Improve Windows desktop integration

app/src/components/ServerTab/GeneralPage.tsx

@@ -1,5 +1,14 @@
 import { zodResolver } from '@hookform/resolvers/zod';
-import { AlertCircle, ArrowUpRight, Book, Download, Loader2, RefreshCw } from 'lucide-react';
+import {
+  AlertCircle,
+  ArrowUpRight,
+  Book,
+  Download,
+  Loader2,
+  Play,
+  RefreshCw,
+  Square,
+} from 'lucide-react';
 import { useEffect, useMemo, useState } from 'react';
 import { useForm } from 'react-hook-form';
 import { Trans, useTranslation } from 'react-i18next';
@@ -36,8 +45,11 @@ export function GeneralPage() {
   const setKeepServerRunningOnClose = useServerStore((state) => state.setKeepServerRunningOnClose);
   const mode = useServerStore((state) => state.mode);
   const setMode = useServerStore((state) => state.setMode);
+  const customModelsDir = useServerStore((state) => state.customModelsDir);
   const { toast } = useToast();
-  const { data: health, isLoading, error: healthError } = useServerHealth();
+  const { data: health, isLoading, error: healthError, refetch: refetchHealth } = useServerHealth();
+  const [serverAction, setServerAction] = useState<'start' | 'stop' | 'restart' | null>(null);
+  const serverOnline = Boolean(health) && !healthError;
 
   const resolver = useMemo(
     () => zodResolver(makeConnectionSchema(t('settings.general.serverUrl.invalidUrl'))),
@@ -70,6 +82,38 @@ export function GeneralPage() {
     });
   }
 
+  async function runServerAction(action: 'start' | 'stop' | 'restart') {
+    setServerAction(action);
+    try {
+      if (action === 'start') {
+        const url = await platform.lifecycle.startServer(false, customModelsDir);
+        setServerUrl(url);
+        window.__voiceboxServerStartedByApp = true;
+      } else if (action === 'stop') {
+        await platform.lifecycle.stopServer();
+        window.__voiceboxServerStartedByApp = false;
+      } else {
+        const url = await platform.lifecycle.restartServer(customModelsDir);
+        setServerUrl(url);
+        window.__voiceboxServerStartedByApp = true;
+      }
+
+      await refetchHealth();
+      toast({
+        title: t(`settings.general.serverControl.${action}Success`),
+      });
+    } catch (error) {
+      const message = error instanceof Error ? error.message : String(error);
+      toast({
+        title: t('settings.general.serverControl.failed'),
+        description: message,
+        variant: 'destructive',
+      });
+    } finally {
+      setServerAction(null);
+    }
+  }
+
   return (
     <div className="space-y-8 max-w-2xl">
       <div className="grid grid-cols-2 gap-3">
@@ -141,6 +185,58 @@ export function GeneralPage() {
           </Form>
         </SettingRow>
 
+        {platform.metadata.isTauri && mode === 'local' && (
+          <SettingRow
+            title={t('settings.general.serverControl.title')}
+            description={t('settings.general.serverControl.description')}
+            action={
+              <div className="flex items-center gap-2">
+                <Button
+                  type="button"
+                  size="sm"
+                  onClick={() => runServerAction('start')}
+                  disabled={serverAction !== null || serverOnline}
+                >
+                  {serverAction === 'start' ? (
+                    <Loader2 className="h-3.5 w-3.5 animate-spin" />
+                  ) : (
+                    <Play className="h-3.5 w-3.5" />
+                  )}
+                  {t('settings.general.serverControl.start')}
+                </Button>
+                <Button
+                  type="button"
+                  size="sm"
+                  variant="outline"
+                  onClick={() => runServerAction('restart')}
+                  disabled={serverAction !== null || !serverOnline}
+                >
+                  {serverAction === 'restart' ? (
+                    <Loader2 className="h-3.5 w-3.5 animate-spin" />
+                  ) : (
+                    <RefreshCw className="h-3.5 w-3.5" />
+                  )}
+                  {t('settings.general.serverControl.restart')}
+                </Button>
+                <Button
+                  type="button"
+                  size="sm"
+                  variant="destructive"
+                  onClick={() => runServerAction('stop')}
+                  disabled={serverAction !== null || !serverOnline}
+                >
+                  {serverAction === 'stop' ? (
+                    <Loader2 className="h-3.5 w-3.5 animate-spin" />
+                  ) : (
+                    <Square className="h-3.5 w-3.5" />
+                  )}
+                  {t('settings.general.serverControl.stop')}
+                </Button>
+              </div>
+            }
+          />
+        )}
+
         <SettingRow
           title={t('settings.general.keepServerRunning.title')}
           description={t('settings.general.keepServerRunning.description')}

