Voxta v1.5.0: Mobile App, Faster Model Support & Vision Upgrades

Hello everyone,

Voxta v1.5.0 is here, and it's a big one! This release marks a major milestone with the introduction of the Voxta Mobile App for Android, lets you stay on the cutting edge of new LLMs with our new llama-cpp Python backend, and brings computer vision to LlamaSharp. On top of that, we've packed in dozens of UX polish items and fixes across the board.

📱 Brand New: Voxta Mobile App (Android)

The most requested feature is finally here. Voxta is now on your phone!

• Android APK Available: Install Voxta directly on your Android device and chat with your companions on the go.

• Voxta Cloud Powered: The mobile app runs on Voxta Cloud, so you get fast, high-quality experiences without needing a beefy GPU in your pocket.

• Same Voxta, Smaller Screen: The full Voxta experience, optimized for mobile.

This opens up a whole new way to interact with your AI companions, whether you're commuting, traveling, or just away from your desk.

🚀 New Features & Integrations

llama-cpp Python (New Backend!)

We've added a brand new llama-cpp module using Python bindings. Why? Because new models drop fast, and Python bindings get updated almost immediately—much faster than LlamaSharp. If you want to run the latest hotness the day it comes out, this is your new best friend.

Computer Vision for LlamaSharp (MTMD)

LlamaSharp now supports multimodal input via MTMD. Show your local model an image and it can actually see it, no cloud required.

Gemma 4 Support, Everywhere

• LlamaSharp updated to 0.27.0 with Gemma 4 support

• ExLlamaV3 updated to 0.0.31 with Gemma 4 support

• Gemma 4 prompt templates added out of the box

Elite Dangerous Smarter Than Ever

The Elite Dangerous module now auto-detects your keybindings from your .binds file and gives the AI knowledge of your key bindings. No more manually telling your companion which keys do what.

ElevenLabs V3

V3 model selection with language code support and an improved stability slider for finer voice control.

Smarter Simple Memory

Simple Memory got a tune-up: word filtering, configurable minimum threshold, and normalized scoring with weight + date ranking. Memory recall just got a lot more relevant.

⚙️ UI & Experience Overhaul

Chat Start Confirmation

Before jumping into a chat, you'll now see a confirmation screen showing your current configuration, so you know exactly what services and settings are active before the conversation begins.

Smarter Copy Button

The copy button now detects the message type:

• Text messages → copies as text

• Image-only messages → copies the image as PNG to your clipboard

Typewriter Toggle

Love the typewriter effect or hate it? Now it's your call—toggle it on or off.

Inline Module Search

When searching the modules list, available modules now show inline. Faster discovery, less clicking.

Image Gen Playground

A max tokens field and a bunch of UI improvements to make experimentation smoother.

HuggingFace & CivitAI Browsers

Both browsers got polish passes for a much nicer browsing experience.

Visual Story Chat Style

Visual Story now respects the character's chat style for a more consistent feel across modes.

Other UI Wins

• GPU device selection added for Coqui and WhisperLive

• Locked collection cards now show a placeholder description overlay

• Edit prompt added to assistant chat image context menu

• Improved login page

• Chats page: crash fixes, filter persistence, confirmation prompts

• Polished edit pages action bar

🛠️ Key Fixes

A massive batch of fixes landed in this release:

• Deepgram STT: Fixed zombie connections and eliminated per-turn reconnections (huge stability win)

• Discord Rich Presence: Now shows the character name, plus fixed callback pump and activity clearing

• VaM: Fixed thinking speech silent error

• Voice previews: Fixed crash with complex models due to large URLs

• Context updates: Fixed context not updating when the participant list changes

• Chain of Thought: Now properly respects disabled "Enable For" checkboxes

• Elite Dangerous: Fixed action inference spam loop

• OpenRouter: Fixed ComputerVision preset creation

• Narration: Fixed first message narration TTS when narrator is None

• Image gen: Fixed image size dropdown in assistant chat and scaling for user-attached images

• Audio: Fixed audio device picker labels and enumeration

• Errors: Fixed error modal when receiving very long error messages

• Database: Fixed DbUpdateConcurrencyException when deleting an already-deleted chat

• Voices: Fixed Catherine voice error and added TTS voice deduplication

• Profile: Fixed profile description being wiped after avatar upload + save

• Narration view: Fixed text size jumping during speech in portrait view

🔒 Backend & Protocol Improvements

• OpenAI: Now merges consecutive messages using the same role (cleaner prompts, fewer tokens)

• Voxta Cloud: Reasoning controls now exposed (implementation coming soon)

• Action inference: Added system_prompt_addons support

• Per-message constraints now supported

• Local Diffusers: Default preset added

• Root user deletion prevented (you can't accidentally lock yourself out!)

📦 Voxta Installer

The NSIS installer got a major upgrade with custom branding and functionality, and we eliminated the staging folder while adding clear status messages during install. A much cleaner first-time experience.

Thank you all for the continued support and feedback. Whether you're trying Voxta on your phone for the first time, running Gemma 4 locally, or showing your local model an image with the new MTMD support—we hope you enjoy v1.5.0!

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