Voxta

Voxta 146 Beta: New Running Modules System, Preset-based Configuration & Image Generation!

Added 2025-06-12 02:04:09 +0000 UTC

Hey everyone,

Beta 146 is a huge one! This update delivers a massive overhaul to how Voxta manages AI models, giving you unprecedented control and flexibility. We've moved all model settings into presets to make switching between LLMs effortless, introduced a new system to see and manage running AI services, and we're launching experimental support for Image Generation! Plus, there’s a new logo!

⚠️ CRITICAL UPDATE INFORMATION:
To enable the new features, your old Text Generation presets will be deleted when you update.
After updating, please go to Settings -> Modules and click "Install All" to get the new default presets for your installed models. We're sorry for the hassle, but this is a necessary step for a much better system!

Here’s the breakdown:

⚙️ Core Overhaul: Unprecedented Control & Flexibility

New "Running Modules" System: Ever wonder what's eating your VRAM? Now you can see exactly which models and services are loaded and running. The new "Running Modules" page under Settings lets you shut down services you're not using to free up resources instantly.
Effortless Model Swapping with Presets: This is a game-changer. All model-specific fields have been moved to presets. This means you can now switch from ExLlamaV2 to LlamaSharp or KoboldAI with a single click, and all the correct settings will load with it.
Run Multiple Chats at Once: You can now run multiple chat sessions in parallel without them interfering with each other. Perfect for power users and testers!
Deeper KoboldAI Integration: You can now point Voxta to your KoboldCPP executable and have it launch automatically.

🎨 New Creative & Power-User Features

Experimental Image Generation: Bring your scenes to life! We've added early support for image generation using OpenAI, KoboldAI, and a future ComfyUI module. Generate character portraits, scene backdrops, and more right from Voxta. Check out the new playground to test it out!
⚠️ To enable this feature, open appsettings.json and set "EnableImageGen": true.
Smarter AI with Thinking Format: OpenAI Compatible and OpenRouter models now support "Thinking Format," giving you better insight into the AI's process.
Proof-of-Concept MCP Support: We've added initial integration for the Model Context Protocol (MCP), paving the way for more powerful and standardized tool use in the future.

✨ UI & Desktop Experience Polish

A Brand-New Logo!
Better Feedback: The UI now has visual displays to show you when modules are loading and a progress bar for file downloads, so you always know what's happening.
Improved Diagnostics: The diagnostics view is now private and shows a sequential log of all inferences, with direct links from a chat message to its specific generation data, making debugging much easier.
Minimize to Tray (Desktop): The desktop app can now be minimized to the system tray, keeping it running neatly in the background.

🛠️ Key Fixes & Engine Updates

Latest Module Versions: We've updated ExLlamaV2 (0.3.1), Coqui (0.26.2), LlamaSharp (0.24.0), and WhisperLive (0.7.1) to their latest versions.
Smarter Text-to-Speech: Voxta is now better at not splitting text inside quotation marks into separate audio clips.
Optimized Avatars: Avatars are now automatically resized to a 2:3 aspect ratio, which reduces memory usage and improves animation performance.
Housekeeping: We've removed obsolete services like ChromaDB and Silero to streamline the app. We also fixed bugs related to tokenizer switching, interrupted downloads, and unfinished sentences being kept in chat history.

Important Notes: 🛠️

Remember, this is still a beta. We rely on your feedback! Please test the new "Running Modules" page and let us know how the new preset system feels.
Image generation is highly experimental. We'd love to see what you create and hear your thoughts on how to improve it!
Hit us up on Discord or comment here with any feedback or issues.

Links: