dbus.implementation = "dbus";
+
     # firmware update
     fwupd.enable = true;
 

common/global/nix.nix

@@ -34,6 +34,7 @@ in
       warn-dirty = false;
       flake-registry = ""; # disable global flake registries
       connect-timeout = 10;
+      download-buffer-size = 536870912;
       fallback = true;
     };
 

common/optional/monitoring-agent.nix

@@ -0,0 +1,256 @@
+{
+  config,
+  lib,
+  pkgs,
+  ...
+}:
+let
+  monitoringInterface = "ztwfunumly";
+  nodeTextfileDir = "/var/lib/prometheus-node-exporter-textfile";
+
+  mkProcessNameTemplate =
+    perPid: template: if perPid then "${template}:{{.PID}}:{{.StartTime}}" else template;
+
+  mkProcessMatchers = perPid: [
+    {
+      name = mkProcessNameTemplate perPid "{{.Username}}:{{.Matches.Module}}";
+      cmdline = [ "^/nix/store[^ ]*/bin/python[^ ]* -m (?P<Module>[^ ]+)" ];
+    }
+    {
+      name = mkProcessNameTemplate perPid "{{.Username}}:{{.Matches.Wrapped}}";
+      cmdline = [
+        "^/nix/store[^ ]*/bin/python[^ ]* /nix/store[^ ]*/bin/\\.?(?P<Wrapped>[^ /]+?)(?:-wrapped)?(?:\\s|$)"
+      ];
+    }
+    {
+      name = mkProcessNameTemplate perPid "{{.Username}}:{{.Matches.Wrapped}}";
+      cmdline = [
+        "^/nix/store[^ ]*/bin/node /nix/store[^ ]*-(?P<Wrapped>[A-Za-z0-9._+-]+)-[0-9][^ /]*/"
+      ];
+    }
+    {
+      name = mkProcessNameTemplate perPid "{{.Username}}:{{.Matches.Wrapped}}";
+      cmdline = [ "^/nix/store[^ ]*/(?:bin/|lib/[^ ]*/)?\\.?(?P<Wrapped>[^ /]+?)(?:-wrapped)?(?:\\s|$)" ];
+    }
+    {
+      name = mkProcessNameTemplate perPid "{{.Username}}:{{.ExeBase}}";
+      cmdline = [ ".+" ];
+    }
+  ];
+
+  perPidConfig = pkgs.writeText "process-exporter-per-pid.yaml" (
+    builtins.toJSON {
+      process_names = mkProcessMatchers true;
+    }
+  );
+
+  zpoolLatencyScript = pkgs.writeShellScript "zpool-latency-exporter" ''
+        set -euo pipefail
+
+        out_dir=${lib.escapeShellArg nodeTextfileDir}
+        host=${lib.escapeShellArg config.networking.hostName}
+        tmp_file="$(mktemp "$out_dir/zpool.prom.XXXXXX")"
+        trap 'rm -f "$tmp_file"' EXIT
+
+        pools="$(zpool list -H -o name | paste -sd, -)"
+
+        cat >"$tmp_file" <<'EOF'
+    # HELP zpool_iostat_total_wait_read_ns Average total read wait time reported by zpool iostat.
+    # TYPE zpool_iostat_total_wait_read_ns gauge
+    # HELP zpool_iostat_total_wait_write_ns Average total write wait time reported by zpool iostat.
+    # TYPE zpool_iostat_total_wait_write_ns gauge
+    # HELP zpool_iostat_disk_wait_read_ns Average disk read wait time reported by zpool iostat.
+    # TYPE zpool_iostat_disk_wait_read_ns gauge
+    # HELP zpool_iostat_disk_wait_write_ns Average disk write wait time reported by zpool iostat.
+    # TYPE zpool_iostat_disk_wait_write_ns gauge
+    # HELP zpool_iostat_syncq_wait_read_ns Average synchronous queue read wait time reported by zpool iostat.
+    # TYPE zpool_iostat_syncq_wait_read_ns gauge
+    # HELP zpool_iostat_syncq_wait_write_ns Average synchronous queue write wait time reported by zpool iostat.
+    # TYPE zpool_iostat_syncq_wait_write_ns gauge
+    # HELP zpool_iostat_asyncq_wait_read_ns Average asynchronous queue read wait time reported by zpool iostat.
+    # TYPE zpool_iostat_asyncq_wait_read_ns gauge
+    # HELP zpool_iostat_asyncq_wait_write_ns Average asynchronous queue write wait time reported by zpool iostat.
+    # TYPE zpool_iostat_asyncq_wait_write_ns gauge
+    EOF
+
+        zpool iostat -Hplvy -y 1 1 | awk -F '\t' -v host="$host" -v pools="$pools" '
+          function esc(str, out) {
+            out = str
+            gsub(/\\/, "\\\\", out)
+            gsub(/"/, "\\\"", out)
+            return out
+          }
+
+          function emit(metric, pool, vdev, value) {
+            if (value == "" || value == "-") {
+              return
+            }
+
+            printf "%s{host=\"%s\",pool=\"%s\",vdev=\"%s\"} %s\n",
+              metric,
+              esc(host),
+              esc(pool),
+              esc(vdev),
+              value
+          }
+
+          BEGIN {
+            split(pools, pool_names, ",")
+            for (idx in pool_names) {
+              if (pool_names[idx] != "") {
+                known_pools[pool_names[idx]] = 1
+              }
+            }
+          }
+
+          NF == 0 {
+            next
+          }
+
+          {
+            row_name = $1
+
+            if (row_name in known_pools) {
+              current_pool = row_name
+              current_vdev = "_pool"
+            } else if (current_pool == "") {
+              next
+            } else {
+              current_vdev = row_name
+            }
+
+            emit("zpool_iostat_total_wait_read_ns", current_pool, current_vdev, $8)
+            emit("zpool_iostat_total_wait_write_ns", current_pool, current_vdev, $9)
+            emit("zpool_iostat_disk_wait_read_ns", current_pool, current_vdev, $10)
+            emit("zpool_iostat_disk_wait_write_ns", current_pool, current_vdev, $11)
+            emit("zpool_iostat_syncq_wait_read_ns", current_pool, current_vdev, $12)
+            emit("zpool_iostat_syncq_wait_write_ns", current_pool, current_vdev, $13)
+            emit("zpool_iostat_asyncq_wait_read_ns", current_pool, current_vdev, $14)
+            emit("zpool_iostat_asyncq_wait_write_ns", current_pool, current_vdev, $15)
+          }
+        ' >>"$tmp_file"
+
+        mv "$tmp_file" "$out_dir/zpool.prom"
+        trap - EXIT
+  '';
+in
+{
+  networking.firewall.interfaces.${monitoringInterface}.allowedTCPPorts = [
+    9100
+    9134
+    9256
+    9257
+    9633
+  ];
+
+  services.prometheus.exporters = {
+    node = {
+      enable = true;
+      enabledCollectors = [
+        "pressure"
+        "processes"
+        "systemd"
+      ];
+      extraFlags = [ "--collector.textfile.directory=${nodeTextfileDir}" ];
+    };
+
+    process = {
+      enable = true;
+      user = "root";
+      group = "root";
+      settings.process_names = mkProcessMatchers false;
+      extraFlags = [
+        "-gather-smaps=false"
+        "-remove-empty-groups=true"
+        "-threads=false"
+      ];
+    };
+
+    smartctl.enable = true;
+    zfs.enable = true;
+  };
+
+  programs.atop = {
+    enable = true;
+    atopService.enable = true;
+    atopRotateTimer.enable = true;
+    atopacctService.enable = true;
+    settings.interval = 30;
+  };
+
+  systemd = {
+    services = {
+      prometheus-process-pid-exporter = {
+        description = "Prometheus process exporter with per-PID naming";
+        wantedBy = [ "multi-user.target" ];
+        after = [ "network.target" ];
+        serviceConfig = {
+          ExecStart = ''
+            ${pkgs.prometheus-process-exporter}/bin/process-exporter \
+              --web.listen-address 0.0.0.0:9257 \
+              --config.path ${perPidConfig} \
+              -children=false \
+              -gather-smaps=false \
+              -remove-empty-groups=true \
+              -threads=false
+          '';
+          User = "root";
+          Group = "root";
+          Restart = "always";
+          WorkingDirectory = "/tmp";
+          CapabilityBoundingSet = [ "" ];
+          DeviceAllow = [ "" ];
+          LockPersonality = true;
+          MemoryDenyWriteExecute = true;
+          NoNewPrivileges = true;
+          PrivateDevices = true;
+          PrivateTmp = true;
+          ProtectClock = true;
+          ProtectControlGroups = true;
+          ProtectHome = true;
+          ProtectHostname = true;
+          ProtectKernelLogs = true;
+          ProtectKernelModules = true;
+          ProtectKernelTunables = true;
+          ProtectSystem = "strict";
+          RemoveIPC = true;
+          RestrictAddressFamilies = [
+            "AF_INET"
+            "AF_INET6"
+          ];
+          RestrictNamespaces = true;
+          RestrictRealtime = true;
+          RestrictSUIDSGID = true;
+          SystemCallArchitectures = "native";
+          UMask = "0077";
+        };
+      };
+
+      zpool-latency-exporter = {
+        description = "Exports ZFS latency metrics for node_exporter textfile collection";
+        after = [ "zfs-import.target" ];
+        requires = [ "zfs-import.target" ];
+        path = [
+          config.boot.zfs.package
+          pkgs.coreutils
+          pkgs.gawk
+        ];
+        serviceConfig = {
+          Type = "oneshot";
+          ExecStart = zpoolLatencyScript;
+        };
+      };
+    };
+
+    timers.zpool-latency-exporter = {
+      wantedBy = [ "timers.target" ];
+      timerConfig = {
+        OnBootSec = "2m";
+        OnUnitActiveSec = "60s";
+        Unit = "zpool-latency-exporter.service";
+      };
+    };
+
+    tmpfiles.rules = [ "d ${nodeTextfileDir} 0755 root root - -" ];
+  };
+}

common/optional/syncthing_base.nix

@@ -12,7 +12,7 @@
       brain.id = "SSCGIPI-IV3VYKB-TRNIJE3-COV4T2H-CDBER7F-I2CGHYA-NWOEUDU-3T5QAAN"; # cspell:disable-line
       ipad.id = "KI76T3X-SFUGV2L-VSNYTKR-TSIUV5L-SHWD3HE-GQRGRCN-GY4UFMD-CW6Z6AX"; # cspell:disable-line
       jeeves.id = "ICRHXZW-ECYJCUZ-I4CZ64R-3XRK7CG-LL2HAAK-FGOHD22-BQA4AI6-5OAL6AG"; # cspell:disable-line
-      phone.id = "TBRULKD-7DZPGGZ-F6LLB7J-MSO54AY-7KLPBIN-QOFK6PX-W2HBEWI-PHM2CQI"; # cspell:disable-line
+      phone.id = "JPVQKQW-CFXOJXT-Q5G5F3H-QIDHDRE-GKHPTQB-GXZUQSP-U7FR7F7-INP3AAH"; # cspell:disable-line
       rhapsody-in-green.id = "ASL3KC4-3XEN6PA-7BQBRKE-A7JXLI6-DJT43BY-Q4WPOER-7UALUAZ-VTPQ6Q4"; # cspell:disable-line
     };
   };

common/optional/update.nix

@@ -4,7 +4,7 @@
     flags = [ "--accept-flake-config" ];
     randomizedDelaySec = "1h";
     persistent = true;
-    flake = "github:RichieCahill/dotfiles";
+    flake = "git+https://gitea.tmmworkshop.com/richie/dotfiles?ref=main";
     allowReboot = true;
     dates = "Sat *-*-* 06:00:00";
   };

docs/zfs_failed_root_import_recovry.md

@@ -0,0 +1,76 @@
+# ZFS failed root import recovery
+
+## Fast path
+
+If the machine fails to boot because ZFS refuses to import `root_pool`:
+
+### GRUB
+
+1. At the bootloader menu, select the normal NixOS entry.
+2. Press `e`.
+3. Find the line that starts with `linux`.
+4. Append this to the end of that line:
+
+```text
+zfs_force=1
+```
+
+5. Boot once with `Ctrl+x` or `F10`.
+
+### systemd-boot
+
+1. At the bootloader menu, highlight the normal NixOS entry.
+2. Press `e`.
+3. Append this to the end of the options line:
+
+```text
+zfs_force=1
+```
+
+4. Press `Enter` to boot once.
+
+## After boot
+
+Run:
+
+```bash
+sudo zpool status
+sudo zpool import
+journalctl -b | rg "ZFS|zfs|import|root_pool"
+```
+
+## Expected result
+
+`sudo zpool status` should show `root_pool` as `ONLINE`.
+
+## Reboot test
+
+Run:
+
+```bash
+sudo reboot
+```
+
+Do not add `zfs_force=1` the second time.
+
+## If it still fails
+
+Boot once more with:
+
+```text
+zfs_force=1
+```
+
+Then run:
+
+```bash
+sudo zpool status -v
+sudo zpool history | tail -n 50
+journalctl -b | rg "ZFS|zfs|import|root_pool"
+```
+
+## Notes
+
+- Root pool name is `root_pool`.
+- This is a one-time recovery path after disk moves, controller changes, dirty exports, or interrupted imports.
+- Some hosts also need the LUKS unlock USB key inserted before boot.

flake.lock

@@ -8,11 +8,11 @@
       },
       "locked": {
         "dir": "pkgs/firefox-addons",
-        "lastModified": 1773979456,
+        "lastModified": 1780733803,
-        "narHash": "sha256-9kBMJ5IvxqNlkkj/swmE8uK1Sc7TL/LIRUI958m7uBM=",
+        "narHash": "sha256-QBJPq12P1DAXFGezoEJaSO/xPUrPlnaI3ddSaMG2JpM=",
         "owner": "rycee",
         "repo": "nur-expressions",
-        "rev": "81e28f47ac18d9e89513929c77e711e657b64851",
+        "rev": "c80b0aa94392c5f3612ac797108f6d952752036d",
         "type": "gitlab"
       },
       "original": {
@@ -29,11 +29,11 @@
         ]
       },
       "locked": {
-        "lastModified": 1774007980,
+        "lastModified": 1780679734,
-        "narHash": "sha256-FOnZjElEI8pqqCvB6K/1JRHTE8o4rer8driivTpq2uo=",
+        "narHash": "sha256-KmRNvpNOb7QEORa06bVgjW9kITcx0VhsI7w0vhmZyD8=",
         "owner": "nix-community",
         "repo": "home-manager",
-        "rev": "9670de2921812bc4e0452f6e3efd8c859696c183",
+        "rev": "b2b7db486e06e098711dc291bb25db82850e1d16",
         "type": "github"
       },
       "original": {
@@ -43,12 +43,15 @@
       }
     },
     "nixos-hardware": {
+      "inputs": {
+        "nixpkgs": "nixpkgs"
+      },
       "locked": {
-        "lastModified": 1774018263,
+        "lastModified": 1780310866,
-        "narHash": "sha256-HHYEwK1A22aSaxv2ibhMMkKvrDGKGlA/qObG4smrSqc=",
+        "narHash": "sha256-fPBRVf6A5xlACYcOI59shGrjURuvwu0lRsDoSCEXt/I=",
         "owner": "nixos",
         "repo": "nixos-hardware",
-        "rev": "2d4b4717b2534fad5c715968c1cece04a172b365",
+        "rev": "4ed851c979641e28597a05086332d75cdc9e395f",
         "type": "github"
       },
       "original": {
@@ -60,27 +63,24 @@
     },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1773821835,
+        "lastModified": 1767892417,
-        "narHash": "sha256-TJ3lSQtW0E2JrznGVm8hOQGVpXjJyXY2guAxku2O9A4=",
+        "narHash": "sha256-8bW3q88CEg2u4hSP66Vf4lpbLonHz7hqDNBMcCY7E9U=",
-        "owner": "nixos",
+        "rev": "3497aa5c9457a9d88d71fa93a4a8368816fbeeba",
-        "repo": "nixpkgs",
+        "type": "tarball",
-        "rev": "b40629efe5d6ec48dd1efba650c797ddbd39ace0",
+        "url": "https://releases.nixos.org/nixos/unstable/nixos-26.05pre924538.3497aa5c9457/nixexprs.tar.xz"
-        "type": "github"
       },
       "original": {
-        "owner": "nixos",
+        "type": "tarball",
-        "ref": "nixos-unstable",
+        "url": "https://channels.nixos.org/nixos-unstable/nixexprs.tar.xz"
-        "repo": "nixpkgs",
-        "type": "github"
       }
     },
     "nixpkgs-master": {
       "locked": {
-        "lastModified": 1774051532,
+        "lastModified": 1780798858,
-        "narHash": "sha256-d3CGMweyYIcPuTj5BKq+1Lx4zwlgL31nVtN647tOZKo=",
+        "narHash": "sha256-4KLc5ZMjfMQosXA2JasUgZTk3i+c/i1zMH4custtmI0=",
         "owner": "nixos",
         "repo": "nixpkgs",
-        "rev": "8620c0b5cc8fbe76502442181be1d0514bc3a1b7",
+        "rev": "92840095e65b9970125843175f4be974b71a92ad",
         "type": "github"
       },
       "original": {
@@ -106,12 +106,28 @@
         "type": "github"
       }
     },
+    "nixpkgs_2": {
+      "locked": {
+        "lastModified": 1780243769,
+        "narHash": "sha256-x5UQuRsH3MqI0U9afaXSNqzTPSeZlRLvFAav2Ux1pNw=",
+        "owner": "nixos",
+        "repo": "nixpkgs",
+        "rev": "331800de5053fcebacf6813adb5db9c9dca22a0c",
+        "type": "github"
+      },
+      "original": {
+        "owner": "nixos",
+        "ref": "nixos-unstable",
+        "repo": "nixpkgs",
+        "type": "github"
+      }
+    },
     "root": {
       "inputs": {
         "firefox-addons": "firefox-addons",
         "home-manager": "home-manager",
         "nixos-hardware": "nixos-hardware",
-        "nixpkgs": "nixpkgs",
+        "nixpkgs": "nixpkgs_2",
         "nixpkgs-master": "nixpkgs-master",
         "nixpkgs-stable": "nixpkgs-stable",
         "sops-nix": "sops-nix",
@@ -125,11 +141,11 @@
         ]
       },
       "locked": {
-        "lastModified": 1773889674,
+        "lastModified": 1780547341,
-        "narHash": "sha256-+ycaiVAk3MEshJTg35cBTUa0MizGiS+bgpYw/f8ohkg=",
+        "narHash": "sha256-Gq8KNx5A7hBB3uGJaj6eQfLDIz5YdLu92gqBcvHvoUo=",
         "owner": "Mic92",
         "repo": "sops-nix",
-        "rev": "29b6519f3e0780452bca0ac0be4584f04ac16cc5",
+        "rev": "9ed65852b6257fbeae4355bc24ecfea307ca759a",
         "type": "github"
       },
       "original": {

overlays/default.nix

@@ -24,7 +24,6 @@
         fastapi
         fastapi-cli
         httpx
-        huggingface-hub
         mypy
         orjson
         polars

pyproject.toml

@@ -12,7 +12,6 @@ dependencies = [
     "alembic",
     "apprise",
     "apscheduler",
-    "huggingface-hub",
     "httpx",
     "python-multipart",
     "polars",
@@ -27,11 +26,7 @@ dependencies = [
 [project.scripts]
 database = "python.database_cli:app"
 van-inventory = "python.van_inventory.main:serve"
-prompt-bench = "python.prompt_bench.main:cli"
+whisper-transcribe = "python.tools.whisper.transcribe:main"
-prompt-bench-download = "python.prompt_bench.downloader:cli"
-finetune = "python.prompt_bench.finetune:cli"
-finetune-container = "python.prompt_bench.finetune_container:cli"
-build-finetune-dataset = "python.prompt_bench.build_finetune_dataset:cli"
 
 [dependency-groups]
 dev = [
@@ -56,6 +51,7 @@ lint.ignore = [
     "COM812", # (TEMP) conflicts when used with the formatter
     "ISC001", # (TEMP) conflicts when used with the formatter
     "S603",   # (PERM) This is known to cause a false positive
+    "S607",   # (PERM) This is becoming a consistent annoyance
 ]
 
