Voxta - Voxta v1.11.0: Content Hub, Workspaces & Portrait Animation

Hello everyone,

This is the biggest release yet. The Content Hub is a new way to get content right inside Voxta: browse, search and install characters, scenarios and packs, and share your own with the community. Workspaces give a chat a real folder on disk that the AI can read, write and work in, from a new Work page with a file tree beside the chat, and tool calling got a security pass to match. Portrait Animation makes a character's picture speak along with its voice. Decisions, a Discord bot, WAN 2.2 video, expression generation and a web UI reworked for phones fill out the rest.

Python was updated to 3.14, so expect a large download after installing: Python modules are reinstalled once after the upgrade.

There is a lot to dig into, so let's get started.

Installed by default and available in the menu.

Browse, search and install characters, scenarios and packs, with creator profiles

Choose which kind of content you allow in the module configuration

Share your own content with the community

Browse it on the web too, at https://hub.voxta.ai/

A chat can now have a folder on disk that the AI reads, writes and edits files in, runs commands and launches programs from.

The new Work page starts and resumes work in a workspace, with the folder chosen before the chat

A files browser in the right panel, with editing and deleting for text files and viewing for images

Shell commands (Bash and PowerShell) and a Python tool for assistants

Optional notifications when an assistant is done, integrated with Windows and Android notifications

Workspaces put a lot of power in a model's hands, so tool calling got a security pass.

When Allow All is enabled, destructive calls are reviewed before they run

Tool calls are screened by what they actually read and write, and secret reads the path rules could not see are flagged

Web fetch checks with allow and deny lists, and redirect chains are inspected

Revoke allowed tools, or list every tool to grant or revoke access ahead of time

API keys have scopes, and device sign-in asks only for the scopes it needs

Requests coming through a tunnel on this machine count as remote

Shared accounts: several people can use one account, each with their own sign-in (when Users Management is enabled)

A new service type answers multiple-choice questions, such as who speaks next, which action to take or whether a tool call may run, with a model that returns an option instead of text. It answers in well under a second for a fraction of the price, and is enabled by default on Voxta Cloud.

Served through Voxta Cloud and OpenRouter's Decisions API

Decisions mode for ExLlamaV3 and llama.cpp: Gemma 4 12B decisions run about 3.4 times faster with 99% letter-probability

Chat flow and tool call review always use decisions when enabled, and action inference uses them when no action in the layer takes arguments

A typesafe Jev-compatible API to run decisions against your running Voxta instance

A character's picture speaks along with its voice, driven by the new Ditto module

Portrait animation is drawn in the app, and can be tested from the service tests with a voice

Generate expression image assets in a character's Assets tab

Emotes and States work without a scenario, for example Normal and Sitting, or a new Aggressive state

Put a character in your Discord channels, with a character and scenario per channel

A cost limit per channel

Built on new bot sessions (Chat Bridges)

WAN 2.2 video generation inside Voxta

The image viewer can edit any image with a prompt, or turn it into a video

Video cropping, and video generation jobs can resume

ComfyCloud supports img2img, video generation, workflow upload and LoRA panels

Local Diffusers supports Z-Image Turbo and Anima

Krea 2, Qwen Image Edit and Local Diffusers use less memory and run faster, especially with LoRAs

Sprint mode in /imagine for native transparency output with Qwen Image Edit and Local Diffusers

New Upscaler module (Spandrel, chaiNNer, ESRGAN and others)

Image models and LoRAs show a preview thumbnail and label from Civitai

Generated images default to the character's avatar

KoboldCpp video generation (untested)

Favorite voices in Voices Studio and the voice picker

Change voice shortcut on the character menu

ElevenLabs V4, and Cartesia Sonic 3.6 on Voxta Cloud

Faster Python voices warmup

Companion characters have a dedicated prompt to reduce hallucinations

Options to generate Say and Do (event) actions in a chat

A scenario can share chat history with characters who were away

Tool results stay in the prompt until summarization, and a tool timeout or model failure no longer loses the turn

Memory book entries can be edited while a chat is writing, and memory books support thumbnails

The built-in Voxta memory book is now actual documentation

Two chats can no longer share the same chat or memory book

Scripts support setTimeout and setInterval

Disabling explicit content now covers more of the app

Some built-in characters were removed

Python 3.14.7, llama.cpp 0.4.1 with MTP draft support, ExLlamaV3 1.5.4 and ONNX Runtime 1.30.0

The Models folder was reorganized, and the models browser has filtering

NovelAI was removed, along with PythonNet, reducing memory and download size

Voxta Cloud adds background removal and upscaling services

Voxta Cloud sign-ins are kept alive and renewed

A refused model load shows its VRAM breakdown

Clearer errors when max tokens is higher than the context window, and when a reply ran out of tokens or was cut short

MCP tool images reach the model at their original size, and are only scaled down past what the provider refuses

Redesigned right panel, Add Module pages and Character Preview page

An "Add features" link in the right panel, and modules that are not configured are hidden

See your Voxta Cloud credits in the left panel

Launching a character that needs an uninstalled augmentation says so and offers to install it

Startup modals step through new versions, Python package updates and migrations

Errors are sorted into unexpected, fixable by you, or the scenario's fault, and the error dialog shows how to fix them

The Assistants page lists only assistants, marks the default one and can change it

Reasoning moved into the model picker

VRM rendering is capped at 60 fps

Windows notifications

Better performance on high refresh rate displays

VRM popout: mic and voice toggles, resize, context menu, lock button and a saved position

The whole web UI was reworked for phones: drawers, cards and bottom sheets instead of squeezed desktop tables.

Lists, editors and settings open as master-detail drawers

Chats, accounts, API keys, resources and tables are shown as cards

Filters, menus and split buttons open as bottom sheets

The vision, text to speech and speech to text playgrounds are step-by-step flows, and the vision playground opens the real camera

Name prompts stay clear of the keyboard, sliders are dragged by the knob, and the image viewer no longer flickers while pinching

Android notifications

Pair the mobile app with your PC from a QR code

Workspaces support

Voxta.Model targets .NET Standard 2.1 for Unity support

Ticket based WebSocket authentication

Tool calling control API

API key scopes

New to Voxta? Lite is free and runs on Voxta Cloud, so you can create an account and start chatting without any local setup. If you like it, a subscription gets you the full version, with a large library of integrations, custom modules, and the option to run everything on your own hardware.

This post also rolls up v1.10.3, and both versions carry a long list of stability and polish fixes, across tool-using turns, image generation, audio and Linux.

Thank you all, as always, for the continued support and feedback. Whether you are installing your first character from the Content Hub, handing an assistant a folder to work in, or watching a portrait speak along with its voice, we hope you enjoy this release.

Browse the Content Hub at: https://hub.voxta.ai/

Voxta is available to subscribers at: https://portal.voxta.ai/