Common issues

Find your symptom below. If you don't see it here, check the Terminal for error messages and ask in the Voxta Discord.

Voxta won't launch — WebView2 COM error

Symptom — Voxta crashes on launch with an error like:

System.Runtime.InteropServices.COMException (0x8007139F):
The group or resource is not in the correct state to perform the requested operation.

Cause — a stale registry entry under AppCompatFlags / Layers targeting msedgewebview2.exe confuses WebView2 (the Edge-based component Voxta's desktop wrapper uses).

Fix:

1. Press Win + R, type regedit, Enter.
2. Navigate to HKEY_CURRENT_USER\Software\Microsoft\Windows NT\CurrentVersion\AppCompatFlags\Layers.
3. Find the entry containing msedgewebview2.exe. Right-click → Delete.
4. Relaunch Voxta.

Reference: Microsoft WebView2 issue #3008.

Voxta won't launch — folder has spaces in path

Symptom — Voxta fails to start with:

System.InvalidOperationException: Cannot run Voxta from a folder with spaces.
Please move your installation elsewhere.

Cause — Voxta does not support installation paths that contain spaces. The internal path resolution chokes on them.

Fix — move the Voxta installation to a path without spaces:

Update any shortcuts pointing at the old path, then relaunch.

"Failed to decrypt data" — API keys broken after reinstall/restore

Symptom — when launching a service that needs an API key, Voxta shows:

Failed to decrypt data. This may happen with API keys when Windows does not have the same key, for example when copying on another machine. Re-enter your API key or secret in the configuration of the module you are trying to use.

The inner exception mentions System.Security.Cryptography.ProtectedData.Unprotect.

When this happens — after any of:

• You reinstalled Windows (most common cause — fresh install wipes DPAPI keys even if you kept the same username).
• You restored Windows from a system image onto new disk / new hardware.
• You copied the Voxta Data/ folder to another PC.
• You signed in as a different Windows user.
• You restored your Windows user profile after corruption.

Cause — Voxta uses Windows DPAPI to encrypt API keys before storing them in Data/. DPAPI keys are tied to the specific Windows account on the specific install. Reinstalling Windows generates new DPAPI keys, even if you reused the same username.

Fix — re-enter each affected secret:

1. Open Manage Services → the affected module (OpenAI, ElevenLabs, etc.).
2. Re-type the API key.
3. Save. Voxta re-encrypts it with the current DPAPI key.
4. Repeat for every module that stores a secret.

Chat is slow / responses take forever

Most common causes:

• Local LLM is hitting VRAM limit — the model is spilling into system RAM and inference falls off a cliff. Open Task Manager (Ctrl+Shift+Esc) → Performance → GPU and watch dedicated VRAM. If it's pegged, switch to a smaller / more quantized model.
• You're running VAM or another GPU-heavy app at the same time — see VAM optimization for the FPS-cap / physics-rate / MSAA trim recipe.
• Cloud LLM is having a bad day — try a different provider or model from the LLM catalog.

Character won't speak (text only, no audio)

• TTS service isn't configured or has wrong credentials — go to Test Services, run the test, see what fails.
• Audio output device is wrong — check NAudio / OpenTK.
• TTS service is rate-limited (cloud) — check the Terminal for HTTP errors.

Character won't hear me (no transcription)

• STT service isn't configured — check Test Services.
• Wrong microphone selected — check NAudio / OpenTK device picker.
• Audio RMS filter is too aggressive — Voxta Utilities: Audio RMS Filter blocks silence. If you speak quietly the filter may eat your audio.
• Mic permission denied at the OS level — check Windows Privacy → Microphone.

A service won't install

• Watch the Terminal during the install. Python services log their dependency-install progress there. Errors usually pinpoint the failing package.
• Antivirus may be blocking model downloads or DLL loads. Try whitelisting the Voxta folder.
• Disk space — local LLM models and TTS models can be tens of gigabytes. Check available space.

Still stuck