 [tool.ruff.lint.per-file-ignores]
@@ -84,14 +80,7 @@ lint.ignore = [
 "python/congress_tracker/**" = [
     "TC003", # (perm) this creates issues because sqlalchemy uses these at runtime
 ]
-"python/eval_warnings/**" = [
+
-    "S607", # (perm) gh and git are expected on PATH in the runner environment
-]
-"python/prompt_bench/**" = [
-    "FBT002",  # (perm) typer requires boolean defaults for --flag/--no-flag options
-    "PLR0913", # (perm) typer CLIs naturally have many parameters
-    "S607",    # (perm) docker and nvidia-smi are expected on PATH
-]
 "python/alembic/**" = [
     "INP001", # (perm) this creates LSP issues for alembic
 ]

python/gitea.py

@@ -0,0 +1,347 @@
+"""Small Gitea API client for repository automation."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Self
+from urllib.parse import quote
+
+import httpx
+
+DEFAULT_PAGE_SIZE = 100
+EXPECTED_NO_CONTENT = 204
+EXPECTED_CREATED = 201
+EXPECTED_OK = 200
+
+
+@dataclass(frozen=True)
+class CreatedIssue:
+    """Issue data returned by Gitea."""
+
+    number: int | None
+    html_url: str | None
+    title: str
+
+
+@dataclass(frozen=True)
+class PullRequest:
+    """Pull request data returned by Gitea."""
+
+    number: int
+    title: str
+    html_url: str | None
+    labels: tuple[str, ...]
+    head_branch: str | None
+    base_branch: str | None
+
+
+@dataclass(frozen=True)
+class WorkflowJob:
+    """Workflow job data returned by Gitea Actions."""
+
+    id: int
+    name: str
+    run_id: int | None
+    status: str | None
+    conclusion: str | None
+
+
+class GiteaError(RuntimeError):
+    """Raised when Gitea rejects an API request."""
+
+
+def split_repo_name(repo: str) -> tuple[str, str]:
+    """Split an owner/repo string into its parts."""
+    owner, separator, repo_name = repo.partition("/")
+    if not separator or not owner or not repo_name:
+        msg = f"Invalid repository name: {repo}"
+        raise ValueError(msg)
+    return owner, repo_name
+
+
+class GiteaClient:
+    """HTTP client for the subset of Gitea APIs used in this repository."""
+
+    def __init__(
+        self,
+        *,
+        base_url: str,
+        token: str,
+        timeout: int = 30,
+        transport: httpx.BaseTransport | None = None,
+    ) -> None:
+        """Initialize the Gitea client."""
+        self._client = httpx.Client(
+            base_url=base_url.rstrip("/"),
+            timeout=timeout,
+            headers={"Authorization": f"token {token}"},
+            transport=transport,
+        )
+
+    def create_issue(
+        self,
+        *,
+        owner: str,
+        repo: str,
+        title: str,
+        body: str,
+        labels: list[int] | None = None,
+    ) -> CreatedIssue:
+        """Create a Gitea issue."""
+        payload: dict[str, object] = {"title": title, "body": body, "labels": labels or []}
+        response = self._request(
+            "POST",
+            f"/api/v1/repos/{owner}/{repo}/issues",
+            expected_statuses={EXPECTED_CREATED},
+            json=payload,
+        )
+        data = response.json()
+        return CreatedIssue(
+            number=_optional_int(data.get("number")),
+            html_url=_optional_str(data.get("html_url")),
+            title=str(data.get("title", title)),
+        )
+
+    def resolve_label_ids(self, *, owner: str, repo: str, labels: list[str]) -> list[int]:
+        """Resolve label names to Gitea label IDs."""
+        if not labels:
+            return []
+
+        available_labels: dict[str, int] = {}
+        page = 1
+        while True:
+            response = self._request(
+                "GET",
+                f"/api/v1/repos/{owner}/{repo}/labels",
+                params={"page": page, "limit": DEFAULT_PAGE_SIZE},
+            )
+            batch = response.json()
+            if not batch:
+                break
+            for label in batch:
+                label_name = str(label.get("name", ""))
+                label_id = _optional_int(label.get("id"))
+                if label_name and label_id is not None:
+                    available_labels[label_name] = label_id
+            if len(batch) < DEFAULT_PAGE_SIZE:
+                break
+            page += 1
+
+        missing = [label for label in labels if label not in available_labels]
+        if missing:
+            missing_names = ", ".join(sorted(missing))
+            msg = f"Missing Gitea labels: {missing_names}"
+            raise GiteaError(msg)
+
+        return [available_labels[label] for label in labels]
+
+    def list_open_pull_requests(
+        self,
+        *,
+        owner: str,
+        repo: str,
+        labels: list[str] | None = None,
+        head: str | None = None,
+    ) -> list[PullRequest]:
+        """List open pull requests for a repository."""
+        expected_labels = set(labels or [])
+        pull_requests: list[PullRequest] = []
+        page = 1
+        while True:
+            response = self._request(
+                "GET",
+                f"/api/v1/repos/{owner}/{repo}/pulls",
+                params={"state": "open", "page": page, "limit": DEFAULT_PAGE_SIZE},
+            )
+            batch = response.json()
+            if not batch:
+                break
+
+            for item in batch:
+                pull_request = _pull_request_from_api(item)
+                if head and pull_request.head_branch != head:
+                    continue
+                if expected_labels and not expected_labels.issubset(set(pull_request.labels)):
+                    continue
+                pull_requests.append(pull_request)
+
+            if len(batch) < DEFAULT_PAGE_SIZE:
+                break
+            page += 1
+
+        return pull_requests
+
+    def create_pull_request(
+        self,
+        *,
+        owner: str,
+        repo: str,
+        title: str,
+        body: str,
+        head: str,
+        base: str,
+        labels: list[str] | None = None,
+    ) -> PullRequest:
+        """Create a pull request."""
+        payload: dict[str, object] = {
+            "title": title,
+            "body": body,
+            "head": head,
+            "base": base,
+        }
+        if labels:
+            payload["labels"] = self.resolve_label_ids(owner=owner, repo=repo, labels=labels)
+
+        response = self._request(
+            "POST",
+            f"/api/v1/repos/{owner}/{repo}/pulls",
+            expected_statuses={EXPECTED_CREATED},
+            json=payload,
+        )
+        return _pull_request_from_api(response.json())
+
+    def merge_pull_request(
+        self,
+        *,
+        owner: str,
+        repo: str,
+        number: int,
+        merge_method: str = "rebase",
+        head_commit_id: str | None = None,
+        delete_branch_after_merge: bool = False,
+    ) -> None:
+        """Merge a pull request."""
+        payload: dict[str, object] = {
+            "Do": merge_method,
+            "delete_branch_after_merge": delete_branch_after_merge,
+        }
+        if head_commit_id:
+            payload["head_commit_id"] = head_commit_id
+
+        self._request(
+            "POST",
+            f"/api/v1/repos/{owner}/{repo}/pulls/{number}/merge",
+            json=payload,
+        )
+
+    def dispatch_workflow(self, *, owner: str, repo: str, workflow_id: str, ref: str) -> None:
+        """Trigger a workflow_dispatch run."""
+        workflow_path = quote(workflow_id, safe="")
+        self._request(
+            "POST",
+            f"/api/v1/repos/{owner}/{repo}/actions/workflows/{workflow_path}/dispatches",
+            expected_statuses={EXPECTED_OK, EXPECTED_NO_CONTENT},
+            json={"ref": ref},
+        )
+
+    def list_run_jobs(self, *, owner: str, repo: str, run_id: str | int) -> list[WorkflowJob]:
+        """List workflow jobs for a specific run."""
+        jobs: list[WorkflowJob] = []
+        page = 1
+        while True:
+            response = self._request(
+                "GET",
+                f"/api/v1/repos/{owner}/{repo}/actions/jobs",
+                params={"page": page, "limit": DEFAULT_PAGE_SIZE},
+            )
+            payload = response.json()
+            batch = payload.get("jobs", [])
+            if not batch:
+                break
+
+            for item in batch:
+                if str(item.get("run_id")) != str(run_id):
+                    continue
+                jobs.append(_workflow_job_from_api(item))
+
+            if len(batch) < DEFAULT_PAGE_SIZE:
+                break
+            page += 1
+
+        return jobs
+
+    def download_job_logs(self, *, owner: str, repo: str, job_id: int) -> str:
+        """Download logs for a workflow job."""
+        response = self._request(
+            "GET",
+            f"/api/v1/repos/{owner}/{repo}/actions/jobs/{job_id}/logs",
+        )
+        return response.text
+
+    def close(self) -> None:
+        """Close the underlying HTTP client."""
+        self._client.close()
+
+    def __enter__(self) -> Self:
+        """Enter the context manager."""
+        return self
+
+    def __exit__(self, *args: object) -> None:
+        """Close the HTTP client."""
+        self.close()
+
+    def _request(
+        self,
+        method: str,
+        path: str,
+        *,
+        expected_statuses: set[int] | None = None,
+        **kwargs: object,
+    ) -> httpx.Response:
+        """Send an HTTP request and validate the response status."""
+        response = self._client.request(method, path, **kwargs)
+        statuses = expected_statuses or {EXPECTED_OK}
+        if response.status_code not in statuses:
+            msg = f"Gitea request failed ({response.status_code}): {response.text}"
+            raise GiteaError(msg)
+        return response
+
+
+def _pull_request_from_api(data: dict[str, object]) -> PullRequest:
+    """Convert Gitea API pull-request data into a dataclass."""
+    number = _optional_int(data.get("number")) or _optional_int(data.get("index"))
+    if number is None:
+        msg = "Gitea pull request payload is missing a number"
+        raise GiteaError(msg)
+
+    labels = tuple(str(label.get("name", "")) for label in data.get("labels", []))
+    head = data.get("head", {})
+    base = data.get("base", {})
+    return PullRequest(
+        number=number,
+        title=str(data.get("title", "")),
+        html_url=_optional_str(data.get("html_url")),
+        labels=tuple(label for label in labels if label),
+        head_branch=_optional_str(head.get("ref")) or _optional_str(data.get("head_branch")),
+        base_branch=_optional_str(base.get("ref")) or _optional_str(data.get("base_branch")),
+    )
+
+
+def _workflow_job_from_api(data: dict[str, object]) -> WorkflowJob:
+    """Convert Gitea API workflow-job data into a dataclass."""
+    job_id = _optional_int(data.get("id"))
+    if job_id is None:
+        msg = "Gitea workflow job payload is missing an ID"
+        raise GiteaError(msg)
+
+    return WorkflowJob(
+        id=job_id,
+        name=str(data.get("name", "")),
+        run_id=_optional_int(data.get("run_id")),
+        status=_optional_str(data.get("status")),
+        conclusion=_optional_str(data.get("conclusion")),
+    )
+
+
+def _optional_int(value: object) -> int | None:
+    """Convert an API value to an integer when present."""
+    if value is None:
+        return None
+    return int(value)
+
+
+def _optional_str(value: object) -> str | None:
+    """Convert an API value to a string when present."""
+    if value is None:
+        return None
+    return str(value)

python/gitea_flake_lock.py

@@ -0,0 +1,148 @@
+"""Automation helpers for flake.lock pull requests on Gitea."""
+
+from __future__ import annotations
+
+import subprocess
+from os import getenv
+from typing import Annotated
+
+import typer
+
+from python.gitea import GiteaClient, PullRequest, split_repo_name
+
+DEFAULT_BASE_BRANCH = "main"
+DEFAULT_BRANCH = "automation/update-flake-lock"
+DEFAULT_GITEA_URL = "https://gitea.tmmworkshop.com"
+PR_LABELS = ["dependencies", "automated", "flake_lock_update"]
+PR_CHECK_WORKFLOWS = ["build_systems.yml", "treefmt.yml", "pytest.yml"]
+PR_TITLE = "Update flake.lock"
+PR_BODY = "Automated flake.lock update."
+
+app = typer.Typer(add_completion=False)
+
+
+def run_cmd(cmd: list[str], *, check: bool = True) -> subprocess.CompletedProcess[str]:
+    """Run a subprocess command."""
+    return subprocess.run(cmd, capture_output=True, text=True, check=check)
+
+
+def ensure_flake_lock_pull_request(
+    client: GiteaClient,
+    *,
+    owner: str,
+    repo: str,
+    branch: str,
+    base: str,
+) -> PullRequest:
+    """Return an existing flake.lock PR for the branch or create one."""
+    pull_requests = client.list_open_pull_requests(owner=owner, repo=repo, head=branch)
+    if pull_requests:
+        return pull_requests[0]
+
+    return client.create_pull_request(
+        owner=owner,
+        repo=repo,
+        title=PR_TITLE,
+        body=PR_BODY,
+        head=branch,
+        base=base,
+        labels=PR_LABELS,
+    )
+
+
+def find_flake_lock_pull_request(client: GiteaClient, *, owner: str, repo: str) -> PullRequest | None:
+    """Find the first open flake.lock pull request."""
+    pull_requests = client.list_open_pull_requests(owner=owner, repo=repo, labels=["flake_lock_update"])
+    if not pull_requests:
+        return None
+    return pull_requests[0]
+
+
+def dispatch_pull_request_checks(client: GiteaClient, *, owner: str, repo: str, branch: str) -> None:
+    """Dispatch the workflows that normally run for pull requests."""
+    for workflow in PR_CHECK_WORKFLOWS:
+        client.dispatch_workflow(owner=owner, repo=repo, workflow_id=workflow, ref=branch)
+
+
+def has_worktree_changes() -> bool:
+    """Return whether `flake.lock` has worktree changes."""
+    result = run_cmd(["git", "diff", "--quiet", "--", "flake.lock"], check=False)
+    return result.returncode != 0
+
+
+def commit_flake_lock_update(*, branch: str) -> None:
+    """Commit the updated lock file to the automation branch."""
+    run_cmd(["git", "config", "user.name", "gitea-actions[bot]"])
+    run_cmd(["git", "config", "user.email", "gitea-actions@tmmworkshop.com"])
+    run_cmd(["git", "checkout", "-B", branch])
+    run_cmd(["git", "add", "flake.lock"])
+    run_cmd(["git", "commit", "-m", "chore: update flake.lock"])
+
+
+def push_branch(*, branch: str) -> None:
+    """Push the automation branch to origin."""
+    run_cmd(["git", "push", "origin", f"HEAD:{branch}", "--force"])
+
+
+def _required_gitea_token() -> str:
+    """Read the required Gitea token from the environment."""
+    token = getenv("GITEA_TOKEN")
+    if token:
+        return token
+
+    msg = "GITEA_TOKEN environment variable is required"
+    raise RuntimeError(msg)
+
+
+@app.command()
+def update(
+    repo: Annotated[str, typer.Option("--repo", help="Gitea repository in owner/repo form")],
+    base: Annotated[str, typer.Option("--base", help="Base branch")] = DEFAULT_BASE_BRANCH,
+    branch: Annotated[str, typer.Option("--branch", help="Automation branch")] = DEFAULT_BRANCH,
+) -> None:
+    """Commit flake.lock changes and ensure a pull request exists."""
+    if not has_worktree_changes():
+        typer.echo("No flake.lock changes detected")
+        return
+
+    commit_flake_lock_update(branch=branch)
+    push_branch(branch=branch)
+
+    owner, repo_name = split_repo_name(repo)
+    with GiteaClient(
+        base_url=getenv("GITEA_URL", DEFAULT_GITEA_URL),
+        token=_required_gitea_token(),
+    ) as client:
+        pull_request = ensure_flake_lock_pull_request(
+            client,
+            owner=owner,
+            repo=repo_name,
+            branch=branch,
+            base=base,
+        )
+        # We can remove this if Gitea fixes the following issue:
+        # https://github.com/go-gitea/gitea/issues/33963
+        dispatch_pull_request_checks(client, owner=owner, repo=repo_name, branch=branch)
+    typer.echo(pull_request.html_url or f"Pull request #{pull_request.number}")
+
+
+@app.command()
+def merge(
+    repo: Annotated[str, typer.Option("--repo", help="Gitea repository in owner/repo form")],
+) -> None:
+    """Merge the first open flake.lock pull request."""
+    owner, repo_name = split_repo_name(repo)
+    with GiteaClient(
+        base_url=getenv("GITEA_URL", DEFAULT_GITEA_URL),
+        token=_required_gitea_token(),
+    ) as client:
+        pull_request = find_flake_lock_pull_request(client, owner=owner, repo=repo_name)
+        if not pull_request:
+            typer.echo("No open PR found with label flake_lock_update")
+            return
+        client.merge_pull_request(owner=owner, repo=repo_name, number=pull_request.number, merge_method="rebase")
+    typer.echo(f"Merged PR #{pull_request.number}")
+
+
+if __name__ == "__main__":
+    app()

python/prompt_bench/Dockerfile.finetune

@@ -1,25 +0,0 @@
-# Unsloth fine-tuning container for Qwen 3.5 4B on RTX 3090.
-#
-# Build:
-#   docker build -f python/prompt_bench/Dockerfile.finetune -t bill-finetune .
-#
-# Run:
-#   docker run --rm --device=nvidia.com/gpu=all --ipc=host \
-#     -v $(pwd)/output:/workspace/output \
-#     -v $(pwd)/output/finetune_dataset.jsonl:/workspace/dataset.jsonl:ro \
-#     -v /zfs/models/hf:/models \
-#     bill-finetune \
-#     --dataset /workspace/dataset.jsonl \
-#     --output-dir /workspace/output/qwen-bill-summarizer
-
-FROM ghcr.io/unslothai/unsloth:latest
-
-RUN pip install --no-cache-dir typer
-
-WORKDIR /workspace
-COPY python/prompt_bench/finetune.py python/prompt_bench/finetune.py
-COPY python/prompt_bench/summarization_prompts.py python/prompt_bench/summarization_prompts.py
-COPY python/prompt_bench/__init__.py python/prompt_bench/__init__.py
-COPY python/__init__.py python/__init__.py
-
-ENTRYPOINT ["python", "-m", "python.prompt_bench.finetune"]

python/prompt_bench/__init__.py

@@ -1 +0,0 @@
-"""Prompt benchmarking system for evaluating LLMs via vLLM."""

python/prompt_bench/batch_bill_summarizer.py

@@ -1,233 +0,0 @@
-"""Submit an OpenAI Batch API bill-summarization job over compressed text.
-
-Reads the first N bills from a CSV with a `text_content` column, compresses
-each via `bill_token_compression.compress_bill_text`, builds a JSONL file of
-summarization requests, and submits it as an asynchronous Batch API job
-against `/v1/chat/completions`. Also writes a CSV of per-bill pre/post-
-compression token counts.
-"""
-
-from __future__ import annotations
-
-import csv
-import json
-import logging
-import re
-import sys
-from os import getenv
-from pathlib import Path
-from typing import Annotated
-
-import httpx
-import typer
-from tiktoken import Encoding, get_encoding
-
-from python.prompt_bench.bill_token_compression import compress_bill_text
-from python.prompt_bench.summarization_prompts import SUMMARIZATION_SYSTEM_PROMPT, SUMMARIZATION_USER_TEMPLATE
-
-logger = logging.getLogger(__name__)
-
-OPENAI_API_BASE = "https://api.openai.com/v1"
-
-
-def load_bills(csv_path: Path, count: int = 0) -> list[tuple[str, str]]:
-    """Return (bill_id, text_content) tuples with non-empty text.
-
-    If `count` is 0 or negative, all rows are returned.
-    """
-    csv.field_size_limit(sys.maxsize)
-    bills: list[tuple[str, str]] = []
-    with csv_path.open(newline="", encoding="utf-8") as handle:
-        reader = csv.DictReader(handle)
-        for row in reader:
-            text_content = (row.get("text_content") or "").strip()
-            if not text_content:
-                continue
-            bill_id = row.get("bill_id") or row.get("id") or f"row-{len(bills)}"
-            version_code = row.get("version_code") or ""
-            unique_id = f"{bill_id}-{version_code}" if version_code else bill_id
-            bills.append((unique_id, text_content))
-            if count > 0 and len(bills) >= count:
-                break
-    return bills
-
-
-def safe_filename(value: str) -> str:
-    """Make a string safe for use as a filename or batch custom_id."""
-    return re.sub(r"[^A-Za-z0-9._-]+", "_", value).strip("_") or "unnamed"
-
-
-def build_request(custom_id: str, model: str, bill_text: str) -> dict:
-    """Build one OpenAI batch request line."""
-    return {
-        "custom_id": custom_id,
-        "method": "POST",
-        "url": "/v1/chat/completions",
-        "body": {
-            "model": model,
-            "messages": [
-                {"role": "system", "content": SUMMARIZATION_SYSTEM_PROMPT},
-                {"role": "user", "content": SUMMARIZATION_USER_TEMPLATE.format(text_content=bill_text)},
-            ],
-        },
-    }
-
-
-def write_jsonl(path: Path, lines: list[dict]) -> None:
-    """Write a list of dicts as JSONL."""
-    with path.open("w", encoding="utf-8") as handle:
-        for line in lines:
-            handle.write(json.dumps(line, ensure_ascii=False))
-            handle.write("\n")
-
-
-def upload_file(client: httpx.Client, path: Path) -> str:
-    """Upload a JSONL file to the OpenAI Files API and return its file id."""
-    with path.open("rb") as handle:
-        response = client.post(
-            f"{OPENAI_API_BASE}/files",
-            files={"file": (path.name, handle, "application/jsonl")},
-            data={"purpose": "batch"},
-        )
-    response.raise_for_status()
-    return response.json()["id"]
-
-
-def prepare_requests(
-    bills: list[tuple[str, str]],
-    *,
-    model: str,
-    encoder: Encoding,
-) -> tuple[list[dict], list[dict]]:
-    """Build (request_lines, token_rows) from bills.
-
-    Each bill is compressed before being turned into a request line.
-    Each `token_rows` entry has chars + token counts for one bill so the caller
-    can write a per-bill CSV.
-    """
-    request_lines: list[dict] = []
-    token_rows: list[dict] = []
-    for bill_id, text_content in bills:
-        raw_token_count = len(encoder.encode(text_content))
-        compressed_text = compress_bill_text(text_content)
-        compressed_token_count = len(encoder.encode(compressed_text))
-        token_rows.append(
-            {
-                "bill_id": bill_id,
-                "raw_chars": len(text_content),
-                "compressed_chars": len(compressed_text),
-                "raw_tokens": raw_token_count,
-                "compressed_tokens": compressed_token_count,
-                "token_ratio": (compressed_token_count / raw_token_count) if raw_token_count else None,
-            },
-        )
-        safe_id = safe_filename(bill_id)
-        request_lines.append(build_request(safe_id, model, compressed_text))
-    return request_lines, token_rows
-
-
-def write_token_csv(path: Path, token_rows: list[dict]) -> tuple[int, int]:
-    """Write per-bill token counts to CSV. Returns (raw_total, compressed_total)."""
-    with path.open("w", newline="", encoding="utf-8") as handle:
-        writer = csv.DictWriter(
-            handle,
-            fieldnames=["bill_id", "raw_chars", "compressed_chars", "raw_tokens", "compressed_tokens", "token_ratio"],
-        )
-        writer.writeheader()
-        writer.writerows(token_rows)
-    raw_total = sum(row["raw_tokens"] for row in token_rows)
-    compressed_total = sum(row["compressed_tokens"] for row in token_rows)
-    return raw_total, compressed_total
-
-
-def create_batch(client: httpx.Client, input_file_id: str, description: str) -> dict:
-    """Create a batch job and return its full response payload."""
-    response = client.post(
-        f"{OPENAI_API_BASE}/batches",
-        json={
-            "input_file_id": input_file_id,
-            "endpoint": "/v1/chat/completions",
-            "completion_window": "24h",
-            "metadata": {"description": description},
-        },
-    )
-    response.raise_for_status()
-    return response.json()
-
-
-def main(
-    csv_path: Annotated[Path, typer.Option("--csv", help="Bills CSV path")] = Path("bills.csv"),
-    output_dir: Annotated[Path, typer.Option("--output-dir", help="Where to write JSONL + metadata")] = Path(
-        "output/openai_batch",
-    ),
-    model: Annotated[str, typer.Option(help="OpenAI model id")] = "gpt-5-mini",
-    count: Annotated[int, typer.Option(help="Max bills to process, 0 = all")] = 0,
-    log_level: Annotated[str, typer.Option(help="Log level")] = "INFO",
-) -> None:
-    """Submit an OpenAI Batch job of compressed bill summaries."""
-    logging.basicConfig(level=log_level, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
-
-    api_key = getenv("CLOSEDAI_TOKEN") or getenv("OPENAI_API_KEY")
-    if not api_key:
-        message = "Neither CLOSEDAI_TOKEN nor OPENAI_API_KEY is set"
-        raise typer.BadParameter(message)
-    if not csv_path.is_file():
-        message = f"CSV not found: {csv_path}"
-        raise typer.BadParameter(message)
-
-    output_dir.mkdir(parents=True, exist_ok=True)
-
-    logger.info("Loading %d bills from %s", count, csv_path)
-    bills = load_bills(csv_path, count)
-    if len(bills) < count:
-        logger.warning("Only %d bills available (requested %d)", len(bills), count)
-
-    encoder = get_encoding("o200k_base")
-    request_lines, token_rows = prepare_requests(bills, model=model, encoder=encoder)
-
-    token_csv_path = output_dir / "token_counts.csv"
-    raw_tokens_total, compressed_tokens_total = write_token_csv(token_csv_path, token_rows)
-    logger.info(
-        "Token counts: raw=%d compressed=%d ratio=%.3f -> %s",
-        raw_tokens_total,
-        compressed_tokens_total,
-        (compressed_tokens_total / raw_tokens_total) if raw_tokens_total else 0.0,
-        token_csv_path,
-    )
-
-    jsonl_path = output_dir / "requests.jsonl"
-    write_jsonl(jsonl_path, request_lines)
-    logger.info("Wrote %s (%d bills)", jsonl_path, len(request_lines))
-
-    headers = {"Authorization": f"Bearer {api_key}"}
-    with httpx.Client(headers=headers, timeout=httpx.Timeout(300.0)) as client:
-        logger.info("Uploading JSONL")
-        file_id = upload_file(client, jsonl_path)
-        logger.info("Uploaded: %s", file_id)
-
-        logger.info("Creating batch")
-        batch = create_batch(client, file_id, f"compressed bill summaries x{len(request_lines)} ({model})")
-        logger.info("Batch created: %s", batch["id"])
-
-    metadata = {
-        "model": model,
-        "count": len(bills),
-        "jsonl": str(jsonl_path),
-        "input_file_id": file_id,
-        "batch_id": batch["id"],
-        "raw_tokens_total": raw_tokens_total,
-        "compressed_tokens_total": compressed_tokens_total,
-        "batch": batch,
-    }
-    metadata_path = output_dir / "batch.json"
-    metadata_path.write_text(json.dumps(metadata, indent=2))
-    logger.info("Wrote metadata to %s", metadata_path)
-
-
-def cli() -> None:
-    """Typer entry point."""
-    typer.run(main)
-
-
-if __name__ == "__main__":
-    cli()

