Voxta v1.8.2: Mobile Grows Up, ARDY Text-to-Motion & HUD You Can Build (Patreon)

Hello everyone,

v1.8.2 is a focused follow-up to the big 1.8 drop, and it lands on three fronts: the mobile android app finally becomes a real Voxta client, a first look at ARDY, our new text-to-motion animation module, and a HUD you can actually build and shape yourself.

📱 Voxta Mobile Grows Up

The mobile app got the attention it has been waiting for.

It's now upgradable in place, and updating no longer wipes your data

Lots of included modules: LlamaSharp and Vosk, Semantic Kernel, and most online services

New Android Speech module for native STT and TTS, straight from the OS

LlamaSharp support, so you can run local inference on your phone (ultra-slow, but it works)

Vosk STT (which, unlike some cloud services, lets you swear)

Reduced install size by a fair amount

Downloads now continue in the background

Back button goes back instead of dumping you out of the app

The webview recovers if Android kills it

HuggingFace and CivitAI browsing now work on mobile, bottom sheets scroll instead of dismissing, and portrait/stage text layout was reworked for small screens

🕺 ARDY: Text-to-Motion (Early Access)

New ARDY animation module: describe a motion in text and get an animation out the other end.

We think this has huge potential for real-time character animation, generating movement on the fly instead of relying on a fixed library of clips. To be upfront about where it stands: we've only implemented the basics for now, and it's heavy. Expect to need around 15GB of VRAM to run it. We're actively looking at paths to optimize it, so treat this release as a first taste rather than the finished feature.

🎛️ A HUD You Can Build

Create and place HUD widgets like buttons, directly in the interface

Resizable and collapsible HUD, and you choose which components stay visible when collapsed

Click HUD items instead of having to right-click them

Button labels are now rendered as templates, so you can use variables and flag conditions inside them

Disabled scenario buttons, driven by their own separate set of flags

Skip text button in stage and portrait view

🎨 UI & Workflow

Models folder browser, and all model directories are now relative to Data/Models

Installed models are shown in the recommended models list

Voxta release notifications you can opt into

Telemetry opt-in: no personal information, it just tells us which modules are actually being used. Thanks to everyone who turns it on, it genuinely shapes what we work on

Open avatar images full screen in portrait view

Service Tests now report VRAM usage plus cold and hot performance

Module settings shortcut and logo in the presets modal

Create characters directly from the picker in scenario roles and package resources

Document canvas toggle moved into the chat input

Max Tokens To Generate can now be unlimited

Confirm and prompt popups now sit above stacked modals

🤖 Scenario Assistant

Chats scroll properly when you open them

The assistant validates and retries when a trigger fails

It now patches scripts instead of rewriting them wholesale

🧩 Modules & Backends

ExLlamaV2 1.2.0, F5-TTS 1.1.22, llama_python_cpp 0.3.44

SenseVoice 1.3.30 with support for the Nano family

Updated model lists for OpenAI, Mistral, XAI, Google, ElevenLabs, Deepgram and AssemblyAI

Discord integration is now a normal module

Refreshed which modules are marked experimental or recommended

⚙️ Platform & Fixes

Smaller download size: native libraries for Semantic Kernel and Vosk are fetched on demand

Lower RAM usage when extracting downloaded native libraries

Stale Vulkan/ROCm SDK env vars are dropped before loading llama_cpp

Chatterbox no longer cuts sentences short, Orpheus reports unsupported GGUF architectures instead of crashing and gets a longer start timeout

Fixed TTS hanging after the first sentence in VaM, MP4s blanking out and never recovering, a microphone downmix memory leak, and the "Failed to acquire chat create/dispose lock" error when cancelling a chat start

Document commands no longer leave a blank hole in the message, and code blocks stay intact when a character writes a document

Owner IDs are stripped from packages embedded in exports

Manage Modules keeps your changes, sliders stop clamping on load, and the Vision Projector dropdown stops reverting to "un-chosen"

...plus a pile of smaller stability and polish fixes throughout.

Thank you all for the continued support and feedback. Whether you're running Voxta from your pocket, poking at ARDY, or laying out your own HUD, we hope you enjoy v1.8.2.

After subscribing to the required Patreon tier, you can get the latest version at: https://portal.voxta.ai/