fix: pin numba for release CI wheel compatibility

fix: await for confirmation before deleting voices and channels

.bumpversion.cfg

@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.1.12
+current_version = 0.1.13
 commit = True
 tag = True
 tag_name = v{new_version}

.github/workflows/build-windows.yml

@@ -0,0 +1,63 @@
+name: Build Windows
+
+on:
+  workflow_dispatch:
+
+jobs:
+  build-windows:
+    permissions:
+      contents: write
+    runs-on: windows-latest
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Setup Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+          cache: "pip"
+
+      - name: Install Python dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install pyinstaller
+          pip install -r backend/requirements.txt
+
+      - name: Build Python server
+        shell: bash
+        run: |
+          cd backend
+          python build_binary.py
+
+          PLATFORM=$(rustc --print host-tuple)
+          mkdir -p ../tauri/src-tauri/binaries
+          cp dist/voicebox-server.exe ../tauri/src-tauri/binaries/voicebox-server-${PLATFORM}.exe
+          echo "Built voicebox-server-${PLATFORM}.exe"
+
+      - name: Setup Bun
+        uses: oven-sh/setup-bun@v2
+
+      - name: Install Rust stable
+        uses: dtolnay/rust-toolchain@stable
+
+      - name: Rust cache
+        uses: swatinem/rust-cache@v2
+        with:
+          workspaces: "./tauri/src-tauri -> target"
+
+      - name: Install dependencies
+        run: bun install
+
+      - uses: tauri-apps/tauri-action@v0
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        with:
+          projectPath: tauri
+          tagName: v__VERSION__
+          releaseName: "voicebox v__VERSION__ (test build)"
+          releaseBody: "Test build for audio export fix"
+          releaseDraft: true
+          prerelease: true
+          args: ""
+          includeUpdaterJson: false

CHANGELOG.md

@@ -53,6 +53,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Fixed
+- Audio export failing when Tauri save dialog returns object instead of string path
+
 ### Added
 - **Makefile** - Comprehensive development workflow automation with commands for setup, development, building, testing, and code quality checks
   - Includes Python version detection and compatibility warnings

README.md

@@ -59,7 +59,7 @@
 
 ## What is Voicebox?
 
-Voicebox is a **local-first voice cloning studio** with DAW-like features for professional voice synthesis. Think of it as the **Ollama for voice** — download models, clone voices, and generate speech entirely on your machine.
+Voicebox is a **local-first voice cloning studio** with DAW-like features for professional voice synthesis. Think of it as a **local, free and open-source alternative to ElevenLabs** — download models, clone voices, and generate speech entirely on your machine.
 
 Unlike cloud services that lock your voice data behind subscriptions, Voicebox gives you:
 
@@ -80,10 +80,10 @@ Voicebox is available now for macOS and Windows.
 
 | Platform | Download |
 |----------|----------|
-| macOS (Apple Silicon) | [voicebox_aarch64.app.tar.gz](https://github.com/jamiepine/voicebox/releases/download/v0.1.0/voicebox_aarch64.app.tar.gz) |
+| macOS (Apple Silicon) | [Voicebox_aarch64.app.tar.gz](https://github.com/jamiepine/voicebox/releases/latest/download/Voicebox_aarch64.app.tar.gz) |
-| macOS (Intel) | [voicebox_x64.app.tar.gz](https://github.com/jamiepine/voicebox/releases/download/v0.1.0/voicebox_x64.app.tar.gz) |
+| macOS (Intel) | [Voicebox_x64.app.tar.gz](https://github.com/jamiepine/voicebox/releases/latest/download/Voicebox_x64.app.tar.gz) |
-| Windows (MSI) | [voicebox_0.1.0_x64_en-US.msi](https://github.com/jamiepine/voicebox/releases/download/v0.1.0/voicebox_0.1.0_x64_en-US.msi) |
+| Windows (MSI) | [Latest Windows MSI](https://github.com/jamiepine/voicebox/releases/latest) |
-| Windows (Setup) | [voicebox_0.1.0_x64-setup.exe](https://github.com/jamiepine/voicebox/releases/download/v0.1.0/voicebox_0.1.0_x64-setup.exe) |
+| Windows (Setup) | [Latest Windows Setup](https://github.com/jamiepine/voicebox/releases/latest) |
 
 > **Linux builds coming soon** — Currently blocked by GitHub runner disk space limitations.
 
@@ -233,7 +233,7 @@ See [CONTRIBUTING.md](CONTRIBUTING.md) for detailed setup and contribution guide
 
 ```bash
 # Clone the repo
-git clone https://github.com/voicebox-sh/voicebox.git
+git clone https://github.com/jamiepine/voicebox.git
 cd voicebox
 
 # Setup everything
@@ -247,7 +247,7 @@ make dev
 
 ```bash
 # Clone the repo
-git clone https://github.com/voicebox-sh/voicebox.git
+git clone https://github.com/jamiepine/voicebox.git
 cd voicebox
 
 # Install dependencies
@@ -260,7 +260,7 @@ cd backend && pip install -r requirements.txt && cd ..
 bun run dev
 ```
 
-**Prerequisites:** [Bun](https://bun.sh), [Rust](https://rustup.rs), [Python 3.11+](https://python.org). 
+**Prerequisites:** [Bun](https://bun.sh), [Rust](https://rustup.rs), [Python 3.11+](https://python.org). [XCode on macOS](https://developer.apple.com/xcode/).
 
 **Performance:** 
 - **Apple Silicon (M1/M2/M3)**: Uses MLX backend with native Metal acceleration for 4-5x faster inference

app/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@voicebox/app",
-  "version": "0.1.12",
+  "version": "0.1.13",
   "private": true,
   "type": "module",
   "scripts": {

app/src/components/AudioTab/AudioTab.tsx

@@ -124,6 +124,13 @@ export function AudioTab() {
     );
   }
 
+  const handleChannelDelete = async (e, channelId) => {
+    e.stopPropagation();
+    if (await confirm('Delete this channel?')) {
+      deleteChannel.mutate(channelId);
+    }
+  }
+
   const allChannels = channels || [];
   const allDevices = devices || [];
   const selectedChannel = selectedChannelId
@@ -241,12 +248,7 @@ export function AudioTab() {
                             variant="ghost"
                             size="sm"
                             className="h-8 w-8 p-0"
-                            onClick={(e) => {
+                            onClick={(e) => handleChannelDelete(e, channel.id)}
-                              e.stopPropagation();
-                              if (confirm('Delete this channel?')) {
-                                deleteChannel.mutate(channel.id);
-                              }
-                            }}
                           >
                             <Trash2 className="h-4 w-4" />
                           </Button>

app/src/components/Generation/FloatingGenerateBox.tsx

@@ -1,6 +1,6 @@
 import { useMatchRoute } from '@tanstack/react-router';
 import { AnimatePresence, motion } from 'framer-motion';
-import { Loader2, MessageSquare, Sparkles } from 'lucide-react';
+import { Loader2, SlidersHorizontal, Sparkles } from 'lucide-react';
 import { useEffect, useRef, useState } from 'react';
 import { Button } from '@/components/ui/button';
 import { Form, FormControl, FormField, FormItem, FormMessage } from '@/components/ui/form';
@@ -187,7 +187,7 @@ export function FloatingGenerateBox({
       }}
     >
       <motion.div
-        className="bg-background/30 backdrop-blur-2xl border border-accent/20 rounded-[2rem] shadow-2xl hover:bg-background/40 hover:border-accent/20 transition-all duration-300 overflow-hidden p-3"
+        className="bg-background/30 backdrop-blur-2xl border border-accent/20 rounded-[2rem] shadow-2xl hover:bg-background/40 hover:border-accent/20 transition-all duration-300 p-3"
         transition={{ duration: 0.6, ease: 'easeInOut' }}
       >
         <Form {...form}>
@@ -274,7 +274,7 @@ export function FloatingGenerateBox({
                                   field.ref(node);
                                 }
                               }}
-                              placeholder="Add delivery instructions..."
+                              placeholder="e.g. very happy and excited"
                               className="resize-none bg-transparent border-none focus-visible:ring-0 focus-visible:ring-offset-0 focus:outline-none focus:ring-0 outline-none ring-0 rounded-2xl text-sm placeholder:text-muted-foreground/60 w-full"
                               style={{
                                 minHeight: isExpanded ? '100px' : '32px',
@@ -294,18 +294,27 @@ export function FloatingGenerateBox({
               </motion.div>
 
               <div className="relative shrink-0">
-                <Button
+                <div className="group relative">
-                  type="submit"
+                  <Button
-                  disabled={isPending || !selectedProfileId}
+                    type="submit"
-                  className="h-10 w-10 rounded-full bg-accent hover:bg-accent/90 hover:scale-105 text-accent-foreground shadow-lg hover:shadow-accent/50 transition-all duration-200"
+                    disabled={isPending || !selectedProfileId}
-                  size="icon"
+                    className="h-10 w-10 rounded-full bg-accent hover:bg-accent/90 hover:scale-105 text-accent-foreground shadow-lg hover:shadow-accent/50 transition-all duration-200"
-                >
+                    size="icon"
-                  {isPending ? (
+                  >
-                    <Loader2 className="h-4 w-4 animate-spin" />
+                    {isPending ? (
-                  ) : (
+                      <Loader2 className="h-4 w-4 animate-spin" />
-                    <Sparkles className="h-4 w-4" />
+                    ) : (
-                  )}
+                      <Sparkles className="h-4 w-4" />
-                </Button>
+                    )}
+                  </Button>
+                  <span className="pointer-events-none absolute bottom-full left-1/2 -translate-x-1/2 mb-2 whitespace-nowrap rounded-md bg-popover px-3 py-1.5 text-xs text-popover-foreground border border-border opacity-0 transition-opacity group-hover:opacity-100 z-[9999]">
+                    {isPending
+                      ? 'Generating...'
+                      : !selectedProfileId
+                        ? 'Select a voice profile first'
+                        : 'Generate speech'}
+                  </span>
+                </div>
                 <AnimatePresence>
                   {isExpanded && (
                     <motion.div
@@ -315,20 +324,25 @@ export function FloatingGenerateBox({
                       transition={{ duration: 0.2 }}
                       className="absolute top-0 right-[calc(100%+0.5rem)]"
                     >
-                      <Button
+                      <div className="group relative">
-                        type="button"
+                        <Button
-                        variant="ghost"
+                          type="button"
-                        size="icon"
+                          variant="ghost"
-                        onClick={() => setIsInstructMode(!isInstructMode)}
+                          size="icon"
-                        className={cn(
+                          onClick={() => setIsInstructMode(!isInstructMode)}
-                          'h-10 w-10 rounded-full transition-all duration-200',
+                          className={cn(
-                          isInstructMode
+                            'h-10 w-10 rounded-full transition-all duration-200',
-                            ? 'bg-accent text-accent-foreground border border-accent hover:bg-accent/90'
+                            isInstructMode
-                            : 'bg-card border border-border hover:bg-background/50',
+                              ? 'bg-accent text-accent-foreground border border-accent hover:bg-accent/90'
-                        )}
+                              : 'bg-card border border-border hover:bg-background/50',
-                      >
+                          )}
-                        <MessageSquare className="h-4 w-4" />
+                        >
-                      </Button>
+                          <SlidersHorizontal className="h-4 w-4" />
+                        </Button>
+                        <span className="pointer-events-none absolute bottom-full left-1/2 -translate-x-1/2 mb-2 whitespace-nowrap rounded-md bg-popover px-3 py-1.5 text-xs text-popover-foreground border border-border opacity-0 transition-opacity group-hover:opacity-100 z-[9999]">
+                          Fine tune instructions
+                        </span>
+                      </div>
                     </motion.div>
                   )}
                 </AnimatePresence>

app/src/components/VoiceProfiles/AudioSampleRecording.tsx

@@ -58,6 +58,7 @@ export function AudioSampleRecording({
   // Request microphone access when component mounts
   useEffect(() => {
     if (!showWaveform) return;
+    if (!navigator.mediaDevices || !navigator.mediaDevices.getUserMedia) return;
 
     let stream: MediaStream | null = null;
 

app/src/components/VoiceProfiles/ProfileForm.tsx

@@ -43,7 +43,7 @@ import {
 } from '@/lib/hooks/useProfiles';
 import { useSystemAudioCapture } from '@/lib/hooks/useSystemAudioCapture';
 import { useTranscription } from '@/lib/hooks/useTranscription';
-import { formatAudioDuration, getAudioDuration } from '@/lib/utils/audio';
+import { convertToWav, formatAudioDuration, getAudioDuration } from '@/lib/utils/audio';
 import { usePlatform } from '@/platform/PlatformContext';
 import { useServerStore } from '@/stores/serverStore';
 import { type ProfileFormDraft, useUIStore } from '@/stores/uiStore';
@@ -505,10 +505,23 @@ export function ProfileForm() {
           language: data.language,
         });
 
+        // Convert non-WAV uploads to WAV so the backend can always use soundfile.
+        // Recorded audio is already WAV (from useAudioRecording's convertToWav call).
+        let fileToUpload: File = sampleFile;
+        if (!sampleFile.type.includes('wav') && !sampleFile.name.toLowerCase().endsWith('.wav')) {
+          try {
+            const wavBlob = await convertToWav(sampleFile);
+            const wavName = sampleFile.name.replace(/\.[^.]+$/, '.wav');
+            fileToUpload = new File([wavBlob], wavName, { type: 'audio/wav' });
+          } catch {
+            // If browser can't decode the format, send the original and let the backend try.
+          }
+        }
+
         try {
           await addSample.mutateAsync({
             profileId: profile.id,
-            file: sampleFile,
+            file: fileToUpload,
             referenceText: referenceText,
           });
 

app/src/components/VoicesTab/VoicesTab.tsx

@@ -79,8 +79,8 @@ export function VoicesTab() {
     setDialogOpen(true);
   };
 
-  const handleDelete = (profileId: string) => {
+  const handleProfileDelete = async (profileId: string) => {
-    if (confirm('Are you sure you want to delete this profile?')) {
+    if (await confirm('Are you sure you want to delete this profile?')) {