python/prompt_bench/bill_token_compression.py

@@ -1,162 +0,0 @@
-"""Lossless-ish text compression for Congressional bill text."""
-
-from __future__ import annotations
-
-import re
-
-STATES = (
-    "Alabama",
-    "Alaska",
-    "Arizona",
-    "Arkansas",
-    "California",
-    "Colorado",
-    "Connecticut",
-    "Delaware",
-    "Florida",
-    "Georgia",
-    "Hawaii",
-    "Idaho",
-    "Illinois",
-    "Indiana",
-    "Iowa",
-    "Kansas",
-    "Kentucky",
-    "Louisiana",
-    "Maine",
-    "Maryland",
-    "Massachusetts",
-    "Michigan",
-    "Minnesota",
-    "Mississippi",
-    "Missouri",
-    "Montana",
-    "Nebraska",
-    "Nevada",
-    "New Hampshire",
-    "New Jersey",
-    "New Mexico",
-    "New York",
-    "North Carolina",
-    "North Dakota",
-    "Ohio",
-    "Oklahoma",
-    "Oregon",
-    "Pennsylvania",
-    "Rhode Island",
-    "South Carolina",
-    "South Dakota",
-    "Tennessee",
-    "Texas",
-    "Utah",
-    "Vermont",
-    "Virginia",
-    "Washington",
-    "West Virginia",
-    "Wisconsin",
-    "Wyoming",
-    "Puerto Rico",
-    "Guam",
-    "American Samoa",
-    "District of Columbia",
-    "US Virgin Islands",
-)
-STATE_PATTERNS = [(re.compile(re.escape(state), re.IGNORECASE), state) for state in STATES]
-
-
-def normalize_state_names(text: str) -> str:
-    """Replace any casing of state names with title case."""
-    for pattern, replacement in STATE_PATTERNS:
-        text = pattern.sub(replacement, text)
-    return text
-
-
-def strip_number_commas(text: str) -> str:
-    """Remove commas from numeric thousands separators."""
-    return re.sub(r"(\d{1,3}(?:,\d{3})+)", lambda match: match.group().replace(",", ""), text)
-
-
-def strip_horizontal_rules(text: str) -> str:
-    """Remove ASCII horizontal-rule lines built from underscores, dashes, equals, or asterisks."""
-    return re.sub(r"^\s*[_\-=\*]{3,}\s*$", "", text, flags=re.MULTILINE)
-
-
-def collapse_double_dashes(text: str) -> str:
-    """Replace ``--`` em-dash stand-ins with a single space so they don't tokenize oddly."""
-    return text.replace("--", " ")
-
-
-def collapse_inline_whitespace(text: str) -> str:
-    """Collapse runs of horizontal whitespace (spaces, tabs) into a single space, leaving newlines intact."""
-    return re.sub(r"[^\S\n]+", " ", text)
-
-
-def collapse_blank_lines(text: str) -> str:
-    """Collapse three-or-more consecutive newlines down to a blank-line separator."""
-    return re.sub(r"\n{3,}", "\n\n", text)
-
-
-def trim_line_edges(text: str) -> str:
-    """Strip spaces immediately before and after newline characters on every line."""
-    text = re.sub(r" +\n", "\n", text)
-    return re.sub(r"\n +", "\n", text)
-
-
-def shorten_section_markers(text: str) -> str:
-    """Rewrite ``Sec. 12.`` style section headings as the more compact ``SEC 12``."""
-    return re.sub(r"(?i)sec\.\s*(\d+[a-zA-Z]?)\.", r"SEC \1", text)
-
-
-def unwrap_parens(text: str) -> str:
-    """Strip parentheses around short alphanumeric labels like ``(a)`` or ``(12)``."""
-    return re.sub(r"\(([a-zA-Z0-9]+)\)", r"\1", text)
-
-
-def strip_typeset_quotes(text: str) -> str:
-    """Remove the `` and '' typeset quote markers used in the GPO bill format."""
-    return text.replace("``", "").replace("''", "")
-
-
-def normalize_usc_acronym(text: str) -> str:
-    """Collapse ``U.S.C.`` to ``USC`` to save tokens on the common citation."""
-    return text.replace("U.S.C.", "USC")
-
-
-def normalize_us_acronym(text: str) -> str:
-    """Normalize the various ``U.S.``/``U. S.`` spellings to the bare ``US`` form."""
-    for acronym in ("U. S.", "u. s.", "U.S. ", "u.s. "):
-        text = text.replace(acronym, "US ")
-    return text
-
-
-def collapse_ellipses(text: str) -> str:
-    """Collapse runs of two-or-more periods (``...``, ``....``) down to a single period."""
-    return re.sub(r"\.{2,}", ".", text)
-
-
-COMPRESSION_STEPS = (
-    strip_horizontal_rules,
-    collapse_double_dashes,
-    collapse_inline_whitespace,
-    collapse_blank_lines,
-    trim_line_edges,
-    shorten_section_markers,
-    unwrap_parens,
-    strip_typeset_quotes,
-    normalize_usc_acronym,
-    normalize_us_acronym,
-    strip_number_commas,
-    collapse_ellipses,
-    normalize_state_names,
-)
-
-
-def compress_bill_text(text: str) -> str:
-    """Apply lossless-ish whitespace and boilerplate compression to bill text.
-
-    Runs every transform in :data:`COMPRESSION_STEPS` in order, then strips
-    leading/trailing whitespace from the final result.
-    """
-    for step in COMPRESSION_STEPS:
-        text = step(text)
-    return text.strip()

python/prompt_bench/compresion_test.py

@@ -1,236 +0,0 @@
-"""Run two interactive OpenAI chat-completion sweeps over bill text.
-
-Reads the first N bills from a CSV with a `text_content` column and sends two
-sweeps through `/v1/chat/completions` concurrently — one with the raw bill
-text, one with the compressed bill text. Each request's prompt is saved to
-disk alongside the OpenAI response id so the prompts and responses can be
-correlated later.
-"""
-
-from __future__ import annotations
-
-import csv
-import json
-import logging
-import re
-import sys
-import time
-from concurrent.futures import ThreadPoolExecutor, as_completed
-from os import getenv
-from pathlib import Path
-from typing import Annotated
-
-import httpx
-import typer
-
-from python.prompt_bench.bill_token_compression import compress_bill_text
-from python.prompt_bench.summarization_prompts import SUMMARIZATION_SYSTEM_PROMPT, SUMMARIZATION_USER_TEMPLATE
-
-logger = logging.getLogger(__name__)
-
-OPENAI_API_BASE = "https://api.openai.com/v1"
-DEFAULT_MODEL = "gpt-5.4-mini"
-DEFAULT_COUNT = 100
-SEED = 42
-
-
-def load_bills(csv_path: Path, count: int) -> list[tuple[str, str]]:
-    """Return up to `count` (bill_id, text_content) tuples with non-empty text."""
-    csv.field_size_limit(sys.maxsize)
-    bills: list[tuple[str, str]] = []
-    with csv_path.open(newline="", encoding="utf-8") as handle:
-        reader = csv.DictReader(handle)
-        for row in reader:
-            text_content = (row.get("text_content") or "").strip()
-            if not text_content:
-                continue
-            bill_id = row.get("bill_id") or row.get("id") or f"row-{len(bills)}"
-            version_code = row.get("version_code") or ""
-            unique_id = f"{bill_id}-{version_code}" if version_code else bill_id
-            bills.append((unique_id, text_content))
-            if len(bills) >= count:
-                break
-    return bills
-
-
-def build_messages(bill_text: str) -> list[dict]:
-    """Return the system + user message pair for a bill."""
-    return [
-        {"role": "system", "content": SUMMARIZATION_SYSTEM_PROMPT},
-        {"role": "user", "content": SUMMARIZATION_USER_TEMPLATE.format(text_content=bill_text)},
-    ]
-
-
-def safe_filename(value: str) -> str:
-    """Make a string safe for use as a filename."""
-    return re.sub(r"[^A-Za-z0-9._-]+", "_", value).strip("_") or "unnamed"
-
-
-def run_one_request(
-    client: httpx.Client,
-    *,
-    bill_id: str,
-    label: str,
-    bill_text: str,
-    model: str,
-    output_path: Path,
-) -> tuple[bool, float, str | None]:
-    """Send one chat-completion request and persist prompt + response.
-
-    Returns (success, elapsed_seconds, response_id).
-    """
-    messages = build_messages(bill_text)
-    payload = {
-        "model": model,
-        "messages": messages,
-        "seed": SEED,
-    }
-    start = time.monotonic()
-    record: dict = {
-        "bill_id": bill_id,
-        "label": label,
-        "model": model,
-        "seed": SEED,
-        "input_chars": len(bill_text),
-        "messages": messages,
-    }
-    try:
-        response = client.post(f"{OPENAI_API_BASE}/chat/completions", json=payload)
-        response.raise_for_status()
-        body = response.json()
-    except httpx.HTTPStatusError as error:
-        elapsed = time.monotonic() - start
-        record["error"] = {
-            "status_code": error.response.status_code,
-            "body": error.response.text,
-            "elapsed_seconds": elapsed,
-        }
-        output_path.write_text(json.dumps(record, ensure_ascii=False, indent=2))
-        logger.exception("HTTP error for %s/%s after %.2fs", label, bill_id, elapsed)
-        return False, elapsed, None
-    except Exception as error:
-        elapsed = time.monotonic() - start
-        record["error"] = {"message": str(error), "elapsed_seconds": elapsed}
-        output_path.write_text(json.dumps(record, ensure_ascii=False, indent=2))
-        logger.exception("Failed: %s/%s after %.2fs", label, bill_id, elapsed)
-        return False, elapsed, None
-
-    elapsed = time.monotonic() - start
-    response_id = body.get("id")
-    record["response_id"] = response_id
-    record["elapsed_seconds"] = elapsed
-    record["usage"] = body.get("usage")
-    record["response"] = body
-    output_path.write_text(json.dumps(record, ensure_ascii=False, indent=2))
-    logger.info("Done: %s/%s id=%s in %.2fs", label, bill_id, response_id, elapsed)
-    return True, elapsed, response_id
-
-
-def main(
-    csv_path: Annotated[Path, typer.Option("--csv", help="Bills CSV path")] = Path("bills.csv"),
-    output_dir: Annotated[Path, typer.Option("--output-dir", help="Where to write per-request JSON")] = Path(
-        "output/openai_runs",
-    ),
-    model: Annotated[str, typer.Option(help="OpenAI model id")] = DEFAULT_MODEL,
-    count: Annotated[int, typer.Option(help="Number of bills per set")] = DEFAULT_COUNT,
-    concurrency: Annotated[int, typer.Option(help="Concurrent in-flight requests")] = 16,
-    log_level: Annotated[str, typer.Option(help="Log level")] = "INFO",
-) -> None:
-    """Run two interactive OpenAI sweeps (compressed + uncompressed) over bill text."""
-    logging.basicConfig(level=log_level, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
-
-    api_key = getenv("CLOSEDAI_TOKEN") or getenv("OPENAI_API_KEY")
-    if not api_key:
-        message = "Neither CLOSEDAI_TOKEN nor OPENAI_API_KEY is set"
-        raise typer.BadParameter(message)
-    if not csv_path.is_file():
-        message = f"CSV not found: {csv_path}"
-        raise typer.BadParameter(message)
-
-    compressed_dir = output_dir / "compressed"
-    uncompressed_dir = output_dir / "uncompressed"
-    compressed_dir.mkdir(parents=True, exist_ok=True)
-    uncompressed_dir.mkdir(parents=True, exist_ok=True)
-
-    logger.info("Loading %d bills from %s", count, csv_path)
-    bills = load_bills(csv_path, count)
-    if len(bills) < count:
-        logger.warning("Only %d bills available (requested %d)", len(bills), count)
-
-    tasks: list[tuple[str, str, str, Path]] = []
-    for bill_id, text_content in bills:
-        filename = f"{safe_filename(bill_id)}.json"
-        tasks.append((bill_id, "compressed", compress_bill_text(text_content), compressed_dir / filename))
-        tasks.append((bill_id, "uncompressed", text_content, uncompressed_dir / filename))
-
-    logger.info("Submitting %d requests at concurrency=%d", len(tasks), concurrency)
-
-    headers = {"Authorization": f"Bearer {api_key}"}
-    completed = 0
-    failed = 0
-    index: list[dict] = []
-    wall_start = time.monotonic()
-    with (
-        httpx.Client(headers=headers, timeout=httpx.Timeout(300.0)) as client,
-        ThreadPoolExecutor(
-            max_workers=concurrency,
-        ) as executor,
-    ):
-        future_to_task = {
-            executor.submit(
-                run_one_request,
-                client,
-                bill_id=bill_id,
-                label=label,
-                bill_text=bill_text,
-                model=model,
-                output_path=output_path,
-            ): (bill_id, label, output_path)
-            for bill_id, label, bill_text, output_path in tasks
-        }
-        for future in as_completed(future_to_task):
-            bill_id, label, output_path = future_to_task[future]
-            success, elapsed, response_id = future.result()
-            if success:
-                completed += 1
-            else:
-                failed += 1
-            index.append(
-                {
-                    "bill_id": bill_id,
-                    "label": label,
-                    "response_id": response_id,
-                    "elapsed_seconds": elapsed,
-                    "success": success,
-                    "path": str(output_path),
-                },
-            )
-    wall_elapsed = time.monotonic() - wall_start
-
-    summary = {
-        "model": model,
-        "count": len(bills),
-        "completed": completed,
-        "failed": failed,
-        "wall_seconds": wall_elapsed,
-        "concurrency": concurrency,
-        "results": index,
-    }
-    summary_path = output_dir / "summary.json"
-    summary_path.write_text(json.dumps(summary, indent=2))
-    logger.info(
-        "Done: completed=%d failed=%d wall=%.1fs summary=%s",
-        completed,
-        failed,
-        wall_elapsed,
-        summary_path,
-    )
-
-
-def cli() -> None:
-    """Typer entry point."""
-    typer.run(main)
-
-
-if __name__ == "__main__":
-    cli()

python/prompt_bench/containers/__init__.py

@@ -1 +0,0 @@
-"""Prompt benchmarking system for evaluating LLMs via vLLM."""

python/prompt_bench/containers/finetune.py

@@ -1,165 +0,0 @@
-"""Docker container lifecycle management for Unsloth fine-tuning."""
-
-from __future__ import annotations
-
-import logging
-import subprocess
-from pathlib import Path
-from typing import Annotated
-
-import typer
-
-from python.prompt_bench.containers.lib import check_gpu_free
-
-logger = logging.getLogger(__name__)
-
-CONTAINER_NAME = "bill-finetune"
-FINETUNE_IMAGE = "bill-finetune:latest"
-DOCKERFILE_PATH = "/home/richie/dotfiles/python/prompt_bench/Dockerfile.finetune"
-DEFAULT_HF_CACHE = Path("/zfs/models/hf")
-
-
-def build_image() -> None:
-    """Build the fine-tuning Docker image."""
-    logger.info("Building fine-tuning image: %s", FINETUNE_IMAGE)
-    result = subprocess.run(
-        ["docker", "build", "-f", DOCKERFILE_PATH, "-t", FINETUNE_IMAGE, "."],
-        text=True,
-        check=False,
-    )
-    if result.returncode != 0:
-        message = "Failed to build fine-tuning image"
-        raise RuntimeError(message)
-    logger.info("Image built: %s", FINETUNE_IMAGE)
-
-
-def start_finetune(
-    *,
-    dataset_path: Path,
-    output_dir: Path,
-    hf_cache: Path = DEFAULT_HF_CACHE,
-) -> None:
-    """Run the fine-tuning container.
-
-    Args:
-        dataset_path: Host path to the fine-tuning JSONL dataset.
-        output_dir: Host path where the trained model will be saved.
-        hf_cache: Host path to HuggingFace model cache (bind-mounted to avoid re-downloading).
-        validation_split: Fraction of data held out for validation.
-    """
-    dataset_path = dataset_path.resolve()
-    output_dir = output_dir.resolve()
-
-    if not dataset_path.is_file():
-        message = f"Dataset not found: {dataset_path}"
-        raise FileNotFoundError(message)
-
-    output_dir.mkdir(parents=True, exist_ok=True)
-    stop_finetune()
-
-    hf_cache = hf_cache.resolve()
-    hf_cache.mkdir(parents=True, exist_ok=True)
-
-    command = [
-        "docker",
-        "run",
-        "--name",
-        CONTAINER_NAME,
-        "--device=nvidia.com/gpu=all",
-        "--ipc=host",
-        "-v",
-        f"{hf_cache}:/root/.cache/huggingface",
-        "-v",
-        f"{output_dir}:/workspace/output/qwen-bill-summarizer",
-        "-v",
-        f"{dataset_path}:/workspace/dataset.jsonl:ro",
-        FINETUNE_IMAGE,
-        "--dataset",
-        "/workspace/dataset.jsonl",
-        "--output-dir",
-        "/workspace/output/qwen-bill-summarizer",
-    ]
-
-    logger.info("Starting fine-tuning container")
-    logger.info("  Dataset:    %s", dataset_path)
-    logger.info("  Output:     %s", output_dir)
-
-    result = subprocess.run(command, text=True, check=False)
-    if result.returncode != 0:
-        message = f"Fine-tuning container exited with code {result.returncode}"
-        raise RuntimeError(message)
-    logger.info("Fine-tuning complete. Model saved to %s", output_dir)
-
-
-def stop_finetune() -> None:
-    """Stop and remove the fine-tuning container."""
-    logger.info("Stopping fine-tuning container")
-    subprocess.run(["docker", "stop", CONTAINER_NAME], capture_output=True, check=False)
-    subprocess.run(["docker", "rm", "-f", CONTAINER_NAME], capture_output=True, check=False)
-
-
-def logs_finetune() -> str | None:
-    """Return recent logs from the fine-tuning container, or None if not running."""
-    result = subprocess.run(
-        ["docker", "logs", "--tail", "50", CONTAINER_NAME],
-        capture_output=True,
-        text=True,
-        check=False,
-    )
-    if result.returncode != 0:
-        return None
-    return result.stdout + result.stderr
-
-
-app = typer.Typer(help="Fine-tuning container management.")
-
-
-@app.command()
-def build() -> None:
-    """Build the fine-tuning Docker image."""
-    build_image()
-
-
-@app.command()
-def run(
-    dataset: Annotated[Path, typer.Option(help="Fine-tuning JSONL")] = Path(
-        "/home/richie/dotfiles/data/finetune_dataset.jsonl"
-    ),
-    output_dir: Annotated[Path, typer.Option(help="Where to save the trained model")] = Path(
-        "/home/richie/dotfiles/data/output/qwen-bill-summarizer",
-    ),
-    hf_cache: Annotated[Path, typer.Option(help="Host path to HuggingFace model cache")] = DEFAULT_HF_CACHE,
-    log_level: Annotated[str, typer.Option(help="Log level")] = "INFO",
-) -> None:
-    """Run fine-tuning inside a Docker container."""
-    logging.basicConfig(level=log_level, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
-    check_gpu_free()
-    start_finetune(
-        dataset_path=dataset,
-        output_dir=output_dir,
-        hf_cache=hf_cache,
-    )
-
-@app.command()
-def stop() -> None:
-    """Stop and remove the fine-tuning container."""
-    stop_finetune()
-
-
-@app.command()
-def logs() -> None:
-    """Show recent logs from the fine-tuning container."""
-    output = logs_finetune()
-    if output is None:
-        typer.echo("No running fine-tuning container found.")
-        raise typer.Exit(code=1)
-    typer.echo(output)
-
-
-def cli() -> None:
-    """Typer entry point."""
-    app()
-
-
-if __name__ == "__main__":
-    cli()

