preppergpt 0.1.0 → 0.1.1

package/README.md

@@ -11,22 +11,30 @@ working network during setup.
 
 ## Install
 
-Until the npm package is published, install from GitHub:
+Install from npm:
 
 ```bash
-git clone https://github.com/teamslop/preppergpt.git
-cd preppergpt
-node bin/preppergpt.js install --profile balanced
-node bin/preppergpt.js start
+npx preppergpt install --profile balanced
+preppergpt start
 ```
 
-After npm publication:
+Or install globally:
 
 ```bash
-npx preppergpt install --profile balanced
+npm install -g preppergpt
+preppergpt install --profile balanced
 preppergpt start
 ```
 
+GitHub source install:
+
+```bash
+git clone https://github.com/teamslop/preppergpt.git
+cd preppergpt
+node bin/preppergpt.js install --profile balanced
+node bin/preppergpt.js start
+```
+
 Other profiles:
 
 ```bash
@@ -54,6 +62,7 @@ preppergpt stop
 preppergpt status
 preppergpt doctor
 preppergpt switch-profile --profile speed
+preppergpt bundle whisper
 ```
 
 ## Profiles
@@ -70,8 +79,13 @@ and route ordering into the generated compose override.
 
 ## Model Assets
 
-Some routes can be pulled by the runtime, while very large routes such as GLM
-5.2 Q4 and Flux weights are marked as manual or external in
+PrepperGPT installs a bundled local Whisper Base STT cache during
+`preppergpt install`. It is stored under `~/.preppergpt/data/models/whisper/base`
+by default and mounted into OpenWebUI, so speech-to-text works from local files
+after setup.
+
+Some other routes can be pulled by the runtime, while very large routes such as
+GLM 5.2 Q4 and Flux weights are marked as manual or external in
 `profiles/models.json`. `preppergpt doctor` reports which selected routes still
 need local files or endpoints.
 
@@ -84,7 +98,7 @@ npm publish --access public
 ```
 
 Publishing requires an authenticated npm account with permission to publish the
-currently unclaimed `preppergpt` package name.
+`preppergpt` package.
 
 The source repository is expected at:
 

package/compose/preppergpt.yaml

@@ -13,6 +13,7 @@ services:
         hard: 1048576
     volumes:
       - ${PREPPERGPT_DATA_DIR:?set PREPPERGPT_DATA_DIR}/openwebui:/app/backend/data
+      - ${PREPPERGPT_MODELS_DIR:?set PREPPERGPT_MODELS_DIR}:/models:ro
       - ../services/comfyui:/app/backend/data/parity-comfyui:ro
       - ../themes/preppergpt/static/favicon.svg:/app/backend/open_webui/static/favicon.svg:ro
       - ../themes/preppergpt/static/logo.svg:/app/backend/open_webui/static/logo.svg:ro
@@ -69,9 +70,10 @@ services:
       CODE_INTERPRETER_JUPYTER_URL: "http://127.0.0.1:8888"
       CODE_INTERPRETER_JUPYTER_AUTH: "token"
       CODE_INTERPRETER_JUPYTER_AUTH_TOKEN: "${JUPYTER_TOKEN:?set JUPYTER_TOKEN}"
-      WHISPER_MODEL: "large-v3"
+      WHISPER_MODEL: "${PREPPERGPT_WHISPER_MODEL_PATH:-/models/whisper/base}"
+      WHISPER_MODEL_DIR: "/models/whisper"
       WHISPER_COMPUTE_TYPE: "int8"
-      WHISPER_MODEL_AUTO_UPDATE: "True"
+      WHISPER_MODEL_AUTO_UPDATE: "False"
       WHISPER_VAD_FILTER: "True"
       WHISPER_MULTILINGUAL: "True"
       ENABLE_IMAGE_GENERATION: "True"

package/docs/bundles.md

@@ -0,0 +1,30 @@
+# Bundles
+
+PrepperGPT keeps npm lightweight but installs small always-on local assets
+during setup.
+
+## Whisper Base
+
+`preppergpt install` downloads the MIT-licensed `Systran/faster-whisper-base`
+CTranslate2 model into:
+
+```text
+~/.preppergpt/data/models/whisper/base
+```
+
+OpenWebUI receives:
+
+```text
+WHISPER_MODEL=/models/whisper/base
+WHISPER_MODEL_DIR=/models/whisper
+WHISPER_MODEL_AUTO_UPDATE=False
+```
+
+To repair or refresh the bundle:
+
+```bash
+preppergpt bundle whisper
+preppergpt bundle whisper --force
+```
+
+Source: https://huggingface.co/Systran/faster-whisper-base

package/docs/model-sources.md

@@ -3,7 +3,8 @@
 PrepperGPT separates routing from model licensing and distribution.
 
 - Ollama models are pulled by the local Ollama runtime when available.
-- OpenWebUI STT models are downloaded by OpenWebUI/faster-whisper.
+- Whisper Base STT is installer-cached from `Systran/faster-whisper-base`
+  under the local PrepperGPT model directory and mounted into OpenWebUI.
 - Hugging Face vision models are downloaded by the local vision sidecar.
 - Very large GLM, Slopcode, and Flux assets are marked as manual or external
   until a license-compatible public download source is configured.

package/installer/cli.mjs

@@ -1,12 +1,13 @@
 import fs from "node:fs";
 import http from "node:http";
+import { ensureWhisperBundle, modelDirs, whisperBundleStatus } from "./lib/bundles.mjs";
 import { detectMachine } from "./lib/detect.mjs";
 import { buildPlan, normalizeProfile } from "./lib/planner.mjs";
 import { packageRoot, runtimePaths } from "./lib/paths.mjs";
 import { renderInstall } from "./lib/render.mjs";
 import { commandResult, parseArgs, readJson, shellQuote } from "./lib/util.mjs";
 
-const VERSION = "0.1.0";
+const VERSION = "0.1.1";
 
 function usage() {
   return `PrepperGPT ${VERSION}
@@ -14,11 +15,12 @@ function usage() {
 Usage:
   preppergpt detect [--json]
   preppergpt plan --profile balanced|intelligence|speed [--json]
-  preppergpt install --profile balanced|intelligence|speed [--dry-run] [--home PATH]
+  preppergpt install --profile balanced|intelligence|speed [--dry-run] [--skip-bundles] [--home PATH]
   preppergpt start [--home PATH]
   preppergpt stop [--home PATH]
   preppergpt status [--home PATH] [--json]
   preppergpt doctor [--profile balanced|intelligence|speed] [--home PATH]
+  preppergpt bundle whisper [--home PATH] [--force]
   preppergpt switch-profile --profile balanced|intelligence|speed [--home PATH]
   preppergpt version
 `;
@@ -122,6 +124,11 @@ async function commandInstall(flags) {
     return;
   }
   const paths = renderInstall(plan, detection, { home });
+  if (!flags.skip_bundles) {
+    console.log("Installing bundled Whisper base STT model...");
