v0.2.4

# OmniVoice Studio v0.2.4

Real-time event architecture, cross-platform desktop hardening.

---

### What's New

**WebSocket Event Bus** — The sidebar now updates instantly when data changes. Projects, profiles, and history reflect backend mutations in real time with zero polling overhead. Graceful reconnect with exponential backoff.

**Default UI Scale → M** — Ships at 1.3× zoom for better readability out of the box.

---

### Fixed

- Windows/Linux MSI: first launch crashed with `Missing bootstrap resources` — Tauri's `_up_` path rewriting wasn't handled ([#28](https://github.com/debpalash/OmniVoice-Studio/issues/28))
- Windows: backend logs silently lost — log directory pointed to a macOS-only path ([#30](https://github.com/debpalash/OmniVoice-Studio/issues/30))
- Windows: PyTorch hung downloading Triton — now auto-sets `TORCHDYNAMO_DISABLE=1` ([#26](https://github.com/debpalash/OmniVoice-Studio/issues/26))
- Windows: startup timeout too aggressive — increased from 3 min → 5 min for first-run CUDA JIT
- Sidebar: data failed to load on startup due to undefined status function
- Network: reduced idle polling by ~80% (model status 5s→10s, log polling disabled)

---

### Downloads

| | Installer | Size |
|---|---|---|
| 🍎 macOS (Apple Silicon) | [`OmniVoice.Studio_0.2.4_aarch64.dmg`](https://github.com/debpalash/OmniVoice-Studio/releases/download/v0.2.4/OmniVoice.Studio_0.2.4_aarch64.dmg) | ~8 MB |
| 🪟 Windows x64 | [`OmniVoice.Studio_0.2.4_x64_en-US.msi`](https://github.com/debpalash/OmniVoice-Studio/releases/download/v0.2.4/OmniVoice.Studio_0.2.4_x64_en-US.msi) | ~6 MB |
| 🐧 Linux x64 (.deb) | [`OmniVoice.Studio_0.2.4_amd64.deb`](https://github.com/debpalash/OmniVoice-Studio/releases/download/v0.2.4/OmniVoice.Studio_0.2.4_amd64.deb) | ~8 MB |
| 🐧 Linux x64 (AppImage) | [`OmniVoice.Studio_0.2.4_amd64.AppImage`](https://github.com/debpalash/OmniVoice-Studio/releases/download/v0.2.4/OmniVoice.Studio_0.2.4_amd64.AppImage) | ~81 MB |

> The installers are lightweight shells (~6–8 MB). On first launch, the app bootstraps a Python environment and downloads model weights (~2.4 GB). This takes 5–10 minutes — the splash screen shows progress.

---

### Post-Install Notes

<details>
<summary><b>macOS — "app is damaged and can't be opened"</b></summary>

macOS quarantines apps downloaded outside the App Store. After dragging to `/Applications`:

```bash
xattr -cr /Applications/OmniVoice\ Studio.app
```

Open normally after. One-time fix.
</details>

<details>
<summary><b>Windows — Triton / torch.compile warnings</b></summary>

PyTorch may log Triton-related warnings on Windows. These are harmless — `TORCHDYNAMO_DISABLE=1` is now set automatically by the desktop app. If running from source, set it manually:

```
set TORCHDYNAMO_DISABLE=1
```
</details>

<details>
<summary><b>Linux — AppImage needs FUSE</b></summary>

If FUSE isn't available, use the `.deb` or extract-and-run:

```bash
chmod +x OmniVoice.Studio_*.AppImage
./OmniVoice.Studio_*.AppImage --appimage-extract-and-run
```
</details>

---

**Full changelog:** [`v0.2.2...v0.2.4`](https://github.com/debpalash/OmniVoice-Studio/compare/v0.2.2...v0.2.4)