python/prompt_bench/containers/lib.py

@@ -1,23 +0,0 @@
-from __future__ import annotations
-
-import logging
-import subprocess
-
-logger = logging.getLogger(__name__)
-
-
-def check_gpu_free() -> None:
-    """Warn if GPU-heavy processes (e.g. Ollama) are running."""
-    result = subprocess.run(
-        ["nvidia-smi", "--query-compute-apps=pid,process_name", "--format=csv,noheader"],
-        capture_output=True,
-        text=True,
-        check=False,
-    )
-    if result.returncode != 0:
-        logger.warning("Could not query GPU processes: %s", result.stderr.strip())
-        return
-    processes = result.stdout.strip()
-    if processes:
-        logger.warning("GPU processes detected:\n%s", processes)
-        logger.warning("Consider stopping Ollama (sudo systemctl stop ollama) before benchmarking")

python/prompt_bench/containers/vllm.py

@@ -1,70 +0,0 @@
-"""Docker container lifecycle management for vLLM."""
-
-from __future__ import annotations
-
-import logging
-import subprocess
-
-logger = logging.getLogger(__name__)
-
-CONTAINER_NAME = "vllm-bench"
-VLLM_IMAGE = "vllm/vllm-openai:v0.19.0"
-
-
-def start_vllm(
-    *,
-    model: str,
-    port: int,
-    model_dir: str,
-    gpu_memory_utilization: float,
-) -> None:
-    """Start a vLLM container serving the given model.
-
-    Args:
-        model: HuggingFace model directory name (relative to model_dir).
-        port: Host port to bind.
-        model_dir: Host path containing HuggingFace model directories.
-        gpu_memory_utilization: Fraction of GPU memory to use (0-1).
-    """
-    command = [
-        "docker",
-        "run",
-        "-d",
-        "--name",
-        CONTAINER_NAME,
-        "--device=nvidia.com/gpu=all",
-        "--ipc=host",
-        "-v",
-        f"{model_dir}:/models",
-        "-p",
-        f"{port}:8000",
-        VLLM_IMAGE,
-        "--model",
-        f"/models/{model}",
-        "--served-model-name",
-        model,
-        "--gpu-memory-utilization",
-        str(gpu_memory_utilization),
-        "--max-model-len",
-        "4096",
-    ]
-    logger.info("Starting vLLM container with model: %s", model)
-    stop_vllm()
-    result = subprocess.run(command, capture_output=True, text=True, check=False)
-    if result.returncode != 0:
-        msg = f"Failed to start vLLM container: {result.stderr.strip()}"
-        raise RuntimeError(msg)
-    logger.info("vLLM container started: %s", result.stdout.strip()[:12])
-
-
-def stop_vllm() -> None:
-    """Stop and remove the vLLM benchmark container."""
-    logger.info("Stopping vLLM container")
-    subprocess.run(["docker", "stop", CONTAINER_NAME], capture_output=True, check=False)
-    subprocess.run(["docker", "rm", "-f", CONTAINER_NAME], capture_output=True, check=False)
-    subprocess.run(
-        ["docker", "network", "disconnect", "-f", "bridge", CONTAINER_NAME],
-        capture_output=True,
-        check=False,
-    )
-    logger.info("vLLM container stopped and removed")

python/prompt_bench/downloader.py

@@ -1,75 +0,0 @@
-"""HuggingFace model downloader."""
-
-from __future__ import annotations
-
-import logging
-from pathlib import Path
-from typing import Annotated
-
-import typer
-from huggingface_hub import snapshot_download
-
-from python.prompt_bench.models import BenchmarkConfig
-
-logger = logging.getLogger(__name__)
-
-
-def local_model_path(repo: str, model_dir: str) -> Path:
-    """Return the local directory path for a HuggingFace repo."""
-    return Path(model_dir) / repo
-
-
-def is_model_present(repo: str, model_dir: str) -> bool:
-    """Check if a model has already been downloaded."""
-    path = local_model_path(repo, model_dir)
-    return path.exists() and any(path.iterdir())
-
-
-def download_model(repo: str, model_dir: str) -> Path:
-    """Download a HuggingFace model to the local model directory.
-
-    Skips the download if the model directory already exists and contains files.
-    """
-    local_path = local_model_path(repo, model_dir)
-
-    if is_model_present(repo, model_dir):
-        logger.info("Model already exists: %s", local_path)
-        return local_path
-
-    logger.info("Downloading model: %s -> %s", repo, local_path)
-    snapshot_download(
-        repo_id=repo,
-        local_dir=str(local_path),
-    )
-    logger.info("Download complete: %s", repo)
-    return local_path
-
-
-def download_all(config: BenchmarkConfig) -> None:
-    """Download every model listed in the config, top to bottom."""
-    for repo in config.models:
-        download_model(repo, config.model_dir)
-
-
-def main(
-    config: Annotated[Path, typer.Option(help="Path to TOML config file")] = Path("bench.toml"),
-    log_level: Annotated[str, typer.Option(help="Log level")] = "INFO",
-) -> None:
-    """Download all models listed in the benchmark config."""
-    logging.basicConfig(level=log_level, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
-
-    if not config.is_file():
-        message = f"Config file does not exist: {config}"
-        raise typer.BadParameter(message)
-
-    benchmark_config = BenchmarkConfig.from_toml(config)
-    download_all(benchmark_config)
-
-
-def cli() -> None:
-    """Typer entry point."""
-    typer.run(main)
-
-
-if __name__ == "__main__":
-    cli()

python/prompt_bench/finetune.py

@@ -1,214 +0,0 @@
-"""Fine-tune Qwen 3.5 4B on bill summarization data using Unsloth.
-
-Loads a ChatML-style JSONL dataset (system/user/assistant messages),
-applies QLoRA with 4-bit quantization, and saves the merged model
-in HuggingFace format. Designed for a single RTX 3090 (24GB).
-
-Usage:
-    python -m python.prompt_bench.finetune \
-        --dataset output/finetune_dataset.jsonl \
-        --output-dir output/qwen-bill-summarizer
-"""
-
-from __future__ import annotations
-
-import json
-import logging
-from dataclasses import dataclass
-from pathlib import Path
-from typing import Annotated
-
-import tomllib
-import typer
-from unsloth import FastLanguageModel
-from datasets import Dataset
-from transformers import TrainingArguments
-from trl import SFTTrainer
-
-logger = logging.getLogger(__name__)
-
-
-@dataclass
-class LoraConfig:
-    """LoRA adapter hyperparameters."""
-
-    rank: int
-    alpha: int
-    dropout: float
-    targets: list[str]
-
-
-@dataclass
-class TrainingConfig:
-    """Training loop hyperparameters."""
-
-    learning_rate: float
-    epochs: int
-    batch_size: int
-    gradient_accumulation: int
-    max_seq_length: int
-    warmup_ratio: float
-    weight_decay: float
-    logging_steps: int
-    save_steps: int
-
-
-@dataclass
-class FinetuneConfig:
-    """Top-level finetune configuration."""
-
-    base_model: str
-    lora: LoraConfig
-    training: TrainingConfig
-
-    @classmethod
-    def from_toml(cls, config_path: Path) -> FinetuneConfig:
-        """Load finetune config from a TOML file."""
-        raw = tomllib.loads(config_path.read_text())["finetune"]
-        return cls(
-            base_model=raw["base_model"],
-            lora=LoraConfig(**raw["lora"]),
-            training=TrainingConfig(**raw["training"]),
-        )
-
-
-def _messages_to_chatml(messages: list[dict]) -> str:
-    r"""Convert a message list to Qwen ChatML format.
-
-    Produces:
-        <|im_start|>system\n...\n<|im_end|>
-        <|im_start|>user\n...\n<|im_end|>
-        <|im_start|>assistant\n...\n<|im_end|>
-    """
-    parts = []
-    for message in messages:
-        role = message["role"]
-        content = message["content"]
-        parts.append(f"<|im_start|>{role}\n{content}<|im_end|>")
-    return "\n".join(parts)
-
-
-def load_dataset_from_jsonl(path: Path) -> Dataset:
-    """Load a ChatML JSONL file into a HuggingFace Dataset.
-
-    Each line must have {"messages": [{"role": ..., "content": ...}, ...]}.
-    Pre-formats into a `text` column with the Qwen ChatML template applied,
-    which SFTTrainer consumes directly.
-    """
-    records = []
-    with path.open(encoding="utf-8") as handle:
-        for raw_line in handle:
-            stripped = raw_line.strip()
-            if stripped:
-                entry = json.loads(stripped)
-                records.append({"text": _messages_to_chatml(entry["messages"])})
-    logger.info("Loaded %d examples from %s", len(records), path)
-    return Dataset.from_list(records)
-
-
-def main(
-    dataset_path: Annotated[Path, typer.Option("--dataset", help="Fine-tuning JSONL")] = Path(
-        "output/finetune_dataset.jsonl",
-    ),
-    validation_split: Annotated[float, typer.Option("--val-split", help="Fraction held out for validation")] = 0.1,
-    output_dir: Annotated[Path, typer.Option("--output-dir", help="Where to save the merged model")] = Path(
-        "output/qwen-bill-summarizer",
-    ),
-    config_path: Annotated[
-        Path,
-        typer.Option("--config", help="TOML config file"),
-    ] = Path(__file__).parent / "config.toml",
-    save_gguf: Annotated[bool, typer.Option("--save-gguf/--no-save-gguf", help="Also save GGUF")] = False,
-) -> None:
-    """Fine-tune Qwen 3.5 4B on bill summarization with Unsloth + QLoRA."""
-    logging.basicConfig(level="INFO", format="%(asctime)s %(levelname)s %(name)s: %(message)s")
-
-    if not dataset_path.is_file():
-        message = f"Dataset not found: {dataset_path}"
-        raise typer.BadParameter(message)
-
-    config = FinetuneConfig.from_toml(config_path)
-
-    logger.info("Loading base model: %s", config.base_model)
-    model, tokenizer = FastLanguageModel.from_pretrained(
-        model_name=config.base_model,
-        max_seq_length=config.training.max_seq_length,
-        load_in_4bit=True,
-        dtype=None,
-    )
-
-    logger.info("Applying LoRA (rank=%d, alpha=%d)", config.lora.rank, config.lora.alpha)
-    model = FastLanguageModel.get_peft_model(
-        model,
-        r=config.lora.rank,
-        lora_alpha=config.lora.alpha,
-        lora_dropout=config.lora.dropout,
-        target_modules=config.lora.targets,
-        bias="none",
-        use_gradient_checkpointing="unsloth",
-        random_state=42,
-    )
-
-    full_dataset = load_dataset_from_jsonl(dataset_path)
-    split = full_dataset.train_test_split(test_size=validation_split, seed=42)
-    train_dataset = split["train"]
-    validation_dataset = split["test"]
-    logger.info("Split: %d train, %d validation", len(train_dataset), len(validation_dataset))
-    training_args = TrainingArguments(
-        output_dir=str(output_dir / "checkpoints"),
-        num_train_epochs=config.training.epochs,
-        per_device_train_batch_size=config.training.batch_size,
-        gradient_accumulation_steps=config.training.gradient_accumulation,
-        learning_rate=config.training.learning_rate,
-        warmup_ratio=config.training.warmup_ratio,
-        weight_decay=config.training.weight_decay,
-        lr_scheduler_type="cosine",
-        logging_steps=config.training.logging_steps,
-        save_steps=config.training.save_steps,
-        save_total_limit=3,
-        eval_strategy="steps",
-        eval_steps=config.training.save_steps,
-        load_best_model_at_end=True,
-        bf16=True,
-        optim="adamw_8bit",
-        seed=42,
-        report_to="none",
-    )
-
-    trainer = SFTTrainer(
-        model=model,
-        tokenizer=tokenizer,
-        train_dataset=train_dataset,
-        eval_dataset=validation_dataset,
-        args=training_args,
-        max_seq_length=config.training.max_seq_length,
-        packing=True,
-    )
-
-    logger.info(
-        "Starting training: %d train, %d val, %d epochs",
-        len(train_dataset),
-        len(validation_dataset),
-        config.training.epochs,
-    )
-    trainer.train()
-
-    merged_path = str(output_dir / "merged")
-    logger.info("Saving merged model to %s", merged_path)
-    model.save_pretrained_merged(merged_path, tokenizer, save_method="merged_16bit")
-
-    if save_gguf:
-        gguf_path = str(output_dir / "gguf")
-        logger.info("Saving GGUF to %s", gguf_path)
-        model.save_pretrained_gguf(gguf_path, tokenizer, quantization_method="q4_k_m")
-
-    logger.info("Done! Model saved to %s", output_dir)
-
-
-def cli() -> None:
-    """Typer entry point."""
-    typer.run(main)
-
-
-if __name__ == "__main__":
-    cli()

python/prompt_bench/main.py

@@ -1,215 +0,0 @@
-"""CLI entry point for the prompt benchmarking system."""
-
-from __future__ import annotations
-
-import json
-import logging
-import time
-from concurrent.futures import ThreadPoolExecutor, as_completed
-from pathlib import Path
-from typing import Annotated
-
-import typer
-
-from python.prompt_bench.containers.lib import check_gpu_free
-from python.prompt_bench.containers.vllm import start_vllm, stop_vllm
-from python.prompt_bench.downloader import is_model_present
-from python.prompt_bench.models import BenchmarkConfig
-from python.prompt_bench.vllm_client import VLLMClient
-
-logger = logging.getLogger(__name__)
-
-
-def discover_prompts(input_dir: Path) -> list[Path]:
-    """Find all .txt files in the input directory."""
-    prompts = list(input_dir.glob("*.txt"))
-    if not prompts:
-        message = f"No .txt files found in {input_dir}"
-        raise FileNotFoundError(message)
-    return prompts
-
-
-def _run_prompt(
-    client: VLLMClient,
-    prompt_path: Path,
-    *,
-    repo: str,
-    model_dir_name: str,
-    model_output: Path,
-    temperature: float,
-) -> tuple[bool, float]:
-    """Run a single prompt. Returns (success, elapsed_seconds)."""
-    filename = prompt_path.name
-    output_path = model_output / filename
-    start = time.monotonic()
-    try:
-        prompt_text = prompt_path.read_text()
-        response = client.complete(prompt_text, model_dir_name, temperature=temperature)
-        output_path.write_text(response)
-        elapsed = time.monotonic() - start
-        logger.info("Completed: %s / %s in %.2fs", repo, filename, elapsed)
-    except Exception:
-        elapsed = time.monotonic() - start
-        error_path = model_output / f"{filename}.error"
-        logger.exception("Failed: %s / %s after %.2fs", repo, filename, elapsed)
-        error_path.write_text(f"Error processing {filename}")
-        return False, elapsed
-    return True, elapsed
-
-
-def benchmark_model(
-    client: VLLMClient,
-    prompts: list[Path],
-    *,
-    repo: str,
-    model_dir_name: str,
-    model_output: Path,
-    temperature: float,
-    concurrency: int,
-) -> tuple[int, int]:
-    """Run all prompts against a single model in parallel.
-
-    vLLM batches concurrent requests internally, so submitting many at once is
-    significantly faster than running them serially.
-    """
-    pending = [prompt for prompt in prompts if not (model_output / prompt.name).exists()]
-    skipped = len(prompts) - len(pending)
-    if skipped:
-        logger.info("Skipping %d prompts with existing output for %s", skipped, repo)
-
-    if not pending:
-        logger.info("Nothing to do for %s", repo)
-        return 0, 0
-
-    completed = 0
-    failed = 0
-    latencies: list[float] = []
-
-    wall_start = time.monotonic()
-    with ThreadPoolExecutor(max_workers=concurrency) as executor:
-        futures = [
-            executor.submit(
-                _run_prompt,
-                client,
-                prompt_path,
-                repo=repo,
-                model_dir_name=model_dir_name,
-                model_output=model_output,
-                temperature=temperature,
-            )
-            for prompt_path in pending
-        ]
-        for future in as_completed(futures):
-            success, elapsed = future.result()
-            latencies.append(elapsed)
-            if success:
-                completed += 1
-            else:
-                failed += 1
-    wall_elapsed = time.monotonic() - wall_start
-
-    attempted = completed + failed
-    avg_latency = sum(latencies) / attempted
-    throughput = attempted / wall_elapsed if wall_elapsed > 0 else 0.0
-    timing = {
-        "repo": repo,
-        "wall_seconds": wall_elapsed,
-        "attempted": attempted,
-        "completed": completed,
-        "failed": failed,
-        "avg_latency_seconds": avg_latency,
-        "throughput_prompts_per_second": throughput,
-        "concurrency": concurrency,
-    }
-    timing_path = model_output / "_timing.json"
-    timing_path.write_text(json.dumps(timing, indent=2))
-
-    return completed, failed
-
-
-def run_benchmark(
-    config: BenchmarkConfig,
-    input_dir: Path,
-    output_dir: Path,
-) -> None:
-    """Execute the benchmark across all models and prompts."""
-    prompts = discover_prompts(input_dir)
-    logger.info("Found %d prompts in %s", len(prompts), input_dir)
-
-    check_gpu_free()
-
-    total_completed = 0
-    total_failed = 0
-
-    for repo in config.models:
-        if not is_model_present(repo, config.model_dir):
-            logger.warning("Skipping (not downloaded): %s", repo)
-            continue
-
-        model_output = output_dir / repo
-        model_output.mkdir(parents=True, exist_ok=True)
-
-        logger.info("=== Benchmarking model: %s ===", repo)
-
-        stop_vllm()
-        try:
-            start_vllm(
-                model=repo,
-                port=config.port,
-                model_dir=config.model_dir,
-                gpu_memory_utilization=config.gpu_memory_utilization,
-            )
-        except RuntimeError:
-            logger.exception("Failed to start vLLM for %s, skipping", repo)
-            continue
-        logger.info("vLLM started for %s", repo)
-        try:
-            with VLLMClient(port=config.port, timeout=config.timeout) as client:
-                client.wait_ready(max_wait=config.vllm_startup_timeout)
-                completed, failed = benchmark_model(
-                    client,
-                    prompts,
-                    repo=repo,
-                    model_dir_name=repo,
-                    model_output=model_output,
-                    temperature=config.temperature,
-                    concurrency=config.concurrency,
-                )
-                total_completed += completed
-                total_failed += failed
-        finally:
-            stop_vllm()
-
-    logger.info("=== Benchmark complete ===")
-    logger.info("Completed: %d | Failed: %d", total_completed, total_failed)
-
-
-def main(
-    input_dir: Annotated[Path, typer.Argument(help="Directory containing input .txt prompt files")],
-    config: Annotated[Path, typer.Option(help="Path to TOML config file")] = Path("bench.toml"),
-    output_dir: Annotated[Path, typer.Option(help="Output directory for results")] = Path("output"),
-    log_level: Annotated[str, typer.Option(help="Log level")] = "INFO",
-) -> None:
-    """Run prompts through multiple LLMs via vLLM and save results."""
-    logging.basicConfig(level=log_level, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
-
-    if not input_dir.is_dir():
-        message = f"Input directory does not exist: {input_dir}"
-        raise typer.BadParameter(message)
-    if not config.is_file():
-        message = f"Config file does not exist: {config}"
-        raise typer.BadParameter(message)
-
-    benchmark_config = BenchmarkConfig.from_toml(config)
-    output_dir.mkdir(parents=True, exist_ok=True)
-
-    run_benchmark(benchmark_config, input_dir, output_dir)
-
-
-def cli() -> None:
-    """Typer entry point."""
-    typer.run(main)
-
-
-if __name__ == "__main__":
-    cli()

python/prompt_bench/models.py

@@ -1,30 +0,0 @@
-"""Pydantic models for benchmark configuration."""
-
-from __future__ import annotations
-
-import tomllib
-from typing import TYPE_CHECKING
-
-from pydantic import BaseModel
-
-if TYPE_CHECKING:
-    from pathlib import Path
-
-
-class BenchmarkConfig(BaseModel):
-    """Top-level benchmark configuration loaded from TOML."""
-
-    models: list[str]
-    model_dir: str = "/zfs/models/hf"
-    port: int = 8000
-    gpu_memory_utilization: float = 0.90
-    temperature: float = 0.0
-    timeout: int = 300
-    concurrency: int = 4
-    vllm_startup_timeout: int = 900
-
-    @classmethod
-    def from_toml(cls, config_path: Path) -> BenchmarkConfig:
-        """Load benchmark config from a TOML file."""
-        raw = tomllib.loads(config_path.read_text())["bench"]
-        return cls(**raw)

python/prompt_bench/summarization_prompts.py

@@ -1,34 +0,0 @@
-SUMMARIZATION_SYSTEM_PROMPT = """You are a legislative analyst extracting policy substance from Congressional bill text.
-
-Your job is to compress a bill into a dense, neutral structured summary that captures every distinct policy action — including secondary effects that might be buried in subsections.
-
-EXTRACTION RULES:
-- IGNORE: whereas clauses, congressional findings that are purely political statements, recitals, preambles, citations of existing law by number alone, and procedural boilerplate.
-- FOCUS ON: operative verbs — what the bill SHALL do, PROHIBIT, REQUIRE, AUTHORIZE, AMEND, APPROPRIATE, or ESTABLISH.
-- SURFACE ALL THREADS: If the bill touches multiple policy areas, list each thread separately. Do not collapse them.
-- BE CONCRETE: Name the affected population, the mechanism, and the direction (expands/restricts/maintains).
-- STAY NEUTRAL: No political framing. Describe what the text does, not what its sponsors claim it does.
-
-OUTPUT FORMAT — plain structured text, not JSON:
-
-OPERATIVE ACTIONS:
-[Numbered list of what the bill actually does, one action per line, max 20 words each]
-
-AFFECTED POPULATIONS:
-[Who gains something, who loses something, or whose behavior is regulated]
-
-MECHANISMS:
-[How it works: new funding, mandate, prohibition, amendment to existing statute, grant program, study commission, etc.]
-
-POLICY THREADS:
-[List each distinct policy domain this bill touches, even minor ones. Use plain language, not domain codes.]
-
-SYMBOLIC/PROCEDURAL ONLY:
-[Yes or No — is this bill primarily a resolution, designation, or awareness declaration with no operative effect?]
-
-LENGTH TARGET: 150-250 words total. Be ruthless about cutting. Density over completeness."""
-
-SUMMARIZATION_USER_TEMPLATE = """Summarize the following Congressional bill according to your instructions.
-
-BILL TEXT:
-{text_content}"""