app/src/i18n/locales/en/translation.json

@@ -774,6 +774,17 @@
         "updatedTitle": "Server URL updated",
         "updatedDescription": "Connected to {{url}}"
       },
+      "serverControl": {
+        "title": "Local backend",
+        "description": "Control the bundled Voicebox server running in the background.",
+        "start": "Start",
+        "stop": "Stop",
+        "restart": "Restart",
+        "startSuccess": "Server started",
+        "stopSuccess": "Server stopped",
+        "restartSuccess": "Server restarted",
+        "failed": "Server action failed"
+      },
       "keepServerRunning": {
         "title": "Keep server running when app closes",
         "description": "The server will continue running in the background after closing the app.",
@@ -1124,7 +1135,8 @@
         "title": "Switch to CPU backend",
         "description": "Disable GPU acceleration. You can re-download the GPU backend later.",
         "button": "Switch"
-      },      "remove": {
+      },
+      "remove": {
         "title": "Remove CUDA backend",
         "description": "Delete the downloaded CUDA binary to free disk space.",
         "button": "Remove"

backend/build_binary.py

@@ -6,14 +6,15 @@
     python build_binary.py --cuda    # Build CUDA-enabled server binary
 """
 
-import PyInstaller.__main__
 import argparse
 import logging
 import os
 import platform
 import sys
 from pathlib import Path
 
+import PyInstaller.__main__
+
 logger = logging.getLogger(__name__)
 
 
@@ -34,6 +35,18 @@ def build_server(cuda=False, rocm=False):
     if cuda and rocm:
         raise ValueError("Cannot build with both CUDA and ROCm support")
 
+    # PyInstaller resolves transitive DLLs using PATH. On Windows, third-party
+    # toolchains such as LLVM can put an older MSVCP140.dll ahead of the current
+    # system VC++ runtime, producing a binary that crashes during extraction.
+    # Prefer the Windows runtime directory while collecting dependencies.
+    if platform.system() == "Windows":
+        system_root = Path(os.environ.get("SYSTEMROOT", r"C:\Windows"))
+        system32 = str(system_root / "System32")
+        path_entries = os.environ.get("PATH", "").split(os.pathsep)
+        os.environ["PATH"] = os.pathsep.join(
+            [system32, *(entry for entry in path_entries if entry.lower() != system32.lower())]
+        )
+
     backend_dir = Path(__file__).parent
 
     if rocm:
@@ -665,7 +678,6 @@ def build_server(cuda=False, rocm=False):
                 check=True,
             )
 
-
     logger.info("Binary built in %s", backend_dir / "dist" / binary_name)
 
 
@@ -781,4 +793,3 @@ def build_shim():
         build_shim()
     else:
         build_server(cuda=cli_args.cuda, rocm=cli_args.rocm)
-

justfile

@@ -228,10 +228,14 @@ build-server: _ensure-venv
     $env:PATH = "{{ venv_bin }};$env:PATH"; \
     & "{{ python }}" backend/build_binary.py; \
     if ($LASTEXITCODE -ne 0) { throw "build_binary.py failed with exit code $LASTEXITCODE" }; \
+    & "{{ python }}" backend/build_binary.py --shim; \
+    if ($LASTEXITCODE -ne 0) { throw "build_binary.py --shim failed with exit code $LASTEXITCODE" }; \
     $triple = (rustc --print host-tuple); \
     New-Item -ItemType Directory -Path "{{ tauri_dir }}/src-tauri/binaries" -Force | Out-Null; \
     Copy-Item "backend/dist/voicebox-server.exe" "{{ tauri_dir }}/src-tauri/binaries/voicebox-server-$triple.exe" -Force; \
-    Write-Host "Copied sidecar: voicebox-server-$triple.exe"
+    Copy-Item "backend/dist/voicebox-mcp.exe" "{{ tauri_dir }}/src-tauri/binaries/voicebox-mcp-$triple.exe" -Force; \
+    Write-Host "Copied sidecar: voicebox-server-$triple.exe"; \
+    Write-Host "Copied sidecar: voicebox-mcp-$triple.exe"
 