       deleteProfile.mutate(profileId);
     }
   };
@@ -147,7 +147,7 @@ export function VoicesTab() {
                 channels={channels || []}
                 onChannelChange={(channelIds) => handleChannelChange(profile.id, channelIds)}
                 onEdit={() => handleEdit(profile.id)}
-                onDelete={() => handleDelete(profile.id)}
+                onDelete={() => handleProfileDelete(profile.id)}
               />
             ))}
           </TableBody>

app/src/lib/hooks/useAudioRecording.ts

@@ -20,11 +20,13 @@ export function useAudioRecording({
   const streamRef = useRef<MediaStream | null>(null);
   const timerRef = useRef<number | null>(null);
   const startTimeRef = useRef<number | null>(null);
+  const cancelledRef = useRef<boolean>(false);
 
   const startRecording = useCallback(async () => {
     try {
       setError(null);
       chunksRef.current = [];
+      cancelledRef.current = false;
       setDuration(0);
 
       // Check if getUserMedia is available
@@ -87,31 +89,34 @@ export function useAudioRecording({
       };
 
       mediaRecorder.onstop = async () => {
+        // Snapshot the cancellation flag and recorded duration immediately —
+        // cancelRecording() clears chunks and sets cancelledRef synchronously
+        // before this async handler runs, so we must check it first.
+        const wasCancelled = cancelledRef.current;
+        const recordedDuration = startTimeRef.current
+          ? (Date.now() - startTimeRef.current) / 1000
+          : undefined;
+
         const webmBlob = new Blob(chunksRef.current, { type: 'audio/webm' });
 
-        // Convert to WAV format to avoid needing ffmpeg on backend
+        // Stop all tracks now that we have the data
-        try {
-          const wavBlob = await convertToWav(webmBlob);
-
-          // Pass the actual recorded duration
-          const recordedDuration = startTimeRef.current
-            ? (Date.now() - startTimeRef.current) / 1000
-            : undefined;
-          onRecordingComplete?.(wavBlob, recordedDuration);
-        } catch (err) {
-          console.error('Error converting audio to WAV:', err);
-          // Fallback to original blob if conversion fails
-          const recordedDuration = startTimeRef.current
-            ? (Date.now() - startTimeRef.current) / 1000
-            : undefined;
-          onRecordingComplete?.(webmBlob, recordedDuration);
-        }
-
-        // Stop all tracks
         streamRef.current?.getTracks().forEach((track) => {
           track.stop();
         });
         streamRef.current = null;
+
+        // Don't fire completion callback if the recording was cancelled
+        if (wasCancelled) return;
+
+        // Convert to WAV format to avoid needing ffmpeg on backend
+        try {
+          const wavBlob = await convertToWav(webmBlob);
+          onRecordingComplete?.(wavBlob, recordedDuration);
+        } catch (err) {
+          console.error('Error converting audio to WAV:', err);
+          // Fallback to original blob if conversion fails
+          onRecordingComplete?.(webmBlob, recordedDuration);
+        }
       };
 
       mediaRecorder.onerror = (event) => {
@@ -167,9 +172,10 @@ export function useAudioRecording({
 
   const cancelRecording = useCallback(() => {
     if (mediaRecorderRef.current) {
+      cancelledRef.current = true; // Must be set before stop() triggers onstop
+      chunksRef.current = [];
       mediaRecorderRef.current.stop();
       setIsRecording(false);
-      chunksRef.current = [];
       setDuration(0);
     }
 

app/src/lib/utils/audio.ts

@@ -22,6 +22,11 @@ export function formatAudioDuration(seconds: number): string {
  * If the file has a recordedDuration property (from recording hooks),
  * use that instead of trying to read metadata. This fixes issues on Windows
  * where WebM files from MediaRecorder don't have proper duration metadata.
+ *
+ * For uploaded files we use AudioContext.decodeAudioData which fully decodes
+ * the audio and returns the exact duration. This is more reliable than
+ * HTMLMediaElement.duration which can return incorrect large values for VBR
+ * MP3 files that lack a proper XING/VBRI header.
  */
 export async function getAudioDuration(
   file: File & { recordedDuration?: number },
@@ -30,26 +35,39 @@ export async function getAudioDuration(
     return file.recordedDuration;
   }
 
-  return new Promise((resolve, reject) => {
+  // Use Web Audio API for accurate duration — avoids VBR MP3 metadata issues.
-    const audio = new Audio();
+  try {
-    const url = URL.createObjectURL(file);
+    const audioContext = new AudioContext();
+    try {
+      const arrayBuffer = await file.arrayBuffer();
+      const audioBuffer = await audioContext.decodeAudioData(arrayBuffer);
+      return audioBuffer.duration;
+    } finally {
+      await audioContext.close();
+    }
+  } catch {
+    // Fallback: read duration from the media element (less accurate but works for WAV).
+    return new Promise((resolve, reject) => {
+      const audio = new Audio();
+      const url = URL.createObjectURL(file);
 
-    audio.addEventListener('loadedmetadata', () => {
+      audio.addEventListener('loadedmetadata', () => {
-      URL.revokeObjectURL(url);
+        URL.revokeObjectURL(url);
-      if (Number.isFinite(audio.duration) && audio.duration > 0) {
+        if (Number.isFinite(audio.duration) && audio.duration > 0) {
-        resolve(audio.duration);
+          resolve(audio.duration);
-      } else {
+        } else {
-        reject(new Error('Audio file has invalid duration metadata'));
+          reject(new Error('Audio file has invalid duration metadata'));
-      }
+        }
+      });
+
+      audio.addEventListener('error', () => {
+        URL.revokeObjectURL(url);
+        reject(new Error('Failed to load audio file'));
+      });
+
+      audio.src = url;
     });
-
+  }
-    audio.addEventListener('error', () => {
-      URL.revokeObjectURL(url);
-      reject(new Error('Failed to load audio file'));
-    });
-
-    audio.src = url;
-  });
 }
 
 /**

backend/__init__.py

@@ -1,3 +1,3 @@
 # Backend package
 
-__version__ = "0.1.12"
+__version__ = "0.1.13"

backend/backends/pytorch_backend.py

@@ -29,9 +29,23 @@ class PyTorchTTSBackend:
         """Get the best available device."""
         if torch.cuda.is_available():
             return "cuda"
-        elif hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
+        # Intel Arc / Intel Xe GPU via intel-extension-for-pytorch (IPEX)
-            # MPS can have issues, use CPU for stability
+        try:
-            return "cpu"
+            import intel_extension_for_pytorch  # noqa: F401
+            if hasattr(torch, 'xpu') and torch.xpu.is_available():
+                return "xpu"
+        except ImportError:
+            pass
+        # Any GPU on Windows via DirectML (torch-directml)
+        try:
+            import torch_directml
+            if torch_directml.device_count() > 0:
+                return torch_directml.device(0)
+        except ImportError:
+            pass
+        # MPS (Apple Silicon) — kept for completeness but MLX backend is preferred
+        if hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
+            return "cpu"  # MPS disabled for stability; MLX backend handles Apple Silicon
         return "cpu"
     
     def is_loaded(self) -> bool:
@@ -166,11 +180,21 @@ class PyTorchTTSBackend:
 
             # Load the model (tqdm is patched, but filters out non-download progress)
             try:
-                self.model = Qwen3TTSModel.from_pretrained(
+                # Don't pass device_map on CPU: accelerate's meta-tensor mechanism
-                    model_path,
+                # causes "Cannot copy out of meta tensor" when moving to CPU.
-                    device_map=self.device,
+                # Instead load directly then call .to(device) if needed.
-                    torch_dtype=torch.float32 if self.device == "cpu" else torch.bfloat16,
+                if self.device == "cpu":
-                )
+                    self.model = Qwen3TTSModel.from_pretrained(
+                        model_path,
+                        torch_dtype=torch.float32,
+                        low_cpu_mem_usage=False,
+                    )
+                else:
+                    self.model = Qwen3TTSModel.from_pretrained(
+                        model_path,
+                        device_map=self.device,
+                        torch_dtype=torch.bfloat16,
+                    )
             finally:
                 # Exit the patch context
                 tracker_context.__exit__(None, None, None)
@@ -358,9 +382,22 @@ class PyTorchSTTBackend:
         """Get the best available device."""
         if torch.cuda.is_available():
             return "cuda"
-        elif hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
+        # Intel Arc / Intel Xe GPU via intel-extension-for-pytorch (IPEX)
-            # MPS support for Whisper
+        try:
-            return "cpu"  # Use CPU for stability
+            import intel_extension_for_pytorch  # noqa: F401
+            if hasattr(torch, 'xpu') and torch.xpu.is_available():
+                return "xpu"
+        except ImportError:
+            pass
+        # Any GPU on Windows via DirectML (torch-directml)
+        try:
+            import torch_directml
+            if torch_directml.device_count() > 0:
+                return torch_directml.device(0)
+        except ImportError:
+            pass
+        if hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
+            return "cpu"  # MPS disabled for stability
         return "cpu"
     
     def is_loaded(self) -> bool:

backend/build_binary.py

@@ -83,9 +83,13 @@ def build_server():
             '--hidden-import', 'mlx_audio.stt',
             '--collect-submodules', 'mlx',
             '--collect-submodules', 'mlx_audio',
-            # Collect MLX data files including Metal shader libraries (.metallib)
+            # Use --collect-all so PyInstaller bundles both data files AND
-            '--collect-data', 'mlx',