python/prompt_bench/tools/build_finetune_dataset.py

@@ -1,114 +0,0 @@
-"""Build a fine-tuning JSONL dataset from batch request + output files.
-
-Joins the original request JSONL (system + user messages) with the batch
-output JSONL (assistant completions) by custom_id to produce a ChatML-style
-messages JSONL suitable for fine-tuning.
-"""
-
-from __future__ import annotations
-
-import json
-import logging
-from pathlib import Path
-from typing import Annotated
-
-import typer
-
-logger = logging.getLogger(__name__)
-
-HTTP_OK = 200
-
-
-def load_requests(path: Path) -> dict[str, list[dict]]:
-    """Parse request JSONL into {custom_id: messages}."""
-    results: dict[str, list[dict]] = {}
-    with path.open(encoding="utf-8") as handle:
-        for raw_line in handle:
-            stripped = raw_line.strip()
-            if not stripped:
-                continue
-            record = json.loads(stripped)
-            custom_id = record["custom_id"]
-            messages = record["body"]["messages"]
-            results[custom_id] = messages
-    return results
-
-
-def load_completions(path: Path) -> dict[str, str]:
-    """Parse batch output JSONL into {custom_id: assistant_content}."""
-    results: dict[str, str] = {}
-    with path.open(encoding="utf-8") as handle:
-        for line_number, raw_line in enumerate(handle, 1):
-            stripped = raw_line.strip()
-            if not stripped:
-                continue
-            record = json.loads(stripped)
-            custom_id = record["custom_id"]
-            response = record.get("response", {})
-            if response.get("status_code") != HTTP_OK:
-                logger.warning("Skipping %s (line %d): status %s", custom_id, line_number, response.get("status_code"))
-                continue
-            body = response.get("body", {})
-            choices = body.get("choices", [])
-            if not choices:
-                logger.warning("Skipping %s (line %d): no choices", custom_id, line_number)
-                continue
-            content = choices[0].get("message", {}).get("content", "")
-            if not content:
-                logger.warning("Skipping %s (line %d): empty content", custom_id, line_number)
-                continue
-            results[custom_id] = content
-    return results
-
-
-def main(
-    requests_path: Annotated[Path, typer.Option("--requests", help="Batch request JSONL")] = Path(
-        "output/openai_batch/requests.jsonl",
-    ),
-    batch_output: Annotated[Path, typer.Option("--batch-output", help="Batch output JSONL")] = Path(
-        "batch_69d84558d91c819091d53f08d78f9fd6_output.jsonl",
-    ),
-    output_path: Annotated[Path, typer.Option("--output", help="Fine-tuning JSONL output")] = Path(
-        "output/finetune_dataset.jsonl",
-    ),
-    log_level: Annotated[str, typer.Option(help="Log level")] = "INFO",
-) -> None:
-    """Build fine-tuning dataset by joining request and output JSONL files."""
-    logging.basicConfig(level=log_level, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
-
-    logger.info("Loading requests from %s", requests_path)
-    requests = load_requests(requests_path)
-    logger.info("Loaded %d requests", len(requests))
-
-    logger.info("Loading completions from %s", batch_output)
-    completions = load_completions(batch_output)
-    logger.info("Loaded %d completions", len(completions))
-
-    output_path.parent.mkdir(parents=True, exist_ok=True)
-    matched = 0
-    skipped = 0
-
-    with output_path.open("w", encoding="utf-8") as handle:
-        for custom_id, messages in requests.items():
-            assistant_content = completions.get(custom_id)
-            if assistant_content is None:
-                skipped += 1
-                continue
-
-            example = {
-                "messages": [*messages, {"role": "assistant", "content": assistant_content}],
-            }
-            handle.write(json.dumps(example, ensure_ascii=False))
-            handle.write("\n")
-            matched += 1
-
-    logger.info("Wrote %d examples to %s (skipped %d unmatched)", matched, output_path, skipped)
-
-
-def cli() -> None:
-    """Typer entry point."""
-    typer.run(main)
-
-
-if __name__ == "__main__":
-    cli()

python/prompt_bench/tools/count_tokens.py

@@ -1,97 +0,0 @@
-"""Sum token usage across compressed and uncompressed run directories."""
-
-from __future__ import annotations
-
-import json
-import logging
-from dataclasses import dataclass, field
-from pathlib import Path
-from typing import Annotated
-
-import typer
-
-logger = logging.getLogger(__name__)
-
-
-@dataclass
-class UsageTotals:
-    """Aggregate usage counters for a directory of run records."""
-
-    files: int = 0
-    errors: int = 0
-    prompt_tokens: int = 0
-    cached_tokens: int = 0
-    completion_tokens: int = 0
-    reasoning_tokens: int = 0
-    total_tokens: int = 0
-    per_file: list[tuple[str, int, int, int]] = field(default_factory=list)
-
-
-def tally_directory(directory: Path) -> UsageTotals:
-    """Return aggregated usage stats for every JSON record in a directory."""
-    totals = UsageTotals()
-    decoder = json.JSONDecoder()
-    for path in sorted(directory.glob("*.json")):
-        text = path.read_text().lstrip()
-        record, _ = decoder.raw_decode(text)
-        totals.files += 1
-        usage = record.get("usage")
-        if not usage:
-            totals.errors += 1
-            continue
-        prompt_tokens = usage.get("prompt_tokens", 0)
-        completion_tokens = usage.get("completion_tokens", 0)
-        total_tokens = usage.get("total_tokens", 0)
-        cached_tokens = (usage.get("prompt_tokens_details") or {}).get("cached_tokens", 0)
-        reasoning_tokens = (usage.get("completion_tokens_details") or {}).get("reasoning_tokens", 0)
-        totals.prompt_tokens += prompt_tokens
-        totals.completion_tokens += completion_tokens
-        totals.total_tokens += total_tokens
-        totals.cached_tokens += cached_tokens
-        totals.reasoning_tokens += reasoning_tokens
-        totals.per_file.append((path.name, prompt_tokens, completion_tokens, total_tokens))
-    return totals
-
-
-def log_totals(label: str, totals: UsageTotals) -> None:
-    """Log a one-block summary for a directory."""
-    counted = totals.files - totals.errors
-    average_total = totals.total_tokens / counted if counted else 0
-    logger.info("[%s]", label)
-    logger.info("  files          : %d (with usage: %d, errors: %d)", totals.files, counted, totals.errors)
-    logger.info("  prompt tokens  : %d", totals.prompt_tokens)
-    logger.info("  cached tokens  : %d", totals.cached_tokens)
-    logger.info("  completion tok : %d", totals.completion_tokens)
-    logger.info("  reasoning tok  : %d", totals.reasoning_tokens)
-    logger.info("  total tokens   : %d", totals.total_tokens)
-    logger.info("  avg total/file : %.1f", average_total)
-
-
-def main(
-    runs_dir: Annotated[Path, typer.Option("--runs-dir")] = Path("output/openai_runs_temp_1"),
-    log_level: Annotated[str, typer.Option("--log-level")] = "INFO",
-) -> None:
-    """Print token usage totals for the compressed and uncompressed run directories."""
-    logging.basicConfig(level=log_level, format="%(message)s")
-
-    grand = UsageTotals()
-    for label in ("compressed", "uncompressed"):
-        directory = runs_dir / label
-        if not directory.is_dir():
-            logger.warning("%s: directory not found at %s", label, directory)
-            continue
-        totals = tally_directory(directory)
-        log_totals(label, totals)
-        grand.files += totals.files
-        grand.errors += totals.errors
-        grand.prompt_tokens += totals.prompt_tokens
-        grand.cached_tokens += totals.cached_tokens
-        grand.completion_tokens += totals.completion_tokens
-        grand.reasoning_tokens += totals.reasoning_tokens
-        grand.total_tokens += totals.total_tokens
-
-    log_totals("grand total", grand)
-
-
-if __name__ == "__main__":
-    typer.run(main)

python/prompt_bench/vllm_client.py

@@ -1,68 +0,0 @@
-"""OpenAI-compatible client for vLLM's API."""
-
-from __future__ import annotations
-
-import logging
-import time
-from typing import Self
-
-import httpx
-
-logger = logging.getLogger(__name__)
-
-READY_POLL_INTERVAL = 2.0
-
-
-class VLLMClient:
-    """Talk to a vLLM server via its OpenAI-compatible API.
-
-    Args:
-        host: vLLM host.
-        port: vLLM port.
-        timeout: Per-request timeout in seconds.
-    """
-
-    def __init__(self, *, host: str = "localhost", port: int = 8000, timeout: int = 300) -> None:
-        """Create a client connected to a vLLM server."""
-        self._client = httpx.Client(base_url=f"http://{host}:{port}", timeout=timeout)
-
-    def wait_ready(self, max_wait: int) -> None:
-        """Poll /v1/models until the server is ready or timeout."""
-        deadline = time.monotonic() + max_wait
-        while time.monotonic() < deadline:
-            try:
-                response = self._client.get("/v1/models")
-                if response.is_success:
-                    logger.info("vLLM server is ready")
-                    return
-            except httpx.TransportError:
-                pass
-            time.sleep(READY_POLL_INTERVAL)
-        msg = f"vLLM server not ready after {max_wait}s"
-        raise TimeoutError(msg)
-
-    def complete(self, prompt: str, model: str, *, temperature: float = 0.0, max_tokens: int = 4096) -> str:
-        """Send a prompt to /v1/completions and return the response text."""
-        payload = {
-            "model": model,
-            "prompt": prompt,
-            "temperature": temperature,
-            "max_tokens": max_tokens,
-        }
-        logger.info("Sending prompt to %s (%d chars)", model, len(prompt))
-        response = self._client.post("/v1/completions", json=payload)
-        response.raise_for_status()
-        data = response.json()
-        return data["choices"][0]["text"]
-
-    def close(self) -> None:
-        """Close the HTTP client."""
-        self._client.close()
-
-    def __enter__(self) -> Self:
-        """Enter the context manager."""
-        return self
-
-    def __exit__(self, *args: object) -> None:
-        """Close the HTTP client on exit."""
-        self.close()

python/tools/whisper/Dockerfile

@@ -0,0 +1,17 @@
+FROM nvidia/cuda:12.4.1-cudnn-runtime-ubuntu22.04
+
+ENV DEBIAN_FRONTEND=noninteractive \
+    PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends python3 python3-pip ffmpeg \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN pip3 install --no-cache-dir --upgrade pip \
+    && pip3 install --no-cache-dir faster-whisper requests
+
+WORKDIR /app
+COPY python/tools/whisper/inference.py /app/inference.py
+
+ENTRYPOINT ["python3", "/app/inference.py"]

python/tools/whisper/Dockerfile.dockerignore

@@ -0,0 +1,2 @@
+*
+!python/tools/whisper/inference.py

python/tools/whisper/__init__.py

@@ -0,0 +1 @@
+"""Whisper transcription tools (host orchestrator and container entrypoint)."""

python/tools/whisper/inference.py

@@ -0,0 +1,136 @@
+"""Container entrypoint that transcribes a directory of audio files with faster-whisper.
+
+Run inside the whisper-transcribe docker image; segment timestamps are grouped
+into one-minute buckets so the output reads as ``[HH:MM:00] text``.
+"""
+
+from __future__ import annotations
+
+import argparse
+import logging
+from pathlib import Path
+
+from faster_whisper import WhisperModel
+
+logger = logging.getLogger(__name__)
+
+AUDIO_EXTENSIONS = {".mp3", ".wav", ".m4a", ".flac", ".ogg", ".opus", ".mp4", ".mkv", ".webm", ".aac"}
+BUCKET_SECONDS = 60
+BEAM_SIZE = 5
+SECONDS_PER_HOUR = 3600
+SECONDS_PER_MINUTE = 60
+
+
+def format_timestamp(total_seconds: float) -> str:
+    """Render a whole-minute timestamp as ``HH:MM:00``.
+
+    Args:
+        total_seconds: Offset in seconds from the start of the audio.
+
+    Returns:
+        A zero-padded ``HH:MM:00`` string.
+    """
+    hours = int(total_seconds // SECONDS_PER_HOUR)
+    minutes = int((total_seconds % SECONDS_PER_HOUR) // SECONDS_PER_MINUTE)
+    return f"{hours:02d}:{minutes:02d}:00"
+
+
+def transcribe_file(model: WhisperModel, audio_path: Path, output_path: Path) -> None:
+    """Transcribe one audio file and write the bucketed transcript to disk.
+
+    Args:
+        model: Loaded faster-whisper model.
+        audio_path: Source audio file.
+        output_path: Destination ``.txt`` path.
+    """
+    logger.info("Transcribing %s", audio_path)
+    segments, info = model.transcribe(
+        str(audio_path),
+        language="en",
+        beam_size=BEAM_SIZE,
+        vad_filter=True,
+    )
+    logger.info("Duration %.1fs", info.duration)
+
+    buckets: dict[int, list[str]] = {}
+    for segment in segments:
+        bucket = int(segment.start // BUCKET_SECONDS)
+        buckets.setdefault(bucket, []).append(segment.text.strip())
+
+    lines = [f"[{format_timestamp(bucket * BUCKET_SECONDS)}] {' '.join(buckets[bucket])}" for bucket in sorted(buckets)]
+    output_path.write_text("\n\n".join(lines) + "\n", encoding="utf-8")
+    logger.info("Wrote %s", output_path)
+
+
+def find_audio_files(input_directory: Path) -> list[Path]:
+    """Collect every audio file under ``input_directory``.
+
+    Args:
+        input_directory: Directory to walk recursively.
+
+    Returns:
+        Sorted list of audio file paths.
+    """
+    return sorted(
+        path for path in input_directory.rglob("*") if path.is_file() and path.suffix.lower() in AUDIO_EXTENSIONS
+    )
+
+
+def configure_container_logger() -> None:
+    """Configure logging for the container (stdout, INFO)."""
+    logging.basicConfig(
+        level=logging.INFO,
+        format="%(asctime)s %(levelname)s %(message)s",
+    )
+
+
+def parse_arguments() -> argparse.Namespace:
+    """Parse CLI arguments for the container entrypoint.
+
+    Returns:
+        Parsed argparse namespace.
+    """
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--input", type=Path, default=Path("/audio"))
+    parser.add_argument("--output", type=Path, default=Path("/output"))
+    parser.add_argument("--model", default="large-v3")
+    parser.add_argument(
+        "--download-only",
+        action="store_true",
+        help="Download the model into the cache volume and exit without transcribing.",
+    )
+    return parser.parse_args()
+
+
+def main() -> None:
+    """Load the model, then either exit (download-only) or transcribe the directory."""
+    configure_container_logger()
+    arguments = parse_arguments()
+
+    logger.info("Loading model %s on CUDA", arguments.model)
+    model = WhisperModel(arguments.model, device="cuda", compute_type="float16")
+
+    if arguments.download_only:
+        logger.info("Model ready; exiting (download-only mode)")
+        return
+
+    arguments.output.mkdir(parents=True, exist_ok=True)
+
+    audio_files = find_audio_files(arguments.input)
+    if not audio_files:
+        logger.warning("No audio files found in %s", arguments.input)
+        return
+
+    logger.info("Found %d audio file(s)", len(audio_files))
+    for audio_path in audio_files:
+        relative = audio_path.relative_to(arguments.input)
+        output_path = arguments.output / relative.with_suffix(".txt")
+        output_path.parent.mkdir(parents=True, exist_ok=True)
+        if output_path.exists():
+            logger.info("Skip %s (already transcribed)", relative)
+            continue
+        transcribe_file(model, audio_path, output_path)
+
+
+if __name__ == "__main__":
+    main()

python/tools/whisper/transcribe.py

@@ -0,0 +1,167 @@
+"""Build and run the whisper transcription docker container on demand.
+
+The container is started fresh for each invocation and removed on exit
+(``docker run --rm``). The model is cached in a named docker volume so
+only the first run pays the download cost.
+"""
+
+from __future__ import annotations
+
+import logging
+import subprocess
+from pathlib import Path
+from typing import Annotated
+
+import typer
+
+from python.common import configure_logger
+
+logger = logging.getLogger(__name__)
+
+
+class Config:
+    """Paths and names for the whisper-transcribe Docker workflow."""
+
+    image_tag = "whisper-transcribe:latest"
+    model_volume = "whisper-models"
+    repo_root = Path(__file__).resolve().parents[3]
+    dockerfile = Path(__file__).resolve().parent / "Dockerfile"
+    huggingface_cache = "/root/.cache/huggingface"
+
+
+def run_docker(arguments: list[str]) -> None:
+    """Run a docker subcommand, streaming output and raising on failure.
+
+    Args:
+        arguments: Arguments to pass to the ``docker`` binary.
+
+    Raises:
+        subprocess.CalledProcessError: If docker exits non-zero.
+    """
+    logger.info("docker %s", " ".join(arguments))
+    subprocess.run(["docker", *arguments], check=True)
+
+
+def build_image() -> None:
+    """Build the whisper-transcribe image using the repo root as build context."""
+    logger.info("Building image %s", Config.image_tag)
+    run_docker(
+        [
+            "build",
+            "--tag",
+            Config.image_tag,
+            "--file",
+            str(Config.dockerfile),
+            str(Config.repo_root),
+        ],
+    )
+
+
+def model_cache_present(model: str) -> bool:
+    """Check whether the given model is already downloaded in the cache volume.
+
+    Args:
+        model: faster-whisper model name (e.g. ``large-v3``).
+
+    Returns:
+        True if the HuggingFace cache directory for the model exists in the volume.
+    """
+    cache_directory = f"hub/models--Systran--faster-whisper-{model}"
+    completed = subprocess.run(
+        [
+            "docker",
+            "run",
+            "--rm",
+            "--volume",
+            f"{Config.model_volume}:/cache",
+            "alpine",
+            "test",
+            "-d",
+            f"/cache/{cache_directory}",
+        ],
+        check=False,
+    )
+    return completed.returncode == 0
+
+
+def download_model(model: str) -> None:
+    """Download the model into the cache volume and exit.
+
+    Args:
+        model: faster-whisper model name.
+    """
+    logger.info("Downloading model %s into volume %s", model, Config.model_volume)
+    run_docker(
+        [
+            "run",
+            "--rm",
+            "--device=nvidia.com/gpu=all",
+            "--ipc=host",
+            "--volume",
+            f"{Config.model_volume}:{Config.huggingface_cache}",
+            Config.image_tag,
+            "--model",
+            model,
+            "--download-only",
+        ],
+    )
+
+
+def transcribe(input_directory: Path, output_directory: Path, model: str) -> None:
+    """Run transcription on every audio file under ``input_directory``.
+
+    Args:
+        input_directory: Host path containing audio files (mounted read-only).
+        output_directory: Host path for ``.txt`` transcripts.
+        model: faster-whisper model name.
+    """
+    logger.info("Transcribing %s -> %s (model=%s)", input_directory, output_directory, model)
+    run_docker(
+        [
+            "run",
+            "--rm",
+            "--device=nvidia.com/gpu=all",
+            "--ipc=host",
+            "--volume",
+            f"{input_directory}:/audio:ro",
+            "--volume",
+            f"{output_directory}:/output",
+            "--volume",
+            f"{Config.model_volume}:{Config.huggingface_cache}",
+            Config.image_tag,
+            "--model",
+            model,
+        ],
+    )
+
+
+def main(
+    input_directory: Annotated[Path, typer.Argument(help="Directory of audio files to transcribe.")],
+    output_directory: Annotated[Path, typer.Argument(help="Directory to write .txt transcripts to.")],
+    model: Annotated[str, typer.Option(help="faster-whisper model name.")] = "large-v3",
+    *,
+    force_download: Annotated[
+        bool,
+        typer.Option("--force-download", help="Re-download the model even if already cached."),
+    ] = False,
+) -> None:
+    """Build the image, ensure the model is cached, then transcribe and stop."""
+    configure_logger()
+
+    resolved_input = input_directory.resolve(strict=True)
+    output_directory.mkdir(parents=True, exist_ok=True)
+    resolved_output = output_directory.resolve()
+
+    build_image()
+
+    if force_download or not model_cache_present(model):
+        download_model(model)
+    else:
+        logger.info("Model %s already cached in volume %s", model, Config.model_volume)
+
+    transcribe(resolved_input, resolved_output, model)
+    logger.info("Done. Container stopped.")
+
+
+if __name__ == "__main__":
+    typer.run(main)

systems/bob/default.nix

@@ -1,11 +1,13 @@
 { inputs, pkgs, ... }:
 {
   imports = [
-    "${inputs.self}/users/richie"
     "${inputs.self}/users/math"
+    "${inputs.self}/users/richie"
+    "${inputs.self}/users/steve"
     "${inputs.self}/common/global"
     "${inputs.self}/common/optional/docker.nix"
     "${inputs.self}/common/optional/scanner.nix"
+    "${inputs.self}/common/optional/monitoring-agent.nix"
     "${inputs.self}/common/optional/steam.nix"
     "${inputs.self}/common/optional/syncthing_base.nix"
     "${inputs.self}/common/optional/systemd-boot.nix"

systems/bob/hardware.nix

@@ -28,9 +28,13 @@
         allowDiscards = true;
         keyFileSize = 4096;
         keyFile = "/dev/disk/by-id/usb-Samsung_Flash_Drive_FIT_0374620080067131-0:0";
-        fallbackToPassword = true;
       };
     };
+
+    zfs.extraPools = [
+      "storage"
+    ];
+
     kernelModules = [ "kvm-amd" ];
     extraModulePackages = [ ];
   };

systems/bob/llms.nix

@@ -42,11 +42,14 @@
       "qwen3:8b"
       "qwen3.5:27b"
       "qwen3.5:35b"
+      "qwen3.6:27b"
+      "qwen3.6:35b"
+      "rinex20/translategemma3:12b"
       "translategemma:12b"
       "translategemma:27b"
       "translategemma:4b"
     ];
-    models = "/zfs/models";
+    models = "/zfs/storage/models";
     openFirewall = true;
   };
 }

systems/bob/scripts/zfs.sh

@@ -1,11 +0,0 @@
-#!/bin/bash
-
-# zpools
-
-# storage
-sudo zpool create -f -o ashift=12 -O acltype=posixacl -O atime=off -O dnodesize=auto -O xattr=sa -O compression=zstd -m /zfs/storage storage mirror
-sudo zpool create -o ashift=12 -O acltype=posixacl -O atime=off -O dnodesize=auto -O xattr=sa -O compression=zstd -m /zfs/storage storage
-
-
-# storage datasets
-sudo zfs create storage/models -o recordsize=1M

systems/bob/snapshot_config.toml

@@ -24,6 +24,6 @@ monthly = 0
 
 ["root_pool/models"]
 15_min = 4
-hourly = 2
+hourly = 24
 daily = 0
 monthly = 0

systems/bob/syncthing.nix

@@ -31,5 +31,15 @@
       ];
       fsWatcherEnabled = true;
     };
+    "recordings" = {
+      path = "/home/richie/recordings";
+      devices = [
+        "jeeves"
+        "phone"
+        "rhapsody-in-green"
+      ];
+      fsWatcherEnabled = true;
+    };
+
   };
 }

systems/brain/hardware.nix

@@ -26,7 +26,6 @@
         allowDiscards = true;
         keyFileSize = 4096;
         keyFile = "/dev/disk/by-id/usb-USB_SanDisk_3.2Gen1_03021630090925173333-0:0";
-        fallbackToPassword = true;
       };
     };
     kernelModules = [ "kvm-intel" ];

systems/jeeves/default.nix

@@ -4,17 +4,21 @@ let
 in
 {
   imports = [
-    "${inputs.self}/users/richie"
-    "${inputs.self}/users/math"
     "${inputs.self}/users/dov"
+    "${inputs.self}/users/math"
+    "${inputs.self}/users/richie"
+    "${inputs.self}/users/steve"
     "${inputs.self}/common/global"
     "${inputs.self}/common/optional/docker.nix"
+    "${inputs.self}/common/optional/monitoring-agent.nix"
     "${inputs.self}/common/optional/ssh_decrypt.nix"
     "${inputs.self}/common/optional/syncthing_base.nix"
     "${inputs.self}/common/optional/update.nix"
     "${inputs.self}/common/optional/zerotier.nix"
+    ./monitoring
     ./docker
     ./services
+    ./web_services
     ./hardware.nix
     ./networking.nix
     ./programs.nix
@@ -35,5 +39,10 @@ in
     zerotierone.joinNetworks = [ "a09acf02330d37b9" ];
   };
 