 # Build CUDA server binary and place in app data dir for local testing
 [windows]

tauri/src-tauri/Cargo.toml

@@ -13,7 +13,7 @@ edition = "2021"
 tauri-build = { version = "2.0", features = [] }
 
 [dependencies]
-tauri = { version = "2.0", features = ["macos-private-api"] }
+tauri = { version = "2.0", features = ["macos-private-api", "tray-icon"] }
 tauri-plugin-dialog = "2.0"
 tauri-plugin-fs = "2.0"
 tauri-plugin-shell = "2.0"

tauri/src-tauri/src/main.rs

@@ -17,9 +17,17 @@ mod synthetic_keys;
 
 use std::sync::Mutex;
 use tauri::{command, State, Manager, WindowEvent, Emitter, Listener, RunEvent, WebviewUrl, WebviewWindowBuilder, PhysicalPosition};
+#[cfg(desktop)]
+use tauri::{
+    menu::{Menu, MenuItem},
+    tray::{MouseButton, MouseButtonState, TrayIconBuilder, TrayIconEvent},
+};
 use tauri_plugin_shell::ShellExt;
 use tokio::sync::mpsc;
 
+static TRAY_QUIT_REQUESTED: std::sync::atomic::AtomicBool =
+    std::sync::atomic::AtomicBool::new(false);
+
 pub const DICTATE_WINDOW_LABEL: &str = "dictate";
 const DICTATE_WINDOW_WIDTH: f64 = 420.0;
 const DICTATE_WINDOW_HEIGHT: f64 = 64.0;
@@ -119,6 +127,15 @@ pub fn show_dictate_window(app: &tauri::AppHandle) {
 const LEGACY_PORT: u16 = 8000;
 pub(crate) const SERVER_PORT: u16 = 17493;
 
+#[cfg(desktop)]
+fn show_main_window(app: &tauri::AppHandle) {
+    if let Some(window) = app.get_webview_window("main") {
+        let _ = window.unminimize();
+        let _ = window.show();
+        let _ = window.set_focus();
+    }
+}
+
 /// Find a voicebox-server process listening on a given port (Windows only).
 ///
 /// Uses PowerShell `Get-NetTCPConnection` to look up the PID owning the port,
@@ -680,6 +697,7 @@ async fn start_server(
     // PyInstaller bundles can be slow on first import, especially torch/transformers
     let timeout = tokio::time::Duration::from_secs(120);
     let start_time = tokio::time::Instant::now();
+    let mut last_health_check = tokio::time::Instant::now() - tokio::time::Duration::from_secs(1);
     let mut error_output = Vec::new();
 
     loop {
@@ -710,6 +728,21 @@ async fn start_server(
             return Err("Server startup timeout - check Console.app for detailed logs".to_string());
         }
 
+        // Windows sidecars are built with PyInstaller's --noconsole flag, so
+        // stdout/stderr may be redirected to NUL and never produce the Uvicorn
+        // readiness lines handled below. Poll the Voicebox-specific health
+        // endpoint as a platform-independent readiness signal as well.
+        if last_health_check.elapsed() >= tokio::time::Duration::from_secs(1) {
+            last_health_check = tokio::time::Instant::now();
+            let healthy = tokio::task::spawn_blocking(|| check_health(SERVER_PORT))
+                .await
+                .unwrap_or(false);
+            if healthy {
+                println!("Server is ready (health check passed)!");
+                break;
+            }
+        }
+
         match tokio::time::timeout(tokio::time::Duration::from_millis(100), rx.recv()).await {
             Ok(Some(event)) => {
                 match event {
@@ -1396,6 +1429,35 @@ pub fn run() {
                 app.handle().plugin(tauri_plugin_updater::Builder::new().build())?;
                 app.handle().plugin(tauri_plugin_process::init())?;
 