+            # native shared libraries (.dylib, .metallib) for MLX.
-            '--collect-data', 'mlx_audio',
+            # Previously only --collect-data was used, which caused MLX to
+            # raise OSError at runtime inside the bundled binary because
+            # the Metal shader libraries were missing.
+            '--collect-all', 'mlx',
+            '--collect-all', 'mlx_audio',
         ])
     else:
         print("Building for non-Apple Silicon platform - PyTorch only")

backend/config.py

@@ -4,8 +4,17 @@ Configuration module for voicebox backend.
 Handles data directory configuration for production bundling.
 """
 
+import os
 from pathlib import Path
 
+# Allow users to override the HuggingFace model download directory.
+# Set VOICEBOX_MODELS_DIR to an absolute path before starting the server.
+# This sets HF_HUB_CACHE so all huggingface_hub downloads go to that path.
+_custom_models_dir = os.environ.get("VOICEBOX_MODELS_DIR")
+if _custom_models_dir:
+    os.environ["HF_HUB_CACHE"] = _custom_models_dir
+    print(f"[config] Model download path set to: {_custom_models_dir}")
+
 # Default data directory (used in development)
 _data_dir = Path("data")
 

backend/main.py

@@ -22,6 +22,24 @@ import uuid
 import asyncio
 import signal
 import os
+from urllib.parse import quote
+
+
+def _safe_content_disposition(disposition_type: str, filename: str) -> str:
+    """Build a Content-Disposition header that is safe for non-ASCII filenames.
+
+    Uses RFC 5987 ``filename*`` parameter so that browsers can decode
+    UTF-8 filenames while the ``filename`` fallback stays ASCII-only.
+    """
+    ascii_name = "".join(
+        c for c in filename if c.isascii() and (c.isalnum() or c in " -_.")
+    ).strip() or "download"
+    utf8_name = quote(filename, safe="")
+    return (
+        f'{disposition_type}; filename="{ascii_name}"; '
+        f"filename*=UTF-8''{utf8_name}"
+    )
+
 
 from . import database, models, profiles, history, tts, transcribe, config, export_import, channels, stories, __version__
 from .database import get_db, Generation as DBGeneration, VoiceProfile as DBVoiceProfile
@@ -77,10 +95,39 @@ async def health():
     tts_model = tts.get_tts_model()
     backend_type = get_backend_type()
 
-    # Check for GPU availability (CUDA or MPS)
+    # Check for GPU availability (CUDA, MPS, Intel Arc XPU, or DirectML)
     has_cuda = torch.cuda.is_available()
     has_mps = hasattr(torch.backends, 'mps') and torch.backends.mps.is_available()
-    gpu_available = has_cuda or has_mps
+
+    # Intel Arc / Intel Xe via intel-extension-for-pytorch (IPEX)
+    has_xpu = False
+    xpu_name = None
+    try:
+        import intel_extension_for_pytorch as ipex  # noqa: F401
+        if hasattr(torch, 'xpu') and torch.xpu.is_available():
+            has_xpu = True
+            try:
+                xpu_name = torch.xpu.get_device_name(0)
+            except Exception:
+                xpu_name = "Intel GPU"
+    except ImportError:
+        pass
+
+    # DirectML backend (torch-directml) for any Windows GPU
+    has_directml = False
+    directml_name = None
+    try:
+        import torch_directml
+        if torch_directml.device_count() > 0:
+            has_directml = True
+            try:
+                directml_name = torch_directml.device_name(0)
+            except Exception:
+                directml_name = "DirectML GPU"
+    except ImportError:
+        pass
+
+    gpu_available = has_cuda or has_mps or has_xpu or has_directml or backend_type == "mlx"
 
     gpu_type = None
     if has_cuda:
@@ -89,6 +136,10 @@ async def health():
         gpu_type = "MPS (Apple Silicon)"
     elif backend_type == "mlx":
         gpu_type = "Metal (Apple Silicon via MLX)"
+    elif has_xpu:
+        gpu_type = f"XPU ({xpu_name})"
+    elif has_directml:
+        gpu_type = f"DirectML ({directml_name})"
 
     vram_used = None
     if has_cuda:
@@ -252,12 +303,17 @@ async def add_profile_sample(
     db: Session = Depends(get_db),
 ):
     """Add a sample to a voice profile."""
-    # Save uploaded file to temporary location
+    # Preserve the uploaded file's extension so librosa can detect format correctly.
-    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
+    # Defaulting to .wav was causing soundfile to reject MP3/WebM content as invalid WAV.
+    _allowed_audio_exts = {'.wav', '.mp3', '.m4a', '.ogg', '.flac', '.aac', '.webm', '.opus'}
+    _uploaded_ext = Path(file.filename or '').suffix.lower()
+    file_suffix = _uploaded_ext if _uploaded_ext in _allowed_audio_exts else '.wav'
+
+    with tempfile.NamedTemporaryFile(suffix=file_suffix, delete=False) as tmp:
         content = await file.read()
         tmp.write(content)
         tmp_path = tmp.name
-    
+
     try:
         sample = await profiles.add_profile_sample(
             profile_id,
@@ -268,6 +324,8 @@ async def add_profile_sample(
         return sample
     except ValueError as e:
         raise HTTPException(status_code=400, detail=str(e))
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Failed to process audio file: {str(e)}")
     finally:
         # Clean up temp file
         Path(tmp_path).unlink(missing_ok=True)
@@ -388,7 +446,7 @@ async def export_profile(
             io.BytesIO(zip_bytes),
             media_type="application/zip",
             headers={
-                "Content-Disposition": f'attachment; filename="{filename}"'
+                "Content-Disposition": _safe_content_disposition("attachment", filename)
             }
         )
     except ValueError as e:
@@ -542,47 +600,50 @@ async def generate_speech(
         if not profile:
             raise HTTPException(status_code=404, detail="Profile not found")
         
-        # Create voice prompt from profile
-        voice_prompt = await profiles.create_voice_prompt_for_profile(
-            data.profile_id,
-            db,
-        )
-        
         # Generate audio
+
+        # Resolve model size and load the correct model FIRST.
+        # This must happen before create_voice_prompt_for_profile because that
+        # function calls load_model_async(None), which falls back to self.model_size.
+        # If the model is already loaded with the right size at that point, it
+        # returns immediately and the voice prompt is created by the correct model.
         tts_model = tts.get_tts_model()
-        # Load the requested model size if different from current (async to not block)
         model_size = data.model_size or "1.7B"
 
         # Check if model needs to be downloaded first
         model_path = tts_model._get_model_path(model_size)
-        if model_path.startswith("Qwen/"):
+        if not tts_model._is_model_cached(model_size):
-            # Model not cached - check if it exists remotely or needs download
+            # Model is not fully cached — kick off a background download and tell
-            from huggingface_hub import constants as hf_constants
+            # the client to retry once it's ready.
-            repo_cache = Path(hf_constants.HF_HUB_CACHE) / ("models--" + model_path.replace("/", "--"))
+            model_name = f"qwen-tts-{model_size}"
-            if not repo_cache.exists():
-                # Start download in background
-                model_name = f"qwen-tts-{model_size}"
 
-                async def download_model_background():
+            async def download_model_background():
-                    try:
+                try:
-                        await tts_model.load_model_async(model_size)
+                    await tts_model.load_model_async(model_size)
-                    except Exception as e:
+                except Exception as e:
-                        task_manager.error_download(model_name, str(e))
+                    task_manager.error_download(model_name, str(e))
 
-                task_manager.start_download(model_name)
+            task_manager.start_download(model_name)
-                asyncio.create_task(download_model_background())
+            asyncio.create_task(download_model_background())
 
-                # Return 202 Accepted with download info
+            raise HTTPException(
-                raise HTTPException(
+                status_code=202,
-                    status_code=202,
+                detail={
-                    detail={
+                    "message": f"Model {model_size} is being downloaded. Please wait and try again.",
-                        "message": f"Model {model_size} is being downloaded. Please wait and try again.",
+                    "model_name": model_name,
-                        "model_name": model_name,
+                    "downloading": True,
-                        "downloading": True
+                },
-                    }
+            )
-                )
 
+        # Load (or switch to) the requested model before building the voice prompt
         await tts_model.load_model_async(model_size)
+
+        # Create voice prompt from profile (model is already loaded with correct size)
+        voice_prompt = await profiles.create_voice_prompt_for_profile(
+            data.profile_id,
+            db,
+        )
+
         audio, sample_rate = await tts_model.generate(
             data.text,
             voice_prompt,
@@ -625,6 +686,59 @@ async def generate_speech(
         raise HTTPException(status_code=500, detail=str(e))
 
 
+@app.post("/generate/stream")
+async def stream_speech(
+    data: models.GenerationRequest,
+    db: Session = Depends(get_db),
+):
+    """
+    Generate speech and stream the WAV audio directly without saving to disk.
+
+    Returns raw WAV bytes via a StreamingResponse so the client can start
+    playing audio before the entire file has been received.  This endpoint
+    does NOT create a history entry — use /generate for that.
+    """
+    profile = await profiles.get_profile(data.profile_id, db)
+    if not profile:
+        raise HTTPException(status_code=404, detail="Profile not found")
+
+    tts_model = tts.get_tts_model()
+    model_size = data.model_size or "1.7B"
+
+    if not tts_model._is_model_cached(model_size):
+        raise HTTPException(
+            status_code=400,