+  users.groups = {
+    nornsight = { };
+    nornsight-admin = { };
+  };
+
   system.stateVersion = "24.05";
 }

systems/jeeves/hardware.nix

@@ -9,7 +9,6 @@ let
     inherit device;
     keyFileSize = 4096;
     keyFile = "/dev/disk/by-id/usb-XIAO_USB_Drive_24587CE29074-0:0";
-    fallbackToPassword = true;
   };
   makeLuksSSD =
     device:

systems/jeeves/monitoring/dashboards/overview.json

@@ -0,0 +1,426 @@
+{
+  "annotations": {
+    "list": [
+      {
+        "builtIn": 1,
+        "datasource": {
+          "type": "grafana",
+          "uid": "-- Grafana --"
+        },
+        "enable": true,
+        "hide": true,
+        "iconColor": "rgba(0, 211, 255, 1)",
+        "name": "Annotations & Alerts",
+        "type": "dashboard"
+      }
+    ]
+  },
+  "editable": true,
+  "fiscalYearStartMonth": 0,
+  "graphTooltip": 0,
+  "links": [],
+  "panels": [
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "percent"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 6,
+        "x": 0,
+        "y": 0
+      },
+      "id": 1,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "100 * (1 - avg by (instance) (rate(node_cpu_seconds_total{mode=\"idle\"}[5m])))",
+          "legendFormat": "{{instance}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "CPU Used",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "percent"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 6,
+        "x": 6,
+        "y": 0
+      },
+      "id": 2,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "100 * (1 - (node_memory_MemAvailable_bytes / node_memory_MemTotal_bytes))",
+          "legendFormat": "{{instance}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "RAM Used",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "percent"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 6,
+        "x": 12,
+        "y": 0
+      },
+      "id": 3,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "100 * (1 - (node_memory_SwapFree_bytes / node_memory_SwapTotal_bytes))",
+          "legendFormat": "{{instance}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Swap Used",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "short"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 6,
+        "x": 18,
+        "y": 0
+      },
+      "id": 4,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "node_load1",
+          "legendFormat": "{{instance}} load1",
+          "range": true,
+          "refId": "A"
+        },
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "node_load5",
+          "legendFormat": "{{instance}} load5",
+          "range": true,
+          "refId": "B"
+        },
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "node_load15",
+          "legendFormat": "{{instance}} load15",
+          "range": true,
+          "refId": "C"
+        }
+      ],
+      "title": "Load",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "Bps"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 9,
+        "w": 12,
+        "x": 0,
+        "y": 8
+      },
+      "id": 5,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "sum by (instance) (rate(node_disk_read_bytes_total[5m]))",
+          "legendFormat": "{{instance}} read",
+          "range": true,
+          "refId": "A"
+        },
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "sum by (instance) (rate(node_disk_written_bytes_total[5m]))",
+          "legendFormat": "{{instance}} write",
+          "range": true,
+          "refId": "B"
+        }
+      ],
+      "title": "Disk Throughput",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "percent"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 9,
+        "w": 12,
+        "x": 12,
+        "y": 8
+      },
+      "id": 6,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "100 * (1 - (node_filesystem_avail_bytes{mountpoint=~\"(/|/home|/var|/zfs.*)\",fstype!=\"\"} / node_filesystem_size_bytes{mountpoint=~\"(/|/home|/var|/zfs.*)\",fstype!=\"\"}))",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{mountpoint}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Filesystem Usage",
+      "type": "table"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "percentunit"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 0,
+        "y": 17
+      },
+      "id": 7,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(10, rate(namedprocess_namegroup_cpu_seconds_total[5m]))",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{groupname}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Top Grouped CPU",
+      "type": "table"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "bytes"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 12,
+        "y": 17
+      },
+      "id": 8,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(10, namedprocess_namegroup_memory_bytes{memtype=\"resident\"})",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{groupname}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Top Grouped Memory",
+      "type": "table"
+    }
+  ],
+  "refresh": "30s",
+  "schemaVersion": 39,
+  "style": "dark",
+  "tags": [
+    "monitoring"
+  ],
+  "templating": {
+    "list": []
+  },
+  "time": {
+    "from": "now-24h",
+    "to": "now"
+  },
+  "timepicker": {},
+  "timezone": "",
+  "title": "Overview",
+  "uid": "monitor-overview",
+  "version": 1,
+  "weekStart": ""
+}

systems/jeeves/monitoring/dashboards/process-history-grouped.json

@@ -0,0 +1,216 @@
+{
+  "annotations": {
+    "list": [
+      {
+        "builtIn": 1,
+        "datasource": {
+          "type": "grafana",
+          "uid": "-- Grafana --"
+        },
+        "enable": true,
+        "hide": true,
+        "iconColor": "rgba(0, 211, 255, 1)",
+        "name": "Annotations & Alerts",
+        "type": "dashboard"
+      }
+    ]
+  },
+  "editable": true,
+  "fiscalYearStartMonth": 0,
+  "graphTooltip": 0,
+  "links": [],
+  "panels": [
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "percentunit"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 0,
+        "y": 0
+      },
+      "id": 1,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(10, rate(namedprocess_namegroup_cpu_seconds_total[5m]))",
+          "legendFormat": "{{instance}} {{groupname}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Grouped CPU",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "bytes"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 12,
+        "y": 0
+      },
+      "id": 2,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(10, namedprocess_namegroup_memory_bytes{memtype=\"resident\"})",
+          "legendFormat": "{{instance}} {{groupname}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Grouped Resident Memory",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "Bps"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 0,
+        "y": 10
+      },
+      "id": 3,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(10, rate(namedprocess_namegroup_read_bytes_total[5m]))",
+          "legendFormat": "{{instance}} {{groupname}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Grouped Read I/O",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "Bps"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 12,
+        "y": 10
+      },
+      "id": 4,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(10, rate(namedprocess_namegroup_write_bytes_total[5m]))",
+          "legendFormat": "{{instance}} {{groupname}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Grouped Write I/O",
+      "type": "timeseries"
+    }
+  ],
+  "refresh": "30s",
+  "schemaVersion": 39,
+  "style": "dark",
+  "tags": [
+    "monitoring",
+    "process"
+  ],
+  "templating": {
+    "list": []
+  },
+  "time": {
+    "from": "now-7d",
+    "to": "now"
+  },
+  "timepicker": {},
+  "timezone": "",
+  "title": "Process History Grouped",
+  "uid": "monitor-process-history",
+  "version": 1,
+  "weekStart": ""
+}

systems/jeeves/monitoring/dashboards/process-live-pid.json

@@ -0,0 +1,224 @@
+{
+  "annotations": {
+    "list": [
+      {
+        "builtIn": 1,
+        "datasource": {
+          "type": "grafana",
+          "uid": "-- Grafana --"
+        },
+        "enable": true,
+        "hide": true,
+        "iconColor": "rgba(0, 211, 255, 1)",
+        "name": "Annotations & Alerts",
+        "type": "dashboard"
+      }
+    ]
+  },
+  "editable": true,
+  "fiscalYearStartMonth": 0,
+  "graphTooltip": 0,
+  "links": [],
+  "panels": [
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-pid-short"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "percentunit"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 0,
+        "y": 0
+      },
+      "id": 1,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-pid-short"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, rate(namedprocess_namegroup_cpu_seconds_total[2m]))",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{groupname}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Top PID CPU",
+      "type": "table"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-pid-short"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "bytes"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 12,
+        "y": 0
+      },
+      "id": 2,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-pid-short"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, namedprocess_namegroup_memory_bytes{memtype=\"resident\"})",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{groupname}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Top PID RSS",
+      "type": "table"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-pid-short"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "Bps"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 0,
+        "y": 10
+      },
+      "id": 3,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-pid-short"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, rate(namedprocess_namegroup_read_bytes_total[2m]))",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{groupname}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Top PID Read I/O",
+      "type": "table"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-pid-short"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "Bps"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 12,
+        "y": 10
+      },
+      "id": 4,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-pid-short"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, rate(namedprocess_namegroup_write_bytes_total[2m]))",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{groupname}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Top PID Write I/O",
+      "type": "table"
+    }
+  ],
+  "refresh": "15s",
+  "schemaVersion": 39,
+  "style": "dark",
+  "tags": [
+    "monitoring",
+    "process"
+  ],
+  "templating": {
+    "list": []
+  },
+  "time": {
+    "from": "now-10m",
+    "to": "now"
+  },
+  "timepicker": {},
+  "timezone": "",
+  "title": "Process Live PID",
+  "uid": "monitor-process-pid",
+  "version": 1,
+  "weekStart": ""
+}

systems/jeeves/monitoring/dashboards/storage-zfs.json

@@ -0,0 +1,351 @@
+{
+  "annotations": {
+    "list": [
+      {
+        "builtIn": 1,
+        "datasource": {
+          "type": "grafana",
+          "uid": "-- Grafana --"
+        },
+        "enable": true,
+        "hide": true,
+        "iconColor": "rgba(0, 211, 255, 1)",
+        "name": "Annotations & Alerts",
+        "type": "dashboard"
+      }
+    ]
+  },
+  "editable": true,
+  "fiscalYearStartMonth": 0,
+  "graphTooltip": 0,
+  "links": [],
+  "panels": [
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "percent"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 8,
+        "x": 0,
+        "y": 0
+      },
+      "id": 1,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "100 * (zfs_pool_allocated_bytes / zfs_pool_size_bytes)",
+          "legendFormat": "{{instance}} {{pool}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Pool Usage",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "bytes"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 8,
+        "x": 8,
+        "y": 0
+      },
+      "id": 2,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "zfs_pool_free_bytes",
+          "legendFormat": "{{instance}} {{pool}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Pool Free Bytes",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "bytes"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 8,
+        "x": 16,
+        "y": 0
+      },
+      "id": 3,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, zfs_dataset_used_bytes{type=\"filesystem\"})",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{name}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Top Filesystems by Used Bytes",
+      "type": "table"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "ns"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 9,
+        "w": 12,
+        "x": 0,
+        "y": 8
+      },
+      "id": 4,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, zpool_iostat_total_wait_read_ns{vdev!=\"_pool\"})",
+          "legendFormat": "{{host}} {{pool}} {{vdev}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "ZFS Read Wait",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "ns"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 9,
+        "w": 12,
+        "x": 12,
+        "y": 8
+      },
+      "id": 5,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, zpool_iostat_total_wait_write_ns{vdev!=\"_pool\"})",
+          "legendFormat": "{{host}} {{pool}} {{vdev}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "ZFS Write Wait",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "celsius"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 9,
+        "w": 12,
+        "x": 0,
+        "y": 17
+      },
+      "id": 6,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "smartctl_device_temperature{temperature_type=\"current\"}",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{device}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Disk Temperature",
+      "type": "table"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "short"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 9,
+        "w": 12,
+        "x": 12,
+        "y": 17
+      },
+      "id": 7,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": false,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "smartctl_device_smart_status",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{device}}",
+          "refId": "A"
+        }
+      ],
+      "title": "SMART Health",
+      "type": "table"
+    }
+  ],
+  "refresh": "30s",
+  "schemaVersion": 39,
+  "style": "dark",
+  "tags": [
+    "monitoring",
+    "zfs"
+  ],
+  "templating": {
+    "list": []
+  },
+  "time": {
+    "from": "now-24h",
+    "to": "now"
+  },
+  "timepicker": {},
+  "timezone": "",
+  "title": "Storage and ZFS",
+  "uid": "monitor-storage",
+  "version": 1,
+  "weekStart": ""
+}

systems/jeeves/monitoring/default.nix

@@ -0,0 +1,186 @@
+{
+  lib,
+  pkgs,
+  ...
+}:
+let
+  vars = import ../vars.nix;
+
+  prometheusDataRoot = "${vars.database}/prometheus";
+  mainPrometheusDataDir = "${prometheusDataRoot}/main";
+  pidPrometheusDataDir = "${prometheusDataRoot}/pid-short";
+
+  prometheusYaml = pkgs.formats.yaml { };
+
+  mkPrometheusConfig =
+    name: cfg:
+    let
+      configFile = prometheusYaml.generate "${name}.yaml" cfg;
+    in
+    pkgs.runCommand "${name}-checked.yaml"
+      {
+        nativeBuildInputs = [ pkgs.prometheus.cli ];
+      }
+      ''
+        promtool check config ${configFile}
+        cp ${configFile} $out
+      '';
+
+  mkTarget = host: address: {
+    targets = [ address ];
+    labels.instance = host;
+  };
+
+  mainPrometheusConfig = mkPrometheusConfig "prometheus-main" {
+    global = {
+      scrape_interval = "30s";
+      scrape_timeout = "10s";
+      evaluation_interval = "30s";
+    };
+    scrape_configs = [
+      {
+        job_name = "node";
+        static_configs = [
+          (mkTarget "jeeves" "192.168.90.40:9100")
+          (mkTarget "bob" "192.168.90.25:9100")
+        ];
+      }
+      {
+        job_name = "process_grouped";
+        static_configs = [
+          (mkTarget "jeeves" "192.168.90.40:9256")
+          (mkTarget "bob" "192.168.90.25:9256")
+        ];
+      }
+      {
+        job_name = "smartctl";
+        static_configs = [
+          (mkTarget "jeeves" "192.168.90.40:9633")
+          (mkTarget "bob" "192.168.90.25:9633")
+        ];
+      }
+      {
+        job_name = "zfs";
+        static_configs = [
+          (mkTarget "jeeves" "192.168.90.40:9134")
+          (mkTarget "bob" "192.168.90.25:9134")
+        ];
+      }
+    ];
+  };
+
+  pidPrometheusConfig = mkPrometheusConfig "prometheus-pid-short" {
+    global = {
+      scrape_interval = "15s";
+      scrape_timeout = "10s";
+      evaluation_interval = "15s";
+    };
+    scrape_configs = [
+      {
+        job_name = "process_pid";
+        static_configs = [
+          (mkTarget "jeeves" "192.168.90.40:9257")
+          (mkTarget "bob" "192.168.90.25:9257")
+        ];
+      }
+    ];
+  };
+
+  mkPrometheusService =
+    {
+      dataDir,
+      configFile,
+      port,
+      retention,
+    }:
+    {
+      after = [
+        "zfs-media-database-prometheus.mount"
+        "network.target"
+      ];
+      requires = [ "zfs-media-database-prometheus.mount" ];
+      wantedBy = [ "multi-user.target" ];
+      unitConfig.RequiresMountsFor = [ dataDir ];
+      serviceConfig = {
+        ExecStart = "${lib.getExe pkgs.prometheus} ${
+          lib.escapeShellArgs [
+            "--config.file=${configFile}"
+            "--storage.tsdb.path=${dataDir}"
+            "--storage.tsdb.retention.time=${retention}"
+            "--web.listen-address=127.0.0.1:${toString port}"
+          ]
+        }";
+        User = "prometheus";
+        Group = "prometheus";
+        Restart = "always";
+        RestartSec = "5s";
+        WorkingDirectory = dataDir;
+        ReadWritePaths = [ dataDir ];
+        CapabilityBoundingSet = [ "" ];
+        DeviceAllow = [ "/dev/null rw" ];
+        DevicePolicy = "strict";
+        LockPersonality = true;
+        MemoryDenyWriteExecute = true;
+        NoNewPrivileges = true;
+        PrivateDevices = true;
+        PrivateTmp = true;
+        ProtectClock = true;
+        ProtectControlGroups = true;
+        ProtectHome = true;
+        ProtectHostname = true;
+        ProtectKernelLogs = true;
+        ProtectKernelModules = true;
+        ProtectKernelTunables = true;
+        ProtectProc = "invisible";
+        ProtectSystem = "strict";
+        RemoveIPC = true;
+        RestrictAddressFamilies = [
+          "AF_INET"
+          "AF_INET6"
+          "AF_UNIX"
+        ];
+        RestrictNamespaces = true;
+        RestrictRealtime = true;
+        RestrictSUIDSGID = true;
+        SystemCallArchitectures = "native";
+        SystemCallFilter = [
+          "@system-service"
+          "~@privileged"
+        ];
+      };
+    };
+in
+{
+  users = {
+    groups.prometheus = { };
+    users.prometheus = {
+      isSystemUser = true;
+      group = "prometheus";
+      description = "Prometheus daemon user";
+    };
+  };
+
+  systemd = {
+    services = {
+      prometheus-main = mkPrometheusService {
+        configFile = mainPrometheusConfig;
+        dataDir = mainPrometheusDataDir;
+        port = 9090;
+        retention = "90d";
+      };
+
+      prometheus-pid-short = mkPrometheusService {
+        configFile = pidPrometheusConfig;
+        dataDir = pidPrometheusDataDir;
+        port = 9092;
+        retention = "10m";
+      };
+    };
+
+    tmpfiles.rules = [
+      "d ${prometheusDataRoot} 0755 root root - -"
+      "d ${mainPrometheusDataDir} 0750 prometheus prometheus - -"
+      "d ${pidPrometheusDataDir} 0750 prometheus prometheus - -"
+    ];
+  };
+}

systems/jeeves/networking.nix

@@ -1,4 +1,13 @@
 {
+  # Docker loads br_netfilter on jeeves. Disable bridge netfilter so
+  # br-nix-builder behaves like a pure L2 bridge and bridged traffic
+  # does not hit the host firewall/rpfilter path.
+  boot.kernel.sysctl = {
+    "net.bridge.bridge-nf-call-arptables" = 0;
+    "net.bridge.bridge-nf-call-ip6tables" = 0;
+    "net.bridge.bridge-nf-call-iptables" = 0;
+  };
+
   networking = {
     hostName = "jeeves";
     hostId = "0e15ce35";
@@ -34,11 +43,18 @@
       };
     };
     networks = {
-      "10-1GB_Primary" = {
+      "10-Primary" = {
-        matchConfig.Name = "enp97s0f1";
+        matchConfig.Name = "enp97s0";
         address = [ "192.168.99.14/24" ];
+        dns = [
+          "192.168.99.1"
+          "2600:4040:abfb:d700::1"
+        ];
         routes = [ { Gateway = "192.168.99.1"; } ];
         vlan = [ "internet-vlan" ];
+        dhcpV4Config.UseDNS = false;
+        dhcpV6Config.UseDNS = false;
+        ipv6AcceptRAConfig.UseDNS = false;
         linkConfig.RequiredForOnline = "routable";
       };
       "50-internet-vlan" = {
@@ -49,23 +65,10 @@
       "60-br-nix-builder" = {
         matchConfig.Name = "br-nix-builder";
         bridgeConfig = { };
-        address = [ "192.168.3.10/24" ];
+        networkConfig = {
-        routingPolicyRules = [
+          IPv6AcceptRA = false;
-          {
+          LinkLocalAddressing = "no";
-            From = "192.168.3.0/24";
+        };
-            Table = 100;
-            Priority = 100;
-          }
-        ];
-        routes = [
-          {
-            Gateway = "192.168.3.1";
-            Table = 100;
-            GatewayOnLink = false;
-            Metric = 2048;
-            PreferredSource = "192.168.3.10";
-          }
-        ];
         linkConfig.RequiredForOnline = "no";
       };
     };

systems/jeeves/programs.nix

@@ -3,5 +3,6 @@
   environment.systemPackages = with pkgs; [
     filebot
     docker-compose
+    ffmpeg
   ];
 }

systems/jeeves/runners/default.nix

@@ -1,20 +1,7 @@
-{ pkgs, ... }:
+{ ... }:
 {
   imports = [ ./nix_builder.nix ];
 