+                let open_item = MenuItem::with_id(app, "open", "Open Voicebox", true, None::<&str>)?;
+                let quit_item = MenuItem::with_id(app, "quit", "Quit Voicebox", true, None::<&str>)?;
+                let tray_menu = Menu::with_items(app, &[&open_item, &quit_item])?;
+
+                TrayIconBuilder::with_id("voicebox-tray")
+                    .icon(app.default_window_icon().expect("Voicebox app icon missing").clone())
+                    .tooltip("Voicebox")
+                    .menu(&tray_menu)
+                    .show_menu_on_left_click(false)
+                    .on_menu_event(|app, event| match event.id.as_ref() {
+                        "open" => show_main_window(app),
+                        "quit" => {
+                            TRAY_QUIT_REQUESTED.store(true, std::sync::atomic::Ordering::SeqCst);
+                            app.exit(0);
+                        }
+                        _ => {}
+                    })
+                    .on_tray_icon_event(|tray, event| {
+                        if let TrayIconEvent::Click {
+                            button: MouseButton::Left,
+                            button_state: MouseButtonState::Up,
+                            ..
+                        } = event
+                        {
+                            show_main_window(tray.app_handle());
+                        }
+                    })
+                    .build(app)?;
+
                 // Resolve the active keyboard layout's V keycode now, on
                 // the main thread, and register an observer for layout
                 // changes. The synthetic-paste hot path then only reads an
@@ -1529,6 +1591,15 @@ pub fn run() {
             let closing = std::sync::Arc::new(std::sync::atomic::AtomicBool::new(false));
             move |window, event| {
             if let WindowEvent::CloseRequested { api, .. } = event {
+                #[cfg(desktop)]
+                if window.label() == "main"
+                    && !TRAY_QUIT_REQUESTED.load(std::sync::atomic::Ordering::SeqCst)
+                {
+                    api.prevent_close();
+                    window.hide().ok();
+                    return;
+                }
+
                 // If we're already in the close flow, let it proceed
                 if closing.load(std::sync::atomic::Ordering::SeqCst) {
                     return;

tauri/vite.config.ts

@@ -1,22 +1,29 @@
+import fs from 'node:fs';
 import path from 'node:path';
 import tailwindcss from '@tailwindcss/vite';
 import react from '@vitejs/plugin-react';
 import { defineConfig } from 'vite';
 import { changelogPlugin } from '../app/plugins/changelog';
 
+function resolveWorkspaceDependency(name: string) {
+  const isolatedPath = path.resolve(__dirname, '../app/node_modules', name);
+  return fs.existsSync(isolatedPath)
+    ? isolatedPath
+    : path.resolve(__dirname, '../node_modules', name);
+}
+
 export default defineConfig({
   plugins: [react(), tailwindcss(), changelogPlugin(path.resolve(__dirname, '..'))],
   resolve: {
     alias: {
       '@': path.resolve(__dirname, '../app/src'),
-      react: path.resolve(__dirname, '../app/node_modules/react'),
-      'react-dom': path.resolve(__dirname, '../app/node_modules/react-dom'),
-      '@tanstack/react-query': path.resolve(__dirname, '../app/node_modules/@tanstack/react-query'),
-      '@tanstack/react-query-devtools': path.resolve(
-        __dirname,
-        '../app/node_modules/@tanstack/react-query-devtools',
+      react: resolveWorkspaceDependency('react'),
+      'react-dom': resolveWorkspaceDependency('react-dom'),
+      '@tanstack/react-query': resolveWorkspaceDependency('@tanstack/react-query'),
+      '@tanstack/react-query-devtools': resolveWorkspaceDependency(
+        '@tanstack/react-query-devtools',
       ),
-      zustand: path.resolve(__dirname, '../app/node_modules/zustand'),
+      zustand: resolveWorkspaceDependency('zustand'),
     },
     dedupe: ['react', 'react-dom', '@tanstack/react-query', 'zustand'],
   },