+            detail=f"Model {model_size} is not downloaded yet. Use /generate to trigger a download.",
+        )
+
+    # Load the correct model before building the voice prompt (fixes issue #96)
+    await tts_model.load_model_async(model_size)
+
+    voice_prompt = await profiles.create_voice_prompt_for_profile(data.profile_id, db)
+
+    audio, sample_rate = await tts_model.generate(
+        data.text,
+        voice_prompt,
+        data.language,
+        data.seed,
+        data.instruct,
+    )
+
+    wav_bytes = tts.audio_to_wav_bytes(audio, sample_rate)
+
+    async def _wav_stream():
+        # Yield in chunks so large responses don't block the event loop
+        chunk_size = 64 * 1024  # 64 KB
+        for i in range(0, len(wav_bytes), chunk_size):
+            yield wav_bytes[i : i + chunk_size]
+
+    return StreamingResponse(
+        _wav_stream(),
+        media_type="audio/wav",
+        headers={"Content-Disposition": 'attachment; filename="speech.wav"'},
+    )
+
+
 # ============================================
 # HISTORY ENDPOINTS
 # ============================================
@@ -753,7 +867,7 @@ async def export_generation(
             io.BytesIO(zip_bytes),
             media_type="application/zip",
             headers={
-                "Content-Disposition": f'attachment; filename="{filename}"'
+                "Content-Disposition": _safe_content_disposition("attachment", filename)
             }
         )
     except ValueError as e:
@@ -786,7 +900,7 @@ async def export_generation_audio(
         audio_path,
         media_type="audio/wav",
         headers={
-            "Content-Disposition": f'attachment; filename="{filename}"'
+            "Content-Disposition": _safe_content_disposition("attachment", filename)
         }
     )
 
@@ -1054,7 +1168,7 @@ async def export_story_audio(
             io.BytesIO(audio_bytes),
             media_type="audio/wav",
             headers={
-                "Content-Disposition": f'attachment; filename="{filename}"'
+                "Content-Disposition": _safe_content_disposition("attachment", filename)
             }
         )
     except HTTPException:

backend/platform_detect.py

@@ -19,15 +19,17 @@ def is_apple_silicon() -> bool:
 def get_backend_type() -> Literal["mlx", "pytorch"]:
     """
     Detect the best backend for the current platform.
-    
+
     Returns:
-        "mlx" on Apple Silicon (if MLX is available), "pytorch" otherwise
+        "mlx" on Apple Silicon (if MLX is available and functional), "pytorch" otherwise
     """
     if is_apple_silicon():
         try:
-            import mlx
+            import mlx.core  # noqa: F401 — triggers native lib loading
             return "mlx"
-        except ImportError:
+        except (ImportError, OSError, RuntimeError):
-            # MLX not installed, fallback to PyTorch
+            # MLX not installed, or native libraries failed to load inside a
+            # PyInstaller bundle (OSError on missing .dylib / .metallib).
+            # Fall through to PyTorch.
             return "pytorch"
     return "pytorch"

backend/requirements.txt

@@ -18,6 +18,7 @@ qwen-tts>=0.0.5
 librosa>=0.10.0
 soundfile>=0.12.0
 numpy>=1.24.0
+numba>=0.60.0,<0.61.0
 
 # Utilities
 python-multipart>=0.0.6

backend/tts.py

@@ -32,11 +32,3 @@ def audio_to_wav_bytes(audio: np.ndarray, sample_rate: int) -> bytes:
     sf.write(buffer, audio, sample_rate, format="WAV")
     buffer.seek(0)
     return buffer.read()
-
-
-def audio_to_wav_bytes(audio: np.ndarray, sample_rate: int) -> bytes:
-    """Convert audio array to WAV bytes."""
-    buffer = io.BytesIO()
-    sf.write(buffer, audio, sample_rate, format="WAV")
-    buffer.seek(0)
-    return buffer.read()

backend/voicebox-server.spec

@@ -6,8 +6,13 @@ from PyInstaller.utils.hooks import copy_metadata
 datas = []
 hiddenimports = ['backend', 'backend.main', 'backend.config', 'backend.database', 'backend.models', 'backend.profiles', 'backend.history', 'backend.tts', 'backend.transcribe', 'backend.platform_detect', 'backend.backends', 'backend.backends.pytorch_backend', 'backend.utils.audio', 'backend.utils.cache', 'backend.utils.progress', 'backend.utils.hf_progress', 'backend.utils.validation', 'torch', 'transformers', 'fastapi', 'uvicorn', 'sqlalchemy', 'librosa', 'soundfile', 'qwen_tts', 'qwen_tts.inference', 'qwen_tts.inference.qwen3_tts_model', 'qwen_tts.inference.qwen3_tts_tokenizer', 'qwen_tts.core', 'qwen_tts.cli', 'pkg_resources.extern', 'backend.backends.mlx_backend', 'mlx', 'mlx.core', 'mlx.nn', 'mlx_audio', 'mlx_audio.tts', 'mlx_audio.stt']
 datas += collect_data_files('qwen_tts')
-datas += collect_data_files('mlx')
+# Use collect_all (not collect_data_files) so native .dylib and .metallib
-datas += collect_data_files('mlx_audio')
+# files are bundled as binaries, not data. Without this, MLX raises OSError
+# when loading Metal shaders inside the PyInstaller bundle.
+from PyInstaller.utils.hooks import collect_all as _collect_all
+_mlx_datas, _mlx_bins, _mlx_hidden = _collect_all('mlx')
+_mlxa_datas, _mlxa_bins, _mlxa_hidden = _collect_all('mlx_audio')
+datas += _mlx_datas + _mlxa_datas
 datas += copy_metadata('qwen-tts')
 hiddenimports += collect_submodules('qwen_tts')
 hiddenimports += collect_submodules('jaraco')
@@ -18,7 +23,7 @@ hiddenimports += collect_submodules('mlx_audio')
 a = Analysis(
     ['server.py'],
     pathex=[],
-    binaries=[],
+    binaries=_mlx_bins + _mlxa_bins,
     datas=datas,
     hiddenimports=hiddenimports,
     hookspath=[],

docs/overview/introduction.mdx

@@ -5,7 +5,7 @@ description: "Welcome to Voicebox - the open-source voice synthesis studio"
 
 ## What is Voicebox?
 
-Voicebox is a **local-first voice cloning studio** with DAW-like features for professional voice synthesis. Think of it as the **Ollama for voice** — download models, clone voices, and generate speech entirely on your machine.
+Voicebox is a **local-first voice cloning studio** with DAW-like features for professional voice synthesis. Think of it as a **local, free and open-source alternative to ElevenLabs** — download models, clone voices, and generate speech entirely on your machine.
 
 <Frame>
   <img src="/images/app-screenshot-1.webp" alt="Voicebox App Screenshot" />

docs/plans/TTS_PROVIDER_ARCHITECTURE.md

@@ -0,0 +1,964 @@
+# TTS Provider Architecture
+
+**Status:** Planned for v0.1.13
+**Created:** 2025-01-31
+**Problem:** GitHub 2GB release limit + poor UX for frequent updates requiring 2.4GB re-downloads
+
+---
+
+## Overview
+
+Split the monolithic backend into modular components:
+
+1. **Main App** (~150-200MB): Tauri + FastAPI backend + Whisper + UI/profiles/history
+2. **TTS Providers** (downloadable plugins): Separate executables for model inference
+
+This architecture solves:
+
+- ✅ GitHub 2GB release artifact limit
+- ✅ Frequent app updates without re-downloading large python binaries
+- ✅ User choice of compute backend (CPU/GPU/Cloud)
+- ✅ External provider support (OpenAI, custom servers)
+- ✅ Future extensibility
+
+---
+
+## Architecture Diagram
+
+```
+┌─────────────────────────────────────────────────────────┐
+│  Voicebox App (Tauri + Backend)           ~150MB        │
+│  ├─ UI Layer (React)                                    │
+│  ├─ Backend (FastAPI)                                   │
+│  │  ├─ Voice Profiles                                   │
+│  │  ├─ Generation History                               │
+│  │  ├─ Audio Editing / Stories                          │
+│  │  └─ Provider Manager ◄──────────────┐                │
+│  └─ Whisper (bundled, tiny ~50MB)      │                │
+└─────────────────────────────────────────┼────────────────┘
+                                          │
+                            HTTP/IPC      │
+                                          │
+         ┌────────────────────────────────┼─────────────────┐
+         │                                │                 │
+         ▼                                ▼                 ▼
+┌─────────────────┐      ┌─────────────────┐   ┌──────────────────┐
+│ TTS Provider:   │      │ TTS Provider:   │   │ TTS Provider:    │
+│ PyTorch CPU     │      │ PyTorch CUDA    │   │ MLX (Apple)      │
+│                 │      │                 │   │                  │
+│ ~300MB          │      │ ~2.4GB          │   │ ~800MB           │
+│                 │      │                 │   │                  │
+│ Local inference │      │ GPU inference   │   │ Metal inference  │
+└─────────────────┘      └─────────────────┘   └──────────────────┘
+         │                        │                     │
+         └────────────────────────┴─────────────────────┘
+                                  │
+                    ┌─────────────▼──────────────┐
+                    │  Future Providers:         │
+                    │  • Remote Server           │
+                    │  • OpenAI API              │
+                    │  • ElevenLabs              │
+                    │  • Custom Docker Container │
+                    └────────────────────────────┘
+```
+
+---
+
+## Problem Statement
+
+### Current Architecture Issues
+
+**Monolithic Binary:**
+
+- CPU version: ~295MB
+- CUDA version: ~2.37GB
+- GitHub releases: 2GB file size limit (BLOCKED)
+- Updates require re-downloading entire binary
+- Poor UX: update app → restart → download CUDA update → restart again
+
+**User Pain Points:**
+
+1. Cannot release CUDA version on GitHub (over 2GB)
+2. Every app update forces 2.4GB re-download for GPU users
+3. No flexibility (can't use OpenAI, remote servers, etc.)
+4. Wastes bandwidth for small bug fixes
+
+---
+
+## Solution: Pluggable TTS Providers
+
+### Component Breakdown
+
+#### 1. Main App (voicebox.exe / .app / .AppImage)
+
+**Size:** ~100-150MB
+
+**Includes:**
+
+- Tauri runtime + React UI
+- FastAPI backend (pure Python, no PyTorch)
+- Whisper model (tiny, ~50MB)
+- SQLite database
+- Profile/history/audio editing logic
+- Provider management system
+
+**Does NOT include:**
+
+- PyTorch (CPU or CUDA)
+- TTS models (Qwen3-TTS)
+- Heavy ML dependencies
+
+**Updates frequently:** UI fixes, feature additions, non-ML changes
+
+---
+