-  users = {
-    users.github-runners = {
-      shell = pkgs.bash;
-      isSystemUser = true;
-      group = "github-runners";
-      uid = 601;
-      openssh.authorizedKeys.keys = [
-        "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIA/S8i+BNX/12JNKg+5EKGX7Aqimt5KM+ve3wt/SyWuO github-runners" # cspell:disable-line
-      ];
-    };
-    groups.github-runners.gid = 601;
-  };
-
   services.nix_builder.containers = {
     nix-builder-00.enable = true;
     nix-builder-01.enable = true;

systems/jeeves/runners/nix_builder.nix

@@ -2,6 +2,7 @@
   config,
   lib,
   outputs,
+  utils,
   ...
 }:
 
@@ -9,6 +10,8 @@ with lib;
 let
   vars = import ../vars.nix;
   cfg = config.services.nix_builder;
+  runnerUsername = "gitea-runner";
+  runnerUserid = 601;
 in
 {
   options.services.nix_builder = {
@@ -23,37 +26,40 @@ in
         types.submodule (
           { name, ... }:
           {
-            options.enable = mkEnableOption "GitHub runner container";
+            options.enable = mkEnableOption "Gitea runner container";
           }
         )
       );
       default = { };
-      description = "GitHub runner container configurations";
+      description = "Gitea runner container configurations";
     };
   };
 
   config = {
+    users = {
+      users.${runnerUsername} = {
+        isSystemUser = true;
+        group = runnerUsername;
+        uid = runnerUserid;
+      };
+      groups.${runnerUsername}.gid = runnerUserid;
+    };
+
     containers = mapAttrs (
       name: containerCfg:
       mkIf containerCfg.enable {
         autoStart = true;
         privateNetwork = true;
         hostBridge = cfg.bridgeName;
-        ephemeral = true;
         bindMounts = {
-          storage = {
-            hostPath = "/zfs/media/github-runners/${name}";
-            mountPoint = "/zfs/media/github-runners/${name}";
-            isReadOnly = false;
-          };
           host-nix = {
             mountPoint = "/host-nix/var/nix/daemon-socket";
             hostPath = "/nix/var/nix/daemon-socket";
             isReadOnly = false;
           };
-          pat = {
+          token = {
-            hostPath = "${vars.secrets}/services/github-runners/runner_pat";
+            hostPath = "${vars.secrets}/services/gitea-runners";
-            mountPoint = "${vars.secrets}/services/github-runners/runner_pat";
+            mountPoint = "/run/secrets/gitea-runners";
             isReadOnly = true;
           };
         };
@@ -92,46 +98,69 @@ in
                 "nix-command"
               ];
               sandbox = true;
-              allowed-users = [ "github-runners" ];
+              allowed-users = [ "gitea-runner" ];
               trusted-users = [
                 "root"
-                "github-runners"
+                "gitea-runner"
               ];
             };
             nixpkgs = {
               overlays = builtins.attrValues outputs.overlays;
               config.allowUnfree = true;
             };
-            services.github-runners.${name} = {
+            users = {
+              users.${runnerUsername} = {
+                isSystemUser = true;
+                group = runnerUsername;
+                uid = runnerUserid;
+              };
+              groups.${runnerUsername}.gid = runnerUserid;
+            };
+            services.gitea-actions-runner.instances.${name} = {
               enable = true;
-              replace = true;
+              name = "jeeves-${name}";
-              workDir = "/zfs/media/github-runners/${name}";
+              url = "http://192.168.99.14:6443/";
-              url = "https://github.com/RichieCahill/dotfiles";
+              labels = [
-              extraLabels = [ "nixos" ];
+                "self-hosted:host"
-              tokenFile = "${vars.secrets}/services/github-runners/runner_pat";
+                "nixos:host"
-              user = "github-runners";
+              ];
-              group = "github-runners";
+              tokenFile = "/run/secrets/gitea-runners/registration-token";
-              extraPackages = with pkgs; [
+              hostPackages = with pkgs; [
+                bash
+                coreutils
+                curl
+                gawk
                 gitMinimal
-                gh
+                gnused
+                my_python
+                nix
                 nixfmt
                 nixos-rebuild
+                nodejs
                 treefmt
-                my_python
+                wget
               ];
             };
-            users = {
+            systemd.services."gitea-runner-${utils.escapeSystemdPath name}" = {
-              users.github-runners = {
+              serviceConfig = {
-                shell = pkgs.bash;
+                DynamicUser = mkForce false;
-                isSystemUser = true;
+                User = mkForce runnerUsername;
-                group = "github-runners";
+                Group = mkForce runnerUsername;
-                uid = 601;
               };
-              groups.github-runners.gid = 601;
             };
             system.stateVersion = "24.05";
           };
       }
     ) cfg.containers;
+
+    systemd.services = builtins.listToAttrs (
+      map (name: {
+        name = "container@${name}";
+        value = {
+          requires = [ "gitea.service" ];
+          after = [ "gitea.service" ];
+        };
+      }) (builtins.attrNames (filterAttrs (_: c: c.enable) cfg.containers))
+    );
   };
 }

systems/jeeves/scripts/zfs.sh

@@ -23,6 +23,7 @@ sudo zfs create media/secure/home_assistant -o compression=zstd-19
 sudo zfs create media/secure/notes -o copies=2
 sudo zfs create media/secure/postgres -o mountpoint=/zfs/media/database/postgres -o recordsize=16k -o primarycache=metadata
 sudo zfs create media/secure/postgres-wal -o mountpoint=/zfs/media/database/postgres-wal -o recordsize=32k -o primarycache=metadata -o special_small_blocks=32K -o compression=lz4 -o secondarycache=none -o logbias=latency
+sudo zfs create media/secure/prometheus -o mountpoint=/zfs/media/database/prometheus -o compression=lz4
 sudo zfs create media/secure/services -o compression=zstd-9
 sudo zfs create media/secure/share -o mountpoint=/zfs/media/share -o exec=off
 
@@ -41,3 +42,4 @@ sudo zfs create storage/secure/plex -o recordsize=1M -o compression=zstd-19
 sudo zfs create storage/secure/secrets -o compression=zstd-19 -o copies=3
 sudo zfs create storage/secure/syncthing -o compression=zstd-19
 sudo zfs create storage/secure/transmission -o recordsize=1M -o compression=zstd-9 -o exec=off -o sync=disabled
+sudo zfs create storage/secure/important -o compression=zstd-19 -o copies=2 -o mountpoint=/zfs/storage/important

systems/jeeves/services/audiobookshelf.nix

@@ -3,7 +3,10 @@ let
   vars = import ../vars.nix;
 in
 {
-  services.audiobookshelf.enable = true;
+  services.audiobookshelf = {
+    enable = true;
+    port = 8000;
+  };
   systemd.services.audiobookshelf.serviceConfig.WorkingDirectory =
     lib.mkForce "${vars.docker_configs}/audiobookshelf";
   users.users.audiobookshelf.home = lib.mkForce "${vars.docker_configs}/audiobookshelf";

systems/jeeves/services/camofox-browser.nix

@@ -0,0 +1,80 @@
+{
+  ...
+}:
+let
+  vars = import ../vars.nix;
+in
+{
+  systemd.tmpfiles.rules = [
+    "d ${vars.docker_configs}/camofox-browser 0750 root root - -"
+  ];
+
+  containers.camofox-browser = {
+    autoStart = true;
+    privateNetwork = false;
+    bindMounts = {
+      camofox-browser = {
+        hostPath = "${vars.docker_configs}/camofox-browser";
+        mountPoint = "/var/lib/camofox-browser";
+        isReadOnly = false;
+      };
+    };
+    config =
+      {
+        pkgs,
+        lib,
+        ...
+      }:
+      {
+        networking.hostName = "camofox-browser";
+
+        environment.systemPackages = with pkgs; [
+          ffmpeg
+          git
+          nodejs
+          python3Packages.yt-dlp
+        ];
+
+        systemd.services.camofox-browser = {
+          description = "Camofox browser server";
+          wantedBy = [ "multi-user.target" ];
+          after = [ "network.target" ];
+          environment = {
+            CAMOFOX_HOST = "127.0.0.1";
+            CAMOFOX_PORT = "9377";
+            HOME = "/var/lib/camofox-browser";
+          };
+          path = with pkgs; [
+            bash
+            coreutils
+            git
+            nodejs
+          ];
+          serviceConfig = {
+            Restart = "always";
+            RestartSec = "5s";
+            WorkingDirectory = "/var/lib/camofox-browser";
+          };
+          script = ''
+            set -eu
+
+            app_dir=/var/lib/camofox-browser/app
+
+            if [ ! -d "$app_dir/.git" ]; then
+              git clone --depth 1 https://github.com/jo-inc/camofox-browser "$app_dir"
+            fi
+
+            cd "$app_dir"
+
+            if [ ! -d node_modules ]; then
+              npm install
+            fi
+
+            exec npm start
+          '';
+        };
+
+        system.stateVersion = lib.mkDefault "24.05";
+      };
+  };
+}

systems/jeeves/services/cloud_flare_tunnel.nix

@@ -1,17 +0,0 @@
-{ pkgs, ... }:
-let
-  vars = import ../vars.nix;
-in
-{
-  systemd.services.cloud_flare_tunnel = {
-    description = "cloud_flare_tunnel proxy's traffic through cloudflare";
-    after = [ "network.target" ];
-    wantedBy = [ "multi-user.target" ];
-    serviceConfig = {
-      Type = "simple";
-      EnvironmentFile = "${vars.secrets}/docker/cloud_flare_tunnel";
-      ExecStart = "${pkgs.cloudflared}/bin/cloudflared --no-autoupdate tunnel run";
-      Restart = "on-failure";
-    };
-  };
-}

systems/jeeves/services/gitea.nix

@@ -2,7 +2,10 @@ let
   vars = import ../vars.nix;
 in
 {
-  networking.firewall.allowedTCPPorts = [ 6443 ];
+  networking.firewall.allowedTCPPorts = [
+    6443
+    2223
+  ];
 
   services.gitea = {
     enable = true;
@@ -18,13 +21,17 @@ in
       createDatabase = false;
     };
     settings = {
+      actions = {
+        ENABLED = true;
+        DEFAULT_ACTIONS_URL = "github";
+      };
       service.DISABLE_REGISTRATION = true;
       server = {
         DOMAIN = "tmmworkshop.com";
         ROOT_URL = "https://gitea.tmmworkshop.com/";
         HTTP_PORT = 6443;
         SSH_PORT = 2223;
-        SSH_LISTEN_PORT = 2224;
+        SSH_LISTEN_PORT = 2223;
         START_SSH_SERVER = true;
         PUBLIC_URL_DETECTION = "auto";
       };

systems/jeeves/services/grafana.nix

@@ -0,0 +1,80 @@
+{
+  ...
+}:
+let
+  vars = import ../vars.nix;
+  grafanaDataDir = "${vars.services}/grafana";
+in
+{
+  networking.firewall.allowedTCPPorts = [ 3000 ];
+
+  services.grafana = {
+    enable = true;
+    dataDir = grafanaDataDir;
+    settings = {
+      database.type = "sqlite3";
+      security = {
+        admin_password = "$__file{${vars.secrets}/services/grafana/admin_password}";
+        admin_user = "admin";
+        secret_key = "$__file{${vars.secrets}/services/grafana/secret_key}";
+      };
+      server = {
+        http_addr = "192.168.90.40";
+        http_port = 3000;
+        root_url = "http://192.168.90.40:3000/";
+      };
+    };
+    provision = {
+      enable = true;
+      dashboards.settings = {
+        apiVersion = 1;
+        providers = [
+          {
+            name = "monitoring";
+            folder = "Monitoring";
+            type = "file";
+            disableDeletion = false;
+            editable = false;
+            allowUiUpdates = false;
+            updateIntervalSeconds = 30;
+            options.path = ../monitoring/dashboards;
+          }
+        ];
+      };
+      datasources.settings = {
+        apiVersion = 1;
+        prune = true;
+        datasources = [
+          {
+            access = "proxy";
+            editable = false;
+            isDefault = true;
+            name = "prom-main";
+            type = "prometheus";
+            uid = "prom-main";
+            url = "http://127.0.0.1:9090";
+          }
+          {
+            access = "proxy";
+            editable = false;
+            name = "prom-pid-short";
+            type = "prometheus";
+            uid = "prom-pid-short";
+            url = "http://127.0.0.1:9092";
+          }
+        ];
+      };
+    };
+  };
+
+  systemd = {
+    services.grafana.after = [
+      "prometheus-main.service"
+      "prometheus-pid-short.service"
+    ];
+
+    tmpfiles.rules = [
+      "d ${grafanaDataDir} 0750 grafana grafana - -"
+    ];
+  };
+}

systems/jeeves/services/hedgedoc.nix

@@ -1,24 +0,0 @@
-{
-  services.hedgedoc = {
-    enable = true;
-    settings = {
-      host = "0.0.0.0";
-      port = 3000;
-      domain = "192.168.90.40";
-      urlAddPort = true;
-      protocolUseSSL = false;
-      db = {
-        dialect = "postgres";
-        database = "hedgedoc";
-        username = "hedgedoc";
-        host = "/run/postgresql";
-      };
-    };
-  };
-  networking.firewall.allowedTCPPorts = [ 3000 ];
-
-  systemd.services.hedgedoc = {
-    after = [ "postgresql.service" ];
-    requires = [ "postgresql.service" ];
-  };
-}

systems/jeeves/services/nornsight.nix

@@ -0,0 +1,107 @@
+{ pkgs, ... }:
+let
+  vars = import ../vars.nix;
+  stateDir = "${vars.services}/nornsight";
+  appDir = "${stateDir}/app";
+  binPath = pkgs.lib.makeBinPath [
+    pkgs.binutils
+    pkgs.libpq
+    pkgs.postgresql
+    pkgs.stdenv.cc
+  ];
+  libraryPath = pkgs.lib.makeLibraryPath [
+    pkgs.libpq
+    pkgs.postgresql.lib
+  ];
+in
+{
+  systemd.tmpfiles.rules = [
+    "d ${stateDir} 0750 nornsight nornsight - -"
+  ];
+
+  users.users.nornsight = {
+    isSystemUser = true;
+    group = "nornsight";
+    home = stateDir;
+  };
+
+  systemd.services.nornsight = {
+    description = "Norn Sight";
+    after = [ "network-online.target" ];
+    wants = [ "network-online.target" ];
+    wantedBy = [ "multi-user.target" ];
+
+    environment = {
+      HOME = stateDir;
+      UV_CACHE_DIR = "${stateDir}/.cache/uv";
+      UV_PROJECT_ENVIRONMENT = "${appDir}/.venv";
+      UV_PYTHON = "${pkgs.python313}/bin/python3.13";
+      UV_PYTHON_DOWNLOADS = "never";
+      LD_LIBRARY_PATH = libraryPath;
+      LIBRARY_PATH = libraryPath;
+      PSYCOPG_IMPL = "python";
+    };
+
+    path = with pkgs; [
+      bash
+      coreutils
+      git
+      uv
+    ];
+
+    serviceConfig = {
+      Type = "simple";
+      User = "nornsight";
+      Group = "nornsight";
+      EnvironmentFile = "-${vars.secrets}/services/nornsight";
+      WorkingDirectory = stateDir;
+      Restart = "on-failure";
+      RestartSec = "5s";
+      StandardOutput = "journal";
+      StandardError = "journal";
+      NoNewPrivileges = true;
+      PrivateTmp = true;
+      ProtectHome = true;
+      ProtectSystem = "strict";
+      ReadWritePaths = [ stateDir ];
+    };
+
+    script = ''
+      set -eu
+      export PATH="${binPath}:$PATH"
+      export LD_LIBRARY_PATH="${libraryPath}:''${LD_LIBRARY_PATH:-}"
+      export LIBRARY_PATH="${libraryPath}:''${LIBRARY_PATH:-}"
+
+      : "''${NORN_SIGHT_REPO_URL:?NORN_SIGHT_REPO_URL is required}"
+      branch="''${NORN_SIGHT_BRANCH:-main}"
+
+      if [ -d "${appDir}/.git" ]; then
+        current_origin="$(git -C "${appDir}" remote get-url origin)"
+        if [ "$current_origin" != "$NORN_SIGHT_REPO_URL" ]; then
+          rm -rf "${appDir}"
+        fi
+      fi
+
+      if [ ! -d "${appDir}/.git" ]; then
+        git clone --branch "$branch" "$NORN_SIGHT_REPO_URL" "${appDir}"
+      else
+        cd "${appDir}"
+        git fetch origin "$branch"
+        git checkout "$branch"
+        git pull --ff-only origin "$branch"
+      fi
+
+      cd "${appDir}"
+      uv sync --upgrade
+      uv run python - <<'PY'
+      import ctypes.util
+      import os
+
+      print(f"LD_LIBRARY_PATH={os.environ.get('LD_LIBRARY_PATH')}")
+      print(f"LIBRARY_PATH={os.environ.get('LIBRARY_PATH')}")
+      print(f"libpq={ctypes.util.find_library('pq')}")
+      PY
+      exec uv run uvicorn pipelines.web.main:app --host 0.0.0.0 --port 8001
+    '';
+  };
+}

systems/jeeves/services/postgress.nix

@@ -12,6 +12,7 @@ in
   services.postgresql = {
     enable = true;
     package = pkgs.postgresql_17_jit;
+    extensions = ps: with ps; [ pgvector ];
     enableTCPIP = true;
     enableJIT = true;
     dataDir = "${vars.database}/postgres";

systems/jeeves/services/signal_bot.nix

@@ -1,57 +0,0 @@
-{
-  pkgs,
-  inputs,
-  ...
-}:
-let
-  vars = import ../vars.nix;
-in
-{
-  users = {
-    users.signalbot = {
-      isSystemUser = true;
-      group = "signalbot";
-    };
-    groups.signalbot = { };
-  };
-
-  systemd.services.signal-bot = {
-    description = "Signal command and control bot";
-    after = [
-      "network.target"
-      "podman-signal_cli_rest_api.service"
-    ];
-    wants = [ "podman-signal_cli_rest_api.service" ];
-    wantedBy = [ "multi-user.target" ];
-
-    environment = {
-      PYTHONPATH = "${inputs.self}";
-      SIGNALBOT_DB = "signalbot";
-      SIGNALBOT_USER = "signalbot";
-      SIGNALBOT_HOST = "/run/postgresql";
-      SIGNALBOT_PORT = "5432";
-    };
-
-    serviceConfig = {
-      Type = "simple";
-      WorkingDirectory = "${inputs.self}";
-      User = "signalbot";
-      Group = "signalbot";
-      EnvironmentFile = "${vars.secrets}/services/signal-bot";
-      ExecStart = "${pkgs.my_python}/bin/python -m python.signal_bot.main";
-      StateDirectory = "signal-bot";
-      Restart = "on-failure";
-      RestartSec = "10s";
-      StandardOutput = "journal";
-      StandardError = "journal";
-      NoNewPrivileges = true;
-      ProtectSystem = "strict";
-      ProtectHome = "read-only";
-      PrivateTmp = true;
-      ReadWritePaths = [ "/var/lib/signal-bot" ];
-      ReadOnlyPaths = [
-        "${inputs.self}"
-      ];
-    };
-  };
-}

systems/jeeves/services/validate_system.toml

@@ -1,7 +1,6 @@
 zpool = ["root_pool", "storage", "media"]
 services = [
     "audiobookshelf",
-    "cloud_flare_tunnel",
     "haproxy",
     "docker",
     "home-assistant",

systems/jeeves/syncthing.nix

@@ -10,6 +10,14 @@ in
     settings = {
       devices.davids-server.id = "7GXTDGR-AOXFW2O-K6J7NM3-XYZNRRW-AKHAFWM-GBOWUPQ-OA6JIWD-ER7RDQL"; # cspell:disable-line
       folders = {
+        photos = {
+          path = "${vars.syncthing}/important";
+          devices = [
+            "rhapsody-in-green"
+            "phone"
+          ];
+          fsWatcherEnabled = true;
+        };
         "dotfiles" = {
           path = "/home/richie/dotfiles";
           devices = [
@@ -89,7 +97,16 @@ in
           ];
           fsWatcherEnabled = true;
         };
-        #
+        "recordings" = {
+          path = "/home/richie/recordings";
+          devices = [
+            "bob"
+            "phone"
+            "rhapsody-in-green"
+          ];
+          fsWatcherEnabled = true;
+        };
+        # davids-server
         "davids-backup1" = {
           id = "8229p-8z3tm"; # cspell:disable-line
           path = "${vars.syncthing}/davids_backups/1";

systems/jeeves/web_services/acme.nix

@@ -0,0 +1,74 @@
+let
+  domains = [
+    "audiobookshelf"
+    "cache"
+    "gitea"
+    "jellyfin"
+    "share"
+  ];
+  extraDomains = [ "www.norn-sight.com" ];
+
+  makeCert = name: {
+    name = "${name}.tmmworkshop.com";
+    value = {
+      webroot = "/var/lib/acme/.challenges";
+      group = "acme";
+      reloadServices = [ "haproxy.service" ];
+    };
+  };
+
+  makeExtraCert = name: {
+    inherit name;
+    value = {
+      webroot = "/var/lib/acme/.challenges";
+      group = "acme";
+      reloadServices = [ "haproxy.service" ];
+    };
+  };
+
+  acmeServices =
+    map (domain: "acme-${domain}.tmmworkshop.com.service") domains
+    ++ map (domain: "acme-${domain}.service") extraDomains;
+in
+{
+  users.users.haproxy.extraGroups = [ "acme" ];
+
+  security.acme = {
+    acceptTerms = true;
+    defaults.email = "Richie@tmmworkshop.com";
+    certs = builtins.listToAttrs ((map makeCert domains) ++ (map makeExtraCert extraDomains));
+  };
+
+  # Minimal nginx to serve ACME HTTP-01 challenge files for HAProxy
+  services.nginx = {
+    enable = true;
+    virtualHosts."acme-challenge" = {
+      listen = [
+        {
+          addr = "127.0.0.1";
+          port = 8402;
+        }
+      ];
+      locations."/.well-known/acme-challenge/" = {
+        root = "/var/lib/acme/.challenges";
+      };
+    };
+  };
+
+  # Ensure the challenge directory exists with correct permissions
+  systemd.tmpfiles.rules = [
+    "d /var/lib/acme/.challenges 0750 acme acme - -"
+    "d /var/lib/acme/.challenges/.well-known 0750 acme acme - -"
+    "d /var/lib/acme/.challenges/.well-known/acme-challenge 0750 acme acme - -"
+  ];
+
+  users.users.nginx.extraGroups = [ "acme" ];
+
+  # HAProxy needs certs to exist before it can bind :443.
+  # NixOS's acme module generates self-signed placeholders on first boot
+  # via acme-<domain>.service — just make HAProxy wait for them.
+  systemd.services.haproxy = {
+    after = acmeServices;
+    wants = acmeServices;
+  };
+}

systems/jeeves/web_services/default.nix

@@ -0,0 +1,9 @@
+{ lib, ... }:
+{
+  imports =
+    let
+      files = builtins.attrNames (builtins.readDir ./.);
+      nixFiles = builtins.filter (name: lib.hasSuffix ".nix" name && name != "default.nix") files;
+    in
+    map (file: ./. + "/${file}") nixFiles;
+}

systems/jeeves/web_services/haproxy.cfg

@@ -6,6 +6,7 @@ global
 defaults
   log global
   mode http
+  option httplog
   retries 3
   maxconn 2000
   timeout connect 5s
@@ -22,24 +23,38 @@ defaults
 #Application Setup
 frontend ContentSwitching
   bind *:80 v4v6
-  bind *:443 v4v6 ssl crt /zfs/storage/secrets/docker/cloudflare.pem
+  bind *:443 v4v6 ssl crt /var/lib/acme/audiobookshelf.tmmworkshop.com/full.pem crt /var/lib/acme/cache.tmmworkshop.com/full.pem crt /var/lib/acme/jellyfin.tmmworkshop.com/full.pem crt /var/lib/acme/share.tmmworkshop.com/full.pem crt /var/lib/acme/gitea.tmmworkshop.com/full.pem crt /var/lib/acme/www.norn-sight.com/full.pem
   mode  http
+
+  # ACME challenge routing (must be first)
+  acl is_acme path_beg /.well-known/acme-challenge/
+
   # tmmworkshop.com
   acl host_audiobookshelf  hdr(host) -i audiobookshelf.tmmworkshop.com
   acl host_cache  hdr(host) -i cache.tmmworkshop.com
   acl host_jellyfin  hdr(host) -i jellyfin.tmmworkshop.com
   acl host_share  hdr(host) -i share.tmmworkshop.com
-  acl host_gcw  hdr(host) -i gcw.tmmworkshop.com
-  acl host_n8n  hdr(host) -i n8n.tmmworkshop.com
   acl host_gitea  hdr(host) -i gitea.tmmworkshop.com
+  acl host_norn_sight  hdr(host) -i www.norn-sight.com
 