+    const bundle = await ensureWhisperBundle(paths.whisperHostDir, { force: Boolean(flags.force_bundle) });
+    console.log(`Whisper bundle: ${bundle.ready ? "ready" : "not ready"} at ${paths.whisperHostDir}`);
+  }
   console.log(`Wrote ${paths.envFile}`);
   console.log(`Wrote ${paths.generatedCompose}`);
   console.log(`Wrote ${paths.modelPlan}`);
@@ -187,6 +194,7 @@ async function commandStatus(flags) {
 }
 
 async function commandDoctor(flags) {
+  const paths = runtimePaths(flags.home);
   const detection = await detectMachine();
   const plan = buildPlan(detection, profileFrom(flags));
   printPlan(plan);
@@ -200,6 +208,28 @@ async function commandDoctor(flags) {
       console.log(`  port ${port}: occupied`);
     }
   }
+  const dirs = modelDirs(paths);
+  const whisper = whisperBundleStatus(dirs.whisperHostDir);
+  console.log(`  whisper-base bundle: ${whisper.ready ? "ok" : `missing ${whisper.missing.length} files`} (${dirs.whisperHostDir})`);
+}
+
+async function commandBundle(flags, positional) {
+  const name = positional[1] || "whisper";
+  if (!["whisper", "whisper-base"].includes(name)) {
+    throw new Error(`Unknown bundle: ${name}`);
+  }
+  const paths = runtimePaths(flags.home);
+  const dirs = modelDirs(paths);
+  const bundle = await ensureWhisperBundle(dirs.whisperHostDir, {
+    force: Boolean(flags.force),
+    dryRun: Boolean(flags.dry_run)
+  });
+  console.log(`Whisper bundle ${bundle.ready ? "ready" : "not ready"} at ${dirs.whisperHostDir}`);
+  if (bundle.missing?.length) {
+    for (const file of bundle.missing) {
+      console.log(`  missing ${file}`);
+    }
+  }
 }
 
 export async function runCli(argv) {
@@ -220,6 +250,7 @@ export async function runCli(argv) {
   if (command === "stop") return commandStop(flags);
   if (command === "status") return commandStatus(flags);
   if (command === "doctor") return commandDoctor(flags);
+  if (command === "bundle") return commandBundle(flags, positional);
   if (command === "switch-profile") return commandSwitchProfile(flags);
   throw new Error(`Unknown command: ${command}\n\n${usage()}`);
 }

package/installer/lib/bundles.mjs

@@ -0,0 +1,107 @@
+import fs from "node:fs";
+import path from "node:path";
+import { Readable } from "node:stream";
+import { pipeline } from "node:stream/promises";
+import { readJson, writeJson } from "./util.mjs";
+
+export const WHISPER_BUNDLE = {
+  id: "whisper-base",
+  name: "Whisper Base STT Bundle",
+  repo: "Systran/faster-whisper-base",
+  revision: "main",
+  license: "MIT",
+  modelPathInContainer: "/models/whisper/base",
+  files: ["config.json", "model.bin", "tokenizer.json", "vocabulary.txt", "README.md"],
+  description: "CTranslate2 faster-whisper conversion of openai/whisper-base for local OpenWebUI STT."
+};
+
+function parseEnvFile(file) {
+  if (!fs.existsSync(file)) {
+    return {};
+  }
+  const entries = {};
+  for (const line of fs.readFileSync(file, "utf8").split(/\r?\n/)) {
+    if (!line || line.trim().startsWith("#") || !line.includes("=")) {
+      continue;
+    }
+    const [key, ...valueParts] = line.split("=");
+    let value = valueParts.join("=");
+    if ((value.startsWith('"') && value.endsWith('"')) || (value.startsWith("'") && value.endsWith("'"))) {
+      value = value.slice(1, -1);
+    }
+    entries[key] = value;
+  }
+  return entries;
+}
+
+export function modelDirs(paths) {
+  const env = parseEnvFile(paths.envFile);
+  const modelsDir = process.env.PREPPERGPT_MODELS_DIR || env.PREPPERGPT_MODELS_DIR || path.join(paths.dataDir, "models");
+  const whisperHostDir = path.join(modelsDir, "whisper", "base");
+  return { modelsDir, whisperHostDir };
+}
+
+export function whisperBundleStatus(targetDir) {
+  const files = WHISPER_BUNDLE.files.map((file) => path.join(targetDir, file));
+  const missing = files.filter((file) => !fs.existsSync(file));
+  let manifest = null;
+  const manifestPath = path.join(targetDir, "preppergpt-bundle.json");
+  if (fs.existsSync(manifestPath)) {
+    try {
+      manifest = readJson(manifestPath);
+    } catch {
+      manifest = null;
+    }
+  }