+#### 2. TTS Provider: PyTorch CPU
+
+**Binary:** `tts-provider-pytorch-cpu.exe`
+**Size:** ~200MB
+
+**Includes:**
+
+- PyTorch CPU build
+- Qwen3-TTS package
+- Transformers
+- No CUDA libraries
+
+**Download source:** Cloudflare R2
+**Updates rarely:** Only when model code changes
+
+---
+
+#### 3. TTS Provider: PyTorch CUDA
+
+**Binary:** `tts-provider-pytorch-cuda.exe`
+**Size:** ~2.4GB
+
+**Includes:**
+
+- PyTorch CUDA build (cu121)
+- Qwen3-TTS package
+- CUDA runtime, cuDNN, cuBLAS
+- Transformers
+
+**Download source:** Cloudflare R2
+**Platform:** Windows + Linux (NVIDIA GPU)
+**Updates rarely:** Only when model code or CUDA version changes
+
+---
+
+#### 4. TTS Provider: MLX
+
+**Binary:** `tts-provider-mlx`
+**Size:** ~150MB
+
+**Includes:**
+
+- MLX framework
+- MLX-optimized Qwen3-TTS
+- Metal acceleration
+
+**Platform:** macOS only (Apple Silicon)
+**Download source:** Cloudflare R2
+
+---
+
+#### 5. TTS Provider: Remote
+
+**Binary:** None (built-in config)
+**Size:** 0MB
+
+**How it works:**
+
+- User provides URL to their own TTS server
+- Backend proxies requests to that server
+- Implements API spec from `EXTERNAL_PROVIDERS.md`
+
+**Use cases:**
+
+- AMD GPU users running their own server
+- Team deployments with shared GPU server
+- Cloud hosting (Modal, RunPod, Replicate)
+
+---
+
+#### 6. TTS Provider: OpenAI
+
+**Binary:** None (API wrapper)
+**Size:** 0MB
+
+**How it works:**
+
+- User provides OpenAI API key
+- Backend wraps OpenAI Audio API
+- Voice profiles map to OpenAI voices
+
+**Benefits:**
+
+- Zero local compute
+- Pay-per-use
+- Instant setup
+
+---
+
+## Communication Protocol
+
+### Provider API Specification
+
+All TTS providers must implement these endpoints:
+
+#### POST /tts/generate
+
+Generate speech from text.
+
+**Request:**
+
+```json
+{
+	"text": "Hello world!",
+	"voice_prompt": {
+		/* voice prompt object */
+	},
+	"language": "en",
+	"seed": 12345,
+	"model_size": "1.7B"
+}
+```
+
+**Response:**
+
+```json
+{
+	"audio": "base64-encoded-audio",
+	"sample_rate": 24000,
+	"duration": 2.5
+}
+```
+
+#### POST /tts/create_voice_prompt
+
+Create voice prompt from reference audio.
+
+**Request:** (multipart/form-data)
+
+- `audio`: Audio file
+- `reference_text`: Transcript
+
+**Response:**
+
+```json
+{
+	"voice_prompt": {
+		/* serialized prompt */
+	}
+}
+```
+
+#### GET /tts/health
+
+Health check.
+
+**Response:**
+
+```json
+{
+	"status": "healthy",
+	"provider": "pytorch-cuda",
+	"version": "1.0.0",
+	"model": "Qwen3-TTS-12Hz-1.7B-Base",
+	"device": "cuda:0"
+}
+```
+
+#### GET /tts/status
+
+Model status.
+
+**Response:**
+
+```json
+{
+	"model_loaded": true,
+	"model_size": "1.7B",
+	"available_sizes": ["0.6B", "1.7B"],
+	"gpu_available": true,
+	"vram_used_mb": 1234
+}
+```
+
+---
+
+## Backend Implementation
+
+### Provider Manager
+
+**File:** `backend/providers/__init__.py`
+
+```python
+class ProviderManager:
+    """Manages TTS provider lifecycle."""
+
+    def __init__(self):
+        self.active_provider: Optional[Provider] = None
+        self.config = load_provider_config()
+
+    async def start_provider(self, provider_type: str) -> str:
+        """Start a TTS provider process."""
+        if provider_type == "pytorch-cpu":
+            return await self._start_local_provider("tts-provider-pytorch-cpu.exe")
+        elif provider_type == "pytorch-cuda":
+            return await self._start_local_provider("tts-provider-pytorch-cuda.exe")
+        elif provider_type == "mlx":
+            return await self._start_local_provider("tts-provider-mlx")
+        elif provider_type == "remote":
+            return self.config["remote_url"]
+        elif provider_type == "openai":
+            return None  # No subprocess, API wrapper
+
+    async def _start_local_provider(self, binary_name: str) -> str:
+        """Start local provider subprocess."""
+        provider_path = get_provider_binary_path(binary_name)
+
+        if not provider_path.exists():
+            raise ProviderNotInstalledException(binary_name)
+
+        # Start subprocess on random port
+        port = get_free_port()
+        process = subprocess.Popen([
+            str(provider_path),
+            "--port", str(port),
+            "--data-dir", str(config.get_data_dir())
+        ])
+
+        # Wait for provider to be ready
+        await wait_for_provider_health(f"http://localhost:{port}")
+
+        self.active_provider = Provider(process, port)
+        return f"http://localhost:{port}"
+
+    async def stop_provider(self):
+        """Stop active provider."""
+        if self.active_provider:
+            self.active_provider.process.terminate()
+            self.active_provider = None
+```
+
+---
+
+### Provider Abstraction
+
+**File:** `backend/providers/base.py`
+
+```python
+class TTSProvider(ABC):
+    """Abstract base for TTS providers."""
+
+    @abstractmethod
+    async def generate(
+        self,
+        text: str,
+        voice_prompt: dict,
+        language: str,
+        seed: Optional[int]
+    ) -> tuple[np.ndarray, int]:
+        """Generate speech audio."""
+        pass
+
+    @abstractmethod
+    async def create_voice_prompt(
+        self,
+        audio_path: str,
+        reference_text: str
+    ) -> dict:
+        """Create voice prompt from reference audio."""
+        pass
+```
+
+**File:** `backend/providers/local.py`
+
+```python
+class LocalProvider(TTSProvider):
+    """Provider that communicates with local subprocess via HTTP."""
+
+    def __init__(self, base_url: str):
+        self.base_url = base_url
+        self.client = httpx.AsyncClient()
+
+    async def generate(self, text, voice_prompt, language, seed):
+        response = await self.client.post(
+            f"{self.base_url}/tts/generate",
+            json={
+                "text": text,
+                "voice_prompt": voice_prompt,
+                "language": language,
+                "seed": seed
+            }
+        )
+        data = response.json()
+        audio = np.frombuffer(base64.b64decode(data["audio"]), dtype=np.float32)
+        return audio, data["sample_rate"]
+```
+
+**File:** `backend/providers/openai.py`
+
+```python
+class OpenAIProvider(TTSProvider):
+    """Provider that wraps OpenAI Audio API."""
+
+    def __init__(self, api_key: str):
+        self.client = OpenAI(api_key=api_key)
+
+    async def generate(self, text, voice_prompt, language, seed):
+        # Map voice_prompt to OpenAI voice name
+        voice = map_profile_to_openai_voice(voice_prompt)
+
+        response = await self.client.audio.speech.create(
+            model="tts-1",
+            voice=voice,
+            input=text
+        )
+
+        # Convert to numpy array
+        audio_data = response.content
+        audio, sr = load_audio_from_bytes(audio_data)
+        return audio, sr
+```
+
+---
+
+## Provider Installation
+
+### Download Manager
+
+**File:** `backend/providers/installer.py`
+
+```python
+class ProviderInstaller:
+    """Handles provider download and installation."""
+
+    async def download_provider(self, provider_type: str):
+        """Download provider binary from R2."""
+
+        binary_name = {
+            "pytorch-cpu": "tts-provider-pytorch-cpu.exe",
+            "pytorch-cuda": "tts-provider-pytorch-cuda.exe",
+            "mlx": "tts-provider-mlx"
+        }[provider_type]
+
+        download_url = f"https://downloads.voicebox.sh/providers/v{PROVIDER_VERSION}/{binary_name}"
+
+        # Download with progress tracking (reuse existing SSE system)
+        await download_with_progress(
+            url=download_url,
+            destination=get_provider_install_path(binary_name),
+            progress_key=f"provider-{provider_type}"
+        )
+```
+
+**Provider Storage Location:**
+
+- Windows: `%APPDATA%/voicebox/providers/`
+- macOS: `~/Library/Application Support/voicebox/providers/`
+- Linux: `~/.local/share/voicebox/providers/`
+
+---
+
+## Frontend Implementation
+
+### Provider Settings UI
+
+**Component:** `app/src/components/ServerSettings/ProviderSettings.tsx`
+
+```tsx
+export function ProviderSettings() {
+	const [selectedProvider, setSelectedProvider] =
+		useState<ProviderType>("auto");
+	const {data: installedProviders} = useQuery({
+		queryKey: ["providers", "installed"],
+		queryFn: () => apiClient.getInstalledProviders(),
+	});
+
+	return (
+		<Card>
+			<CardHeader>
+				<CardTitle>TTS Provider</CardTitle>
+				<CardDescription>Choose how Voicebox generates speech</CardDescription>
+			</CardHeader>
+			<CardContent>
+				<RadioGroup
+					value={selectedProvider}
+					onValueChange={setSelectedProvider}
+				>
+					{/* Auto-detect */}
+					<div className="flex items-center space-x-2">
+						<RadioGroupItem value="auto" id="auto" />
+						<Label htmlFor="auto">
+							<div className="font-medium">Auto-detect (Recommended)</div>
+							<div className="text-sm text-muted-foreground">
+								Automatically choose the best available provider
+							</div>
+						</Label>
+					</div>
+
+					{/* PyTorch CUDA */}
+					<div className="flex items-center justify-between">
+						<div className="flex items-center space-x-2">
+							<RadioGroupItem
+								value="pytorch-cuda"
+								id="cuda"
+								disabled={!gpuAvailable}
+							/>
+							<Label htmlFor="cuda">
+								<div className="font-medium">PyTorch CUDA (NVIDIA GPU)</div>
+								<div className="text-sm text-muted-foreground">
+									4-5x faster inference on NVIDIA GPUs
+								</div>
+							</Label>
+						</div>
+						{!installedProviders?.includes("pytorch-cuda") && gpuAvailable && (
+							<Button
+								onClick={() => downloadProvider("pytorch-cuda")}
+								size="sm"
+							>
+								Download (2.4GB)
+							</Button>
+						)}
+					</div>
+
+					{/* PyTorch CPU */}
+					<div className="flex items-center justify-between">
+						<div className="flex items-center space-x-2">
+							<RadioGroupItem value="pytorch-cpu" id="cpu" />
+							<Label htmlFor="cpu">
+								<div className="font-medium">PyTorch CPU</div>
+								<div className="text-sm text-muted-foreground">
+									Works on any system, slower inference
+								</div>