+  # Hosts allowed to serve plain HTTP (add entries to skip the HTTPS redirect)
+  acl allow_http hdr(host) -i __none__
+  # acl allow_http hdr(host) -i example.tmmworkshop.com
+
+  # Redirect all HTTP to HTTPS unless on the allow list or ACME challenge
+  http-request redirect scheme https code 301 if !{ ssl_fc } !allow_http !is_acme
+
+  use_backend acme_challenge if is_acme
   use_backend audiobookshelf_nodes if host_audiobookshelf
   use_backend cache_nodes  if host_cache
   use_backend jellyfin if host_jellyfin
   use_backend share_nodes  if host_share
-  use_backend gcw_nodes  if host_gcw
-  use_backend n8n  if host_n8n
   use_backend gitea  if host_gitea
+  use_backend norn_sight  if host_norn_sight
+
+backend acme_challenge
+  mode http
+  server acme 127.0.0.1:8402
 
 backend audiobookshelf_nodes
   mode http
@@ -60,14 +75,10 @@ backend share_nodes
   mode http
   server server 127.0.0.1:8091
 
-backend gcw_nodes
-  mode http
-  server server 127.0.0.1:8092
-
-backend n8n
-  mode http
-  server server 127.0.0.1:5678
-
 backend gitea
   mode http
   server server 127.0.0.1:6443
+
+backend norn_sight
+  mode http
+  server server 127.0.0.1:8001

systems/rhapsody-in-green/agent_logger.nix

@@ -0,0 +1,35 @@
+{
+  pkgs,
+  inputs,
+  ...
+}:
+{
+  systemd.services.agent-logger = {
+    description = "Unified agent logger";
+    after = [ "local-fs.target" ];
+    wantedBy = [ "multi-user.target" ];
+
+    environment = {
+      AGENT_LOG_DB = "/var/lib/agent-logger/agent_log.sqlite";
+      HOME = "/home/richie";
+      PYTHONPATH = "${inputs.self}";
+    };
+
+    serviceConfig = {
+      Type = "simple";
+      User = "richie";
+      WorkingDirectory = "/home/richie";
+      ExecStart = "${pkgs.my_python}/bin/python -m python.agent_logger.main";
+      StateDirectory = "agent-logger";
+      Restart = "on-failure";
+      RestartSec = "5s";
+      StandardOutput = "journal";
+      StandardError = "journal";
+      NoNewPrivileges = true;
+      ProtectSystem = "strict";
+      ProtectHome = "read-only";
+      PrivateTmp = true;
+      ReadOnlyPaths = [ "${inputs.self}" ];
+    };
+  };
+}

systems/rhapsody-in-green/default.nix

@@ -11,10 +11,9 @@
     "${inputs.self}/common/optional/yubikey.nix"
     "${inputs.self}/common/optional/zerotier.nix"
     ./hardware.nix
-    ./llms.nix
     ./open_webui.nix
+    ./programs.nix
     ./qmk.nix
-    ./sunshine.nix
     ./syncthing.nix
     inputs.nixos-hardware.nixosModules.framework-13-7040-amd
   ];
@@ -24,11 +23,20 @@
     hostId = "6404140d";
     firewall = {
       enable = true;
-      allowedTCPPorts = [ ];
+      allowedTCPPorts = [
+        8000
+        8080
+        8081
+      ];
     };
     networkmanager.enable = true;
   };
 
+  programs.appimage = {
+    enable = true;
+    binfmt = true; # allows *.AppImage to be run directly
+  };
+
   services = {
     openssh.ports = [ 922 ];
     flatpak.enable = true;

systems/rhapsody-in-green/llms.nix

@@ -1,29 +0,0 @@
-{
-  services.ollama = {
-    user = "ollama";
-    enable = true;
-    host = "127.0.0.1";
-    syncModels = true;
-    loadModels = [
-      "deepscaler:1.5b"
-      "deepseek-r1:8b"
-      "gemma3:12b"
-      "lfm2:24b"
-      "nemotron-3-nano:4b"
-      "qwen3:14b"
-      "qwen3.5:27b"
-    ];
-  };
-  systemd.services = {
-    ollama.serviceConfig = {
-      Nice = 19;
-      IOSchedulingPriority = 7;
-    };
-    ollama-model-loader.serviceConfig = {
-      Nice = 19;
-      CPUWeight = 50;
-      IOSchedulingClass = "idle";
-      IOSchedulingPriority = 7;
-    };
-  };
-}

systems/rhapsody-in-green/open_webui.nix

@@ -1,6 +1,7 @@
 {
   services.open-webui = {
     enable = true;
+    host = "0.0.0.0";
     environment = {
       ANONYMIZED_TELEMETRY = "False";
       DO_NOT_TRACK = "True";

systems/rhapsody-in-green/programs.nix

@@ -0,0 +1,6 @@
+{ pkgs, ... }:
+{
+  environment.systemPackages = with pkgs; [
+    ffmpeg
+  ];
+}

systems/rhapsody-in-green/sunshine.nix

@@ -1,24 +0,0 @@
-{ pkgs, ... }:
-{
-  services.sunshine = {
-    enable = true;
-    openFirewall = true;
-    capSysAdmin = true;
-  };
-  environment.systemPackages = [ pkgs.kdePackages.libkscreen ];
-
-  boot.kernelParams = [
-    "drm.edid_firmware=DP-4:edid/virtual-display.bin"
-    "video=DP-4:e"
-  ];
-
-  hardware = {
-    firmwareCompression = "none";
-    firmware = [
-      (pkgs.runCommandLocal "virtual-display-edid" { } ''
-        mkdir -p $out/lib/firmware/edid
-        cp ${./edid/virtual-display.bin} $out/lib/firmware/edid/virtual-display.bin
-      '')
-    ];
-  };
-}

systems/rhapsody-in-green/syncthing.nix

@@ -39,6 +39,14 @@
       ];
       fsWatcherEnabled = true;
     };
+    photos = {
+      path = "/home/richie/photos";
+      devices = [
+        "jeeves"
+        "phone"
+      ];
+      fsWatcherEnabled = true;
+    };
     "projects" = {
       id = "vyma6-lqqrz"; # cspell:disable-line
       path = "/home/richie/projects";
@@ -55,6 +63,15 @@
       ];
       fsWatcherEnabled = true;
     };
+    "recordings" = {
+      path = "/home/richie/recordings";
+      devices = [
+        "bob"
+        "jeeves"
+        "phone"
+      ];
+      fsWatcherEnabled = true;
+    };
     "vault" = {
       path = "/home/richie/vault";
       devices = [

tests/test_gitea_flake_lock.py

@@ -0,0 +1,113 @@
+"""Tests for Gitea flake.lock automation."""
+
+from __future__ import annotations
+
+from python.gitea import PullRequest
+from python.gitea_flake_lock import (
+    PR_CHECK_WORKFLOWS,
+    PR_LABELS,
+    dispatch_pull_request_checks,
+    ensure_flake_lock_pull_request,
+    find_flake_lock_pull_request,
+)
+
+
+def _pull_request(number=1, head_branch="automation/update-flake-lock"):
+    return PullRequest(
+        number=number,
+        title="Update flake.lock",
+        html_url=f"https://gitea.example.test/pulls/{number}",
+        labels=(),
+        head_branch=head_branch,
+        base_branch="main",
+    )
+
+
+class FakeGiteaClient:
+    def __init__(self, pull_requests=None):
+        self.pull_requests = pull_requests or []
+        self.dispatch_calls = []
+        self.list_calls = []
+        self.create_calls = []
+
+    def list_open_pull_requests(self, **kwargs):
+        self.list_calls.append(kwargs)
+        return self.pull_requests
+
+    def create_pull_request(self, **kwargs):
+        self.create_calls.append(kwargs)
+        return _pull_request()
+
+    def dispatch_workflow(self, **kwargs):
+        self.dispatch_calls.append(kwargs)
+
+
+def test_ensure_flake_lock_pull_request_finds_by_branch():
+    pull_request = _pull_request()
+    client = FakeGiteaClient([pull_request])
+
+    result = ensure_flake_lock_pull_request(
+        client,
+        owner="Richie",
+        repo="dotfiles",
+        branch="automation/update-flake-lock",
+        base="main",
+    )
+
+    assert result == pull_request
+    assert client.list_calls == [
+        {"owner": "Richie", "repo": "dotfiles", "head": "automation/update-flake-lock"},
+    ]
+    assert client.create_calls == []
+
+
+def test_ensure_flake_lock_pull_request_creates_with_labels():
+    client = FakeGiteaClient()
+
+    ensure_flake_lock_pull_request(
+        client,
+        owner="Richie",
+        repo="dotfiles",
+        branch="automation/update-flake-lock",
+        base="main",
+    )
+
+    assert client.create_calls == [
+        {
+            "owner": "Richie",
+            "repo": "dotfiles",
+            "title": "Update flake.lock",
+            "body": "Automated flake.lock update.",
+            "head": "automation/update-flake-lock",
+            "base": "main",
+            "labels": PR_LABELS,
+        },
+    ]
+
+
+def test_find_flake_lock_pull_request_finds_by_label():
+    pull_request = _pull_request()
+    client = FakeGiteaClient([pull_request])
+
+    result = find_flake_lock_pull_request(client, owner="Richie", repo="dotfiles")
+
+    assert result == pull_request
+    assert client.list_calls == [
+        {"owner": "Richie", "repo": "dotfiles", "labels": ["flake_lock_update"]},
+    ]
+
+
+def test_dispatch_pull_request_checks_runs_each_workflow():
+    client = FakeGiteaClient()
+
+    dispatch_pull_request_checks(client, owner="Richie", repo="dotfiles", branch="automation/update-flake-lock")
+
+    assert client.dispatch_calls == [
+        {
+            "owner": "Richie",
+            "repo": "dotfiles",
+            "workflow_id": workflow,
+            "ref": "automation/update-flake-lock",
+        }
+        for workflow in PR_CHECK_WORKFLOWS
+    ]

users/dov/home/cli/git.nix

@@ -1,6 +1,7 @@
 {
   programs.git = {
     enable = true;
+    signing.format = null;
     settings = {
       user = {
         email = "dov.kruger@gmail.com";

users/elise/home/cli/git.nix

@@ -1,6 +1,7 @@
 {
   programs.git = {
     enable = true;
+    signing.format = null;
     settings = {
       user = {
         email = "DumbPuppy208@gmail.com";

users/math/default.nix

@@ -36,6 +36,8 @@ in
         "hass"
         "libvirtd"
         "networkmanager"
+        "nornsight"
+        "nornsight-admin"
         "plugdev"
         "scanner"
         "transmission"

users/math/home/cli/git.nix

@@ -1,6 +1,7 @@
 {
   programs.git = {
     enable = true;
+    signing.format = null;
     settings = {
       user = {
         email = "matthew.michal11@gmail.com";

users/richie/default.nix

@@ -36,6 +36,8 @@ in
         "hass"
         "libvirtd"
         "networkmanager"
+        "nornsight"
+        "nornsight-admin"
         "ollama"
         "plugdev"
         "scanner"

users/richie/home/cli/git.nix

@@ -1,6 +1,7 @@
 {
   programs.git = {
     enable = true;
+    signing.format = null;
     settings = {
       user = {
         email = "Richie@tmmworkshop.com";

users/richie/home/gui/default.nix

@@ -6,6 +6,7 @@
     "${inputs.self}/users/shared/sweet.nix"
     ./firefox
     ./kitty.nix
+    ./llm_tools.nix
     ./vscode
   ];
 
@@ -19,12 +20,11 @@
     qalculate-gtk
     vlc
     # browser
+    brave
     chromium
     # dev tools
-    claude-code
     gparted
     jetbrains.datagrip
     proxychains
-    opencode
   ];
 }

users/richie/home/gui/firefox/default.nix

@@ -1,8 +1,9 @@
-{ inputs, ... }:
+{ config, inputs, ... }:
 {
   imports = [ ./search_engines.nix ];
 
   programs.firefox = {
+    configPath = "${config.xdg.configHome}/mozilla/firefox";
     enable = true;
     profiles.richie = {
       extensions.packages = with inputs.firefox-addons.packages.x86_64-linux; [

users/richie/home/gui/kitty.nix

@@ -6,11 +6,13 @@
     settings = {
       allow_remote_control = "yes";
       shell = "${pkgs.zsh}/bin/zsh";
+      scrollback_lines = 50000;
       wayland_titlebar_color = "background";
       background_opacity = "0.75";
       tab_bar_edge = "top";
       tab_bar_style = "powerline";
       enabled_layouts = "splits";
+      enable_audio_bell = "no";
     };
     keybindings = {
       "ctrl+alt+1" = "launch --type=tab --tab-title jeeves kitten ssh jeeves";

users/richie/home/gui/llm_tools.nix

@@ -0,0 +1,9 @@
+{ pkgs, ... }:
+{
+  home.packages = [
+    pkgs.master.claude-code
+    pkgs.master.codex
+    pkgs.master.opencode
+    pkgs.master.pi-coding-agent
+  ];
+}

users/richie/home/gui/vscode/keybindings.json

@@ -2,28 +2,32 @@
   {
     "key": "shift+alt+f",
     "command": "editor.action.formatDocument",
-    "when": "editorHasDocumentFormattingProvider && editorTextFocus && !editorReadonly && !inCompositeEditor"
+    "when": "editorHasDocumentFormattingProvider && editorTextFocus && !editorReadonly && !inCompositeEditor",
   },
   {
     "key": "alt+a d",
-    "command": "cSpell.addWordToWorkspaceSettings"
+    "command": "cSpell.addWordToWorkspaceSettings",
   },
   {
     "key": "ctrl+shift+`",
-    "command": "workbench.action.createTerminalEditor"
+    "command": "workbench.action.createTerminalEditor",
   },
   {
     "key": "ctrl+shift+`",
     "command": "-workbench.action.terminal.new",
-    "when": "terminalProcessSupported || terminalWebExtensionContributedProfile"
+    "when": "terminalProcessSupported || terminalWebExtensionContributedProfile",
   },
   {
     "key": "ctrl+shift+g r",
-    "command": "gitlens.git.rebase"
+    "command": "gitlens.git.rebase",
   },
   {
     "key": "ctrl+shift+g c",
     "command": "-gitlens.showQuickCommitFileDetails",
-    "when": "editorTextFocus && !gitlens:disabled && config.gitlens.keymap == 'chorded'"
+    "when": "editorTextFocus && !gitlens:disabled && config.gitlens.keymap == 'chorded'",
-  }
+  },
+  {
+    "key": "ctrl+shift+g p",
+    "command": "gitlens.pushRepositories",
+  },
 ]

users/richie/home/gui/vscode/settings.json

@@ -14,7 +14,7 @@
   "git.pruneOnFetch": true,
   "terminal.integrated.scrollback": 10000,
   "update.mode": "none",
-  "workbench.colorTheme": "Default Dark+",
+  "workbench.colorTheme": "Dark+",
   "workbench.secondarySideBar.showLabels": false,
 
   // turns off all sounds and announcements
@@ -78,7 +78,8 @@
     "Corvidae",
     "drivername",
     "fastapi",
-    "Qwen",
+    "Michal",
+    "Nornsight",
     "sandboxing",
     "syncthing",
   ],

users/richie/home/ssh_config.nix

@@ -2,46 +2,46 @@
   programs.ssh = {
     enable = true;
     enableDefaultConfig = false;
-    matchBlocks = {
+    settings = {
       jeeves = {
-        hostname = "192.168.90.40";
+        HostName = "192.168.90.40";
-        user = "richie";
+        User = "richie";
-        identityFile = "~/.ssh/id_ed25519";
+        IdentityFile = "~/.ssh/id_ed25519";
-        port = 629;
+        Port = 629;
-        dynamicForwards = [ { port = 9050; } ];
+        DynamicForward = [ { port = 9050; } ];
-        compression = true;
+        Compression = true;
       };
       unlock-jeeves = {
-        hostname = "192.168.99.14";
+        HostName = "192.168.99.14";
-        user = "root";
+        User = "root";
-        identityFile = "~/.ssh/id_ed25519";
+        IdentityFile = "~/.ssh/id_ed25519";
-        port = 2222;
+        Port = 2222;
       };
       brain = {
-        hostname = "192.168.90.35";
+        HostName = "192.168.90.35";
-        user = "richie";
+        User = "richie";
-        identityFile = "~/.ssh/id_ed25519";
+        IdentityFile = "~/.ssh/id_ed25519";
-        port = 129;
+        Port = 129;
-        dynamicForwards = [ { port = 9050; } ];
+        DynamicForward = [ { port = 9050; } ];
       };
       unlock-brain = {
-        hostname = "192.168.95.35";
+        HostName = "192.168.95.35";
-        user = "root";
+        User = "root";
-        identityFile = "~/.ssh/id_ed25519";
+        IdentityFile = "~/.ssh/id_ed25519";
-        port = 2222;
+        Port = 2222;
       };
       bob = {
-        hostname = "192.168.90.25";
+        HostName = "192.168.90.25";
-        user = "richie";
+        User = "richie";
-        identityFile = "~/.ssh/id_ed25519";
+        IdentityFile = "~/.ssh/id_ed25519";
-        port = 262;
+        Port = 262;
-        dynamicForwards = [ { port = 9050; } ];
+        DynamicForward = [ { port = 9050; } ];
       };
       rhapsody-in-green = {
-        hostname = "192.168.90.221";
+        HostName = "192.168.90.221";
-        user = "richie";
+        User = "richie";
-        identityFile = "~/.ssh/id_ed25519";
+        IdentityFile = "~/.ssh/id_ed25519";
-        port = 922;
+        Port = 922;
       };
     };
   };

users/steve/default.nix

@@ -0,0 +1,44 @@
+{
+  pkgs,
+  config,
+  ...
+}:
+let
+  ifTheyExist = groups: builtins.filter (group: builtins.hasAttr group config.users.groups) groups;
+in
+{
+
+  users = {
+    users.steve = {
+      isNormalUser = true;
+
+      shell = pkgs.zsh;
+      group = "steve";
+      openssh.authorizedKeys.keys = [
+        "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIJH03VzDbUhzfhvwD+OsYh6GobODYaI9jdNdzWQoqFsp matth@Jove" # cspell:disable-line
+      ];
+      extraGroups = [
+        "audio"
+        "video"
+        "wheel"
+        "users"
+      ]
+      ++ ifTheyExist [
+        "dialout"
+        "docker"
+        "hass"
+        "libvirtd"
+        "networkmanager"
+        "plugdev"
+        "scanner"
+        "transmission"
+        "uaccess"
+        "wireshark"
+      ];
+      uid = 1005;
+    };
+
+    groups.steve.gid = 1005;
+  };
+  home-manager.users.steve = import ./systems/${config.networking.hostName}.nix;
+}

users/steve/home/cli/default.nix

@@ -0,0 +1,9 @@
+{
+  imports = [
+    ./direnv.nix
+    ./git.nix
+    ./zsh.nix
+  ];
+
+  programs.starship.enable = true;
+}

users/steve/home/cli/direnv.nix

@@ -0,0 +1,8 @@
+{
+
+  programs.direnv = {
+    enable = true;
+    enableZshIntegration = true;
+    nix-direnv.enable = true;
+  };
+}

users/steve/home/cli/git.nix

@@ -0,0 +1,15 @@
+{
+  programs.git = {
+    enable = true;
+    signing.format = null;
+    settings = {
+      user = {
+        email = "matthew.michal11@gmail.com";
+        name = "Matthew Michal";
+      };
+      pull.rebase = true;
+      color.ui = true;
+    };
+    lfs.enable = true;
+  };
+}