+  return {
+    id: WHISPER_BUNDLE.id,
+    targetDir,
+    ready: missing.length === 0,
+    missing,
+    manifest
+  };
+}
+
+async function downloadFile(url, targetFile) {
+  const response = await fetch(url, {
+    headers: {
+      "User-Agent": "preppergpt/0.1"
+    },
+    redirect: "follow"
+  });
+  if (!response.ok || !response.body) {
+    throw new Error(`Failed to download ${url}: HTTP ${response.status}`);
+  }
+  fs.mkdirSync(path.dirname(targetFile), { recursive: true });
+  const tmp = `${targetFile}.tmp-${process.pid}`;
+  await pipeline(Readable.fromWeb(response.body), fs.createWriteStream(tmp));
+  fs.renameSync(tmp, targetFile);
+}
+
+export async function ensureWhisperBundle(targetDir, options = {}) {
+  const status = whisperBundleStatus(targetDir);
+  if (status.ready && !options.force) {
+    return { ...status, changed: false };
+  }
+  fs.mkdirSync(targetDir, { recursive: true });
+  if (options.dryRun) {
+    return { ...status, changed: false, dryRun: true };
+  }
+  for (const file of WHISPER_BUNDLE.files) {
+    const targetFile = path.join(targetDir, file);
+    if (fs.existsSync(targetFile) && !options.force) {
+      continue;
+    }
+    const url = `https://huggingface.co/${WHISPER_BUNDLE.repo}/resolve/${WHISPER_BUNDLE.revision}/${file}`;
+    if (!options.quiet) {
+      console.log(`Downloading ${WHISPER_BUNDLE.repo}/${file}`);
+    }
+    await downloadFile(url, targetFile);
+  }
+  writeJson(path.join(targetDir, "preppergpt-bundle.json"), {
+    ...WHISPER_BUNDLE,
+    installedAt: new Date().toISOString(),
+    source: `https://huggingface.co/${WHISPER_BUNDLE.repo}`
+  });
+  return { ...whisperBundleStatus(targetDir), changed: true };
+}

package/installer/lib/render.mjs

@@ -1,5 +1,7 @@
 import crypto from "node:crypto";
 import fs from "node:fs";
+import path from "node:path";
+import { WHISPER_BUNDLE } from "./bundles.mjs";
 import { packagedPath, runtimePaths } from "./paths.mjs";
 import { envQuote, writeJson, writeText } from "./util.mjs";
 
@@ -10,6 +12,7 @@ function secret(bytes = 24) {
 function envFile(plan, paths, detection) {
   const dataDir = process.env.PREPPERGPT_DATA_DIR || paths.dataDir;
   const modelsDir = process.env.PREPPERGPT_MODELS_DIR || `${dataDir}/models`;
+  const whisperHostDir = path.join(modelsDir, "whisper", "base");
   const adminPassword = process.env.PREPPERGPT_ADMIN_PASSWORD || secret(18);
   const jupyterToken = process.env.JUPYTER_TOKEN || secret(18);
   const searxngSecret = process.env.SEARXNG_SECRET_KEY || secret(24);
@@ -17,6 +20,9 @@ function envFile(plan, paths, detection) {
     PREPPERGPT_PROFILE: plan.profile,
     PREPPERGPT_DATA_DIR: dataDir,
     PREPPERGPT_MODELS_DIR: modelsDir,
+    PREPPERGPT_WHISPER_HOST_DIR: whisperHostDir,
+    PREPPERGPT_WHISPER_MODEL: WHISPER_BUNDLE.id,
+    PREPPERGPT_WHISPER_MODEL_PATH: WHISPER_BUNDLE.modelPathInContainer,
     PREPPERGPT_PORT: process.env.PREPPERGPT_PORT || "8080",
     PREPPERGPT_DEFAULT_MODEL: plan.defaultModel,
     PREPPERGPT_MODEL_ORDER_LIST: JSON.stringify(plan.routeIds),
@@ -60,17 +66,23 @@ function generatedCompose(plan, detection) {
 
 export function renderInstall(plan, detection, options = {}) {
   const paths = runtimePaths(options.home);
+  const dataDir = process.env.PREPPERGPT_DATA_DIR || paths.dataDir;
+  const modelsDir = process.env.PREPPERGPT_MODELS_DIR || `${dataDir}/models`;
+  const whisperHostDir = path.join(modelsDir, "whisper", "base");
   fs.mkdirSync(paths.root, { recursive: true });
   fs.mkdirSync(paths.dataDir, { recursive: true });
   fs.mkdirSync(paths.composeDir, { recursive: true });
   fs.mkdirSync(`${paths.dataDir}/preppergpt`, { recursive: true });
-  fs.mkdirSync(`${paths.dataDir}/models`, { recursive: true });
+  fs.mkdirSync(modelsDir, { recursive: true });
+  fs.mkdirSync(whisperHostDir, { recursive: true });
   writeText(paths.envFile, envFile(plan, paths, detection), 0o600);
   writeText(paths.generatedCompose, generatedCompose(plan, detection));
   writeJson(paths.modelPlan, plan);
   writeJson(paths.detectReport, detection);
   return {
     ...paths,
+    modelsDir,
+    whisperHostDir,