+							</Label>
+						</div>
+						{!installedProviders?.includes("pytorch-cpu") && (
+							<Button onClick={() => downloadProvider("pytorch-cpu")} size="sm">
+								Download (300MB)
+							</Button>
+						)}
+					</div>
+
+					{/* MLX (macOS only) */}
+					{isMacOS && (
+						<div className="flex items-center justify-between">
+							<div className="flex items-center space-x-2">
+								<RadioGroupItem value="mlx" id="mlx" />
+								<Label htmlFor="mlx">
+									<div className="font-medium">MLX (Apple Silicon)</div>
+									<div className="text-sm text-muted-foreground">
+										Optimized for M1/M2/M3 chips
+									</div>
+								</Label>
+							</div>
+							{!installedProviders?.includes("mlx") && (
+								<Button onClick={() => downloadProvider("mlx")} size="sm">
+									Download (800MB)
+								</Button>
+							)}
+						</div>
+					)}
+
+					{/* Remote */}
+					<div className="space-y-2">
+						<div className="flex items-center space-x-2">
+							<RadioGroupItem value="remote" id="remote" />
+							<Label htmlFor="remote">
+								<div className="font-medium">Remote Server</div>
+								<div className="text-sm text-muted-foreground">
+									Connect to your own TTS server
+								</div>
+							</Label>
+						</div>
+						{selectedProvider === "remote" && (
+							<Input placeholder="http://your-server:8000" className="ml-6" />
+						)}
+					</div>
+
+					{/* OpenAI */}
+					<div className="space-y-2">
+						<div className="flex items-center space-x-2">
+							<RadioGroupItem value="openai" id="openai" />
+							<Label htmlFor="openai">
+								<div className="font-medium">OpenAI API</div>
+								<div className="text-sm text-muted-foreground">
+									Use OpenAI's TTS API (requires API key)
+								</div>
+							</Label>
+						</div>
+						{selectedProvider === "openai" && (
+							<Input type="password" placeholder="sk-..." className="ml-6" />
+						)}
+					</div>
+				</RadioGroup>
+			</CardContent>
+		</Card>
+	);
+}
+```
+
+---
+
+## File Structure
+
+```
+voicebox/
+├── backend/
+│   ├── main.py                    # Main FastAPI app (no TTS code)
+│   ├── providers/
+│   │   ├── __init__.py            # ProviderManager
+│   │   ├── base.py                # TTSProvider ABC
+│   │   ├── local.py               # LocalProvider (subprocess)
+│   │   ├── remote.py              # RemoteProvider (HTTP)
+│   │   ├── openai.py              # OpenAIProvider (API wrapper)
+│   │   └── installer.py           # Provider download logic
+│   ├── profiles.py                # Voice profile management
+│   ├── history.py                 # Generation history
+│   ├── transcribe.py              # Whisper (still bundled)
+│   └── ... (other backend modules)
+│
+├── providers/
+│   ├── pytorch-cpu/
+│   │   ├── main.py                # FastAPI server for TTS
+│   │   ├── tts_backend.py         # PyTorch TTS logic
+│   │   ├── requirements.txt       # torch (CPU), qwen-tts, transformers
+│   │   └── build.spec             # PyInstaller spec
+│   │
+│   ├── pytorch-cuda/
+│   │   ├── main.py                # FastAPI server for TTS
+│   │   ├── tts_backend.py         # PyTorch TTS logic
+│   │   ├── requirements.txt       # torch+cu121, qwen-tts, transformers
+│   │   └── build.spec             # PyInstaller spec
+│   │
+│   └── mlx/
+│       ├── main.py                # FastAPI server for TTS
+│       ├── mlx_backend.py         # MLX TTS logic
+│       ├── requirements.txt       # mlx, qwen-tts-mlx
+│       └── build.spec             # PyInstaller spec
+│
+├── app/                           # Frontend (Tauri + React)
+│   └── src/
+│       └── components/
+│           └── ServerSettings/
+│               └── ProviderSettings.tsx
+│
+└── tauri/
+    └── src-tauri/
+        └── tauri.conf.json        # No externalBin for providers
+```
+
+---
+
+## Migration Path
+
+### Phase 1: Refactor Backend (No User Changes)
+
+**Goal:** Abstract TTS behind provider interface
+
+1. Create `backend/providers/` module structure
+2. Implement `TTSProvider` abstract base class
+3. Create `LocalProvider` wrapper for current PyTorch code
+4. Modify `backend/tts.py` to use provider abstraction
+5. Keep PyTorch bundled in main app
+
+**Result:** Code is prepared, but user experience unchanged
+
+---
+
+### Phase 2: Build Provider Binaries
+
+**Goal:** Create standalone TTS provider executables
+
+1. Create separate PyInstaller specs for each provider
+2. Build provider executables:
+   - `tts-provider-pytorch-cpu.exe` (~300MB)
+   - `tts-provider-pytorch-cuda.exe` (~2.4GB)
+   - `tts-provider-mlx` (~800MB, macOS)
+3. Test subprocess communication
+4. Upload providers to Cloudflare R2
+
+**Result:** Provider binaries exist but aren't used yet
+
+---
+
+### Phase 3: Remove PyTorch from Main App
+
+**Goal:** Split main app from providers
+
+1. Exclude PyTorch/Qwen3-TTS from main app PyInstaller spec
+2. Main app now requires provider download
+3. Update GitHub CI to build multiple artifacts:
+   - `voicebox-{version}-{platform}.exe` (~150MB)
+   - `tts-provider-pytorch-cpu-{version}.exe`
+   - `tts-provider-pytorch-cuda-{version}.exe`
+   - `tts-provider-mlx-{version}` (macOS)
+
+**Result:** Main app is small, providers downloaded separately
+
+---
+
+### Phase 4: Add Provider UI
+
+**Goal:** User-facing provider management
+
+1. Create Provider Settings page
+2. Implement provider download UI
+3. Add provider status indicators
+4. Show active provider in UI
+
+**Result:** Users can choose and download providers
+
+---
+
+### Phase 5: External Providers
+
+**Goal:** Enable remote and cloud providers
+
+1. Implement `RemoteProvider` (HTTP client)
+2. Implement `OpenAIProvider` (API wrapper)
+3. Add provider configuration UI (URLs, API keys)
+4. Document external provider API spec
+
+**Result:** Full provider ecosystem
+
+---
+
+## Provider Versioning
+
+### Independent Versioning
+
+Providers have their own version numbers, independent of the main app:
+
+- **App version:** `v0.2.0` (frequent updates)
+- **Provider version:** `v1.0.0` (rare updates)
+
+### Compatibility Matrix
+
+**Example:**
+
+| App Version | Min Provider Version | Max Provider Version |
+| ----------- | -------------------- | -------------------- |
+| v0.2.0      | v1.0.0               | v1.x.x               |
+| v0.3.0      | v1.0.0               | v1.x.x               |
+| v0.4.0      | v1.2.0               | v1.x.x               |
+| v1.0.0      | v2.0.0               | v2.x.x               |
+
+**Backend checks compatibility:**
+
+```python
+async def check_provider_compatibility(provider_version: str) -> bool:
+    """Check if provider version is compatible with current app."""
+    min_version = "1.0.0"
+    max_version = "1.999.999"
+    return min_version <= provider_version < max_version
+```
+
+**UI shows warning if incompatible:**
+
+```
+⚠️ Provider version 0.9.0 is outdated. Update to v1.0.0+
+```
+
+---
+
+## User Flows
+
+### First-Time Setup
+
+1. User downloads and installs Voicebox (~150MB)
+2. App launches → detects no TTS provider installed
+3. Shows setup wizard:
+
+   ```
+   Choose your TTS provider:
+
+   [ ] PyTorch CUDA (2.4GB)    [Download]
+       ✓ Fastest on NVIDIA GPUs
+       ✗ Requires NVIDIA GPU
+
+   [●] PyTorch CPU (300MB)     [Download]
+       ✓ Works on any system
+       ✗ Slower inference
+
+   [ ] MLX (800MB)             [Download]
+       ✓ Fast on Apple Silicon
+       ✗ macOS only (M1/M2/M3)
+
+   [ ] Remote Server
+       URL: ___________________
+
+   [ ] OpenAI API
+       API Key: ________________
+   ```
+
+4. User selects provider → downloads with progress bar
+5. Provider installs to AppData/Application Support
+6. App starts provider → ready to use
+
+---
+
+### App Update Flow (No Provider Change)
+
+**Scenario:** Bug fix in UI, no backend changes
+
+1. User gets update notification: "Voicebox v0.2.1 available"
+2. Downloads update (~150MB, not 2.4GB!)
+3. Installs and restarts
+4. **Provider stays the same** (no re-download needed)
+5. App starts using existing provider
+
+**User experience:** Fast updates, no multi-GB downloads
+
+---
+
+### Provider Update Flow
+
+**Scenario:** New Qwen3-TTS model version released
+
+1. User opens Settings → Provider tab
+2. Sees notification: "Provider update available (v1.1.0)"
+3. Clicks "Update Provider"
+4. Downloads new provider binary
+5. Old provider binary is replaced
+6. Restart app to use new provider
+
+**Frequency:** Rare (only when TTS model/backend changes)
+
+---
+
+### Switching Providers
+
+**Scenario:** User upgrades to NVIDIA GPU
+
+1. User goes to Settings → Provider
+2. Selects "PyTorch CUDA"
+3. Clicks "Download" → downloads 2.4GB
+4. Download completes → restarts app
+5. App now uses CUDA provider
+
+---
+
+## Benefits
+
+| Benefit                       | Details                                                   |
+| ----------------------------- | --------------------------------------------------------- |
+| **GitHub Releases Work**      | Main app ~150MB << 2GB limit                              |
+| **Fast Updates**              | UI/feature updates don't require re-downloading providers |
+| **User Choice**               | CPU, CUDA, MLX, OpenAI, remote server                     |
+| **External Provider Support** | Users can run their own TTS servers                       |
+| **Bandwidth Savings**         | Only download provider once, app updates are small        |
+| **Future-Proof**              | Easy to add new providers (ElevenLabs, custom models)     |
+| **Team Deployments**          | Multiple users share one remote provider                  |
+| **Cloud-Ready**               | Works with Modal, Replicate, RunPod, etc.                 |
+
+---
+
+## Open Questions
+
+### 1. Provider Versioning
+
+**Question:** Should providers have independent versions or match app version?
+
+**Options:**
+
+- A. Independent (providers: v1.x, app: v0.2.x)
+- B. Matched (both use v0.2.x)
+
+**Recommendation:** Independent versioning with compatibility matrix
+
+---
+
+### 2. Auto-Update Providers
+
+**Question:** Should providers auto-update separately from app?
+
+**Options:**
+
+- A. Manual updates only (user clicks "Update Provider")
+- B. Optional auto-update (user can enable)
+- C. Always auto-update
+
+**Recommendation:** Optional auto-update (default off)
+
+---
+
+### 3. Provider Discovery
+
+**Question:** How does app find installed providers?
+
+**Options:**
+
+- A. Check standard paths in AppData/Application Support
+- B. Registry (Windows) / plist (macOS)
+- C. Config file with provider locations
+
+**Recommendation:** Standard paths + config fallback
+
+---
+
+### 4. Fallback Behavior
+
+**Question:** What if no provider is installed?
+
+**Options:**
+
+- A. Show setup wizard on first launch
+- B. Block app until provider installed
+- C. Allow app to run in "demo mode" (transcription only)
+
+**Recommendation:** Setup wizard on first launch
+
+---
+
+### 5. Provider Auto-Start
+
+**Question:** Should provider start automatically with app?
+
+**Options:**
+
+- A. Always start selected provider on app launch
+- B. Start on-demand (when user generates speech)
+- C. User preference
+
+**Recommendation:** Auto-start (configurable in settings)
+
+---
+
+## Future Enhancements
+
+- [ ] **Provider Marketplace:** Built-in directory of community providers
+- [ ] **Multi-Provider Support:** Use different providers per voice/language
+- [ ] **Provider Health Monitoring:** Automatic failover if provider crashes
+- [ ] **Cost Tracking:** Monitor API usage for OpenAI/cloud providers
+- [ ] **Performance Metrics:** Latency, throughput, VRAM usage dashboards
+- [ ] **Docker Providers:** Run providers in Docker containers
+- [ ] **Provider Plugins:** Load custom providers from user scripts
+
+---
+
+## Related Documents
+
+- [EXTERNAL_PROVIDERS.md](./EXTERNAL_PROVIDERS.md) - External provider support plan
+- [OPENAI_SUPPORT.md](./OPENAI_SUPPORT.md) - OpenAI API compatibility
+- [github-2gb-limit-issue.md](../github-2gb-limit-issue.md) - Original problem
+- [r2-setup.md](../r2-setup.md) - Cloudflare R2 configuration
+
+---
+
+## Contributing
+
+If you want to build a custom TTS provider:
+
+1. Implement the provider API spec (see above)
+2. Test with Voicebox locally
+3. Package as executable (PyInstaller, Docker, etc.)
+4. Share in GitHub Discussions
+
+**Questions?**
+
+- GitHub Issues: [voicebox/issues](https://github.com/jamiepine/voicebox/issues)
+- Discord: Coming soon

landing/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@voicebox/landing",
-  "version": "0.1.12",
+  "version": "0.1.13",
   "description": "Landing page for voicebox.sh",
   "scripts": {
     "dev": "bun --bun next dev --turbo",

landing/src/app/layout.tsx

@@ -1,6 +1,7 @@
 import type { Metadata } from 'next';
 import { Inter } from 'next/font/google';
 import './globals.css';
+import { Banner } from '@/components/Banner';
 import { Footer } from '@/components/Footer';
 import { Header } from '@/components/Header';
 
@@ -31,6 +32,7 @@ export default function RootLayout({ children }: { children: React.ReactNode })
     <html lang="en" suppressHydrationWarning className="dark">
       <body className={inter.variable}>
         <div className="relative min-h-screen bg-background font-sans flex flex-col">
+          <Banner />
           <Header />
           <main className="container mx-auto px-4 sm:px-6 md:px-4 flex-1 py-4 sm:py-6 md:py-0">
             {children}

landing/src/app/page.tsx

@@ -239,8 +239,9 @@ export default function Home() {
           <div className="space-y-6 text-lg text-foreground/80 text-center">
             <p>
               Voicebox is a <strong>local-first voice cloning studio</strong> with DAW-like features
-              for professional voice synthesis. Think of it as the <strong>Ollama for voice</strong>{' '}
+              for professional voice synthesis. Think of it as a{' '}
-              — download models, clone voices, and generate speech entirely on your machine.
+              <strong>local, free and open-source alternative to ElevenLabs</strong> — download
+              models, clone voices, and generate speech entirely on your machine.
             </p>
             <p>
               Unlike cloud services that lock your voice data behind subscriptions, Voicebox gives

landing/src/components/Banner.tsx

@@ -0,0 +1,25 @@
+import { ArrowRight } from 'lucide-react';
+
+export function Banner() {
+  return (
+    <div className="bg-primary/[0.06] border-b border-border backdrop-blur-sm">
+      <div className="container mx-auto px-4">
+        <div className="flex items-center justify-center h-10 text-sm">
+          <a
+            href="https://spacebot.sh"
+            target="_blank"
+            rel="noopener noreferrer"
+            className="flex items-center gap-2 text-muted-foreground hover:text-foreground transition-colors group"
+          >
+            <span>
+              Also by the creator of Voicebox:{' '}
+              <strong className="text-foreground/90">Spacebot</strong>, an AI agent OS for teams.
+              Connect Discord, Slack, or Telegram in one click.
+            </span>
+            <ArrowRight className="h-3.5 w-3.5 transition-transform group-hover:translate-x-0.5" />
+          </a>
+        </div>
+      </div>
+    </div>
+  );
+}

package.json

@@ -1,6 +1,6 @@
 {
   "name": "voicebox",
-  "version": "0.1.12",
+  "version": "0.1.13",
   "private": true,
   "workspaces": [
     "app",

requirements.txt

@@ -0,0 +1,9 @@
+uvicorn
+fastapi
+sqlalchemy
+torch
+torchvision
+soundfile
+librosa
+python-multipart
+huggingface_hub

tauri/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@voicebox/tauri",
   "private": true,
-  "version": "0.1.12",
+  "version": "0.1.13",
   "type": "module",
   "scripts": {
     "dev": "vite",

tauri/src-tauri/Cargo.lock

@@ -5041,7 +5041,7 @@ checksum = "0b928f33d975fc6ad9f86c8f283853ad26bdd5b10b7f1542aa2fa15e2289105a"
 
 [[package]]
 name = "voicebox"
-version = "0.1.11"
+version = "0.1.12"
 dependencies = [
  "base64 0.22.1",
  "core-foundation-sys",

tauri/src-tauri/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "voicebox"
-version = "0.1.12"
+version = "0.1.13"
 description = "A production-quality desktop app for Qwen3-TTS voice cloning and generation"
 authors = ["you"]
 license = ""

tauri/src-tauri/src/audio_capture/linux.rs

@@ -0,0 +1,16 @@
+use crate::audio_capture::AudioCaptureState;
+
+pub async fn start_capture(
+    state: &AudioCaptureState,
+    max_duration_secs: u32,
+) -> Result<(), String> {
+    todo!("implement Linux audio capture")
+}
+
+pub async fn stop_capture(state: &AudioCaptureState) -> Result<String, String> {
+    todo!("implement Linux audio capture stop")
+}
+
+pub fn is_supported() -> bool {
+    false
+}

tauri/src-tauri/src/audio_capture/mod.rs

@@ -2,11 +2,15 @@
 mod macos;
 #[cfg(target_os = "windows")]
 mod windows;
+#[cfg(target_os = "linux")]
+mod linux;
 
 #[cfg(target_os = "macos")]
 pub use macos::*;
 #[cfg(target_os = "windows")]
 pub use windows::*;
+#[cfg(target_os = "linux")]
+pub use linux::*;
 
 use std::sync::{Arc, Mutex};
 

tauri/src-tauri/tauri.conf.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://schema.tauri.app/config/2",
   "productName": "Voicebox",
-  "version": "0.1.12",
+  "version": "0.1.13",
   "identifier": "sh.voicebox.app",
   "build": {
     "beforeDevCommand": "bun run dev",
@@ -12,7 +12,7 @@
   "bundle": {
     "active": true,
     "targets": "all",
-    "createUpdaterArtifacts": true,
+    "createUpdaterArtifacts": false,
     "externalBin": ["binaries/voicebox-server"],
     "icon": [
       "icons/32x32.png",

tauri/src/platform/filesystem.ts

@@ -2,29 +2,25 @@ import type { PlatformFilesystem, FileFilter } from '@/platform/types';
 
 export const tauriFilesystem: PlatformFilesystem = {
   async saveFile(filename: string, blob: Blob, filters?: FileFilter[]) {
-    try {
+    const { save } = await import('@tauri-apps/plugin-dialog');
-      const { save } = await import('@tauri-apps/plugin-dialog');
+    const { writeFile } = await import('@tauri-apps/plugin-fs');
-      const filePath = await save({
-        defaultPath: filename,
-        filters: filters || [],
-      });
 
-      if (filePath) {
+    const filePath = await save({
-        const { writeBinaryFile } = await import('@tauri-apps/plugin-fs');
+      defaultPath: filename,
-        const arrayBuffer = await blob.arrayBuffer();
+      filters: filters || [],
-        await writeBinaryFile(filePath, new Uint8Array(arrayBuffer));
+    });
-      }
+
-    } catch (error) {
+    if (!filePath) return; // User cancelled the dialog
-      console.error('Failed to use Tauri dialog, falling back to browser download:', error);
+
-      // Fall back to browser download if Tauri dialog fails
+    const resolvedPath = typeof filePath === 'string'
-      const url = window.URL.createObjectURL(blob);
+      ? filePath
-      const a = document.createElement('a');
+      : (filePath as { path: string }).path;
-      a.href = url;
+
-      a.download = filename;
+    if (!resolvedPath) {
-      document.body.appendChild(a);
+      throw new Error('Failed to resolve save path from dialog');
-      a.click();
-      window.URL.revokeObjectURL(url);
-      document.body.removeChild(a);
     }
+
+    const arrayBuffer = await blob.arrayBuffer();
+    await writeFile(resolvedPath, new Uint8Array(arrayBuffer));
   },
 };

web/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@voicebox/web",
   "private": true,
-  "version": "0.1.12",
+  "version": "0.1.13",
   "type": "module",
   "scripts": {
     "dev": "vite",
@@ -21,6 +21,7 @@
     "@types/react-dom": "^18.3.0",
     "@typescript-eslint/eslint-plugin": "^7.0.0",
     "@typescript-eslint/parser": "^7.0.0",
+    "@tailwindcss/vite": "^4.0.0",
     "@vitejs/plugin-react": "^4.3.0",
     "eslint": "^8.57.0",
     "eslint-plugin-react-hooks": "^4.6.0",

web/vite.config.ts

@@ -1,9 +1,10 @@
 import path from 'node:path';
 import react from '@vitejs/plugin-react';
+import tailwindcss from '@tailwindcss/vite';
 import { defineConfig } from 'vite';
 
 export default defineConfig({
-  plugins: [react()],
+  plugins: [react(), tailwindcss()],
   resolve: {
     alias: {
       '@': path.resolve(__dirname, '../app/src'),