     packageCompose: packagedPath("compose", "preppergpt.yaml")
   };
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "preppergpt",
-  "version": "0.1.0",
+  "version": "0.1.1",
   "description": "A local-first ChatGPT-like field kit built on OpenWebUI and local models.",
   "type": "module",
   "bin": {
@@ -19,7 +19,7 @@
   ],
   "scripts": {
     "test": "node --test",
-    "check": "npm run test && node bin/preppergpt.js plan --profile balanced --json >/dev/null && node bin/preppergpt.js install --profile balanced --dry-run",
+    "check": "npm run test && node bin/preppergpt.js plan --profile balanced --json >/dev/null && node bin/preppergpt.js install --profile balanced --dry-run && node bin/preppergpt.js bundle whisper --dry-run >/dev/null",
     "pack:dry-run": "npm pack --dry-run"
   },
   "keywords": [

package/profiles/models.json

@@ -13,7 +13,7 @@
         "agent": ["local-agent-glm52"],
         "vision": ["local-vision-gemma4-12b", "local-vision-moondream2"],
         "image": ["flux-2-klein-9b-fp8"],
-        "stt": ["whisper-large-v3"]
+        "stt": ["whisper-base-bundled"]
       }
     },
     "balanced": {
@@ -28,7 +28,7 @@
         "agent": ["local-agent-glm52"],
         "vision": ["local-vision-gemma4-12b", "local-vision-moondream2"],
         "image": ["flux-2-klein-9b-fp8"],
-        "stt": ["whisper-large-v3"]
+        "stt": ["whisper-base-bundled"]
       }
     },
     "speed": {
@@ -43,7 +43,7 @@
         "agent": ["local-agent-glm52"],
         "vision": ["local-vision-moondream2", "local-vision-gemma4-12b"],
         "image": ["flux-2-klein-9b-fp8"],
-        "stt": ["whisper-large-v3"]
+        "stt": ["whisper-base-bundled"]
       }
     }
   },
@@ -256,21 +256,23 @@
       }
     },
     {
-      "id": "whisper-large-v3",
-      "name": "Whisper Large v3 STT",
+      "id": "whisper-base-bundled",
+      "name": "Bundled Whisper Base STT",
       "roles": ["stt"],
       "backend": "openwebui-faster-whisper",
       "contextTokens": 0,
-      "qualityScore": 82,
-      "speedScore": 65,
-      "tpsEstimate": "audio transcription speed depends on CPU/GPU and clip length",
+      "qualityScore": 72,
+      "speedScore": 82,
+      "tpsEstimate": "local speech-to-text speed depends on CPU/GPU and clip length",
       "requires": {
-        "minRamGb": 16,
-        "diskGb": 4
+        "minRamGb": 8,
+        "diskGb": 1
       },
       "source": {
-        "type": "openwebui",
-        "description": "OpenWebUI faster-whisper local STT model."
+        "type": "bundled-download",
+        "model": "Systran/faster-whisper-base",
+        "license": "MIT",
+        "description": "Installer-cached faster-whisper base model mounted into OpenWebUI for offline local STT after install."
       }
     }
   ]

package/services/local-scheduler/app.py

@@ -3792,7 +3792,7 @@ def local_parity_recommended_model(feature_family: str, primary_models: list[str
     if "image understanding" in family:
         return first_available(["local-vision-gemma4-12b", "local-vision-moondream2"])
     if "voice" in family or "record mode" in family:
-        return first_available(["whisper-large-v3", "local-agent-glm52"])
+        return first_available(["whisper-base-bundled", "whisper-large-v3", "local-agent-glm52"])
     if "shopping" in family:
         return first_available(["glm52-shopping-research-local", "glm52-q4-local"])
     if "job search" in family or "resume" in family or "finance" in family:
@@ -5317,7 +5317,7 @@ def local_parity_dashboard() -> dict:
             {"id": "local-agent-glm52", "route": "local_agent", "best_for": "tool and agent workflows"},
             {"id": "local-vision-gemma4-12b", "route": "local_vision", "best_for": "image understanding"},
             {"id": "flux-2-klein-9b-fp8", "route": "comfyui_flux", "best_for": "image generation"},
-            {"id": "whisper-large-v3", "route": "local_whisper_stt", "best_for": "speech-to-text"},
+            {"id": "whisper-base-bundled", "route": "local_whisper_stt", "best_for": "speech-to-text"},
         ],
         "route_profiles": {
             key: {