Voxta v1.4.0: OpenAI API Compatibility & Elite Dangerous Integration
Hello everyone,
Voxta v1.4.0 is now available. You can download the latest release at the Voxta Portal.
This update introduces an OpenAI-compatible API to use Voxta as a backend, a full Elite Dangerous game integration, and several system fixes.
Here is the full list of changes in this build:
✨ New Features & Integrations
OpenAI Compatible API (Voxta as a Backend)
You can now generate your own Voxta API key and use it to power external clients with Voxta as your backend. Currently, this API exposes standard /v1/chat/completions and /v1/models endpoints for text completions only. We plan to expand this later.
Elite Dangerous Integration
This is a full integration with the game Elite Dangerous. The AI has access to live ship telemetry data, reacts to critical in-game events, and can use actions to control specific ship functions. We will publish dedicated characters for Elite later.
Dual Vision Capabilities
Added a new 'Both' option for vision sources, enabling Voxta to simultaneously capture both your Webcam and Screen.
⚙️ UI and Experience Updates
Refined Selection Screens: Updated the selection screens for Augmentations to improve visibility. Added new icons for the Chat Style selection.
Voice Selector Refactor: Overhauled the Voice Selector UI. Service identification is clearer, and truncation issues have been fixed.
Assets Folder: The Assets folder is now created automatically on demand when the tab is opened.
🛠️ Key Fixes & Technical Enhancements
Server & Modules:
LlamaSharp: Updated to version 0.26.0.
AssemblyAI: Added keyterms prompting support.
OpenAI Compatible Module: Added the ability to pass custom headers and body parameters.
Prompting: Implemented structural changes to the text generation prompt.
Bug Fixes:
ExLlamaV3 Installation: Fixed a breakage caused by setuptools by pinning setuptools < 82.0.0.
Memory & Chat Crashes: Fixed a CharacterMemoryInstance.InitializeMemoryAsync crash that occurred when processing previous chat memories. Fixed a ThinkPass crash that triggered on the first message if the chat history was empty.
Audio Queue Deadlock: Fixed a deadlock in the audio queue that occurred if speech was interrupted while an image was generating.
Visual UI Fixes: Fixed a UI jump when rolling over video character cards if the center positioning didn't match. Fixed chat history sorting. Resolved an issue where the default avatar went blank on non-automatic settings. Fixed a bug where the image gen playground prompt wouldn't send if the editor wasn't blurred.
Thank you to everyone testing and reporting bugs.
Comments (11)
Since it has been roughly 2 months now since V1.4, just curious if there is any future updates planned on being released soon-ish? All your new releases so far have been spot on, hoping for another one soon!
Yes, we're working hard in multiple directions right now. The next build should include a basic Voxta Android app, along with many server improvements. We're also releasing the Minecraft companion client as free and open source, and putting a lot of work into the VAM plugin and integrations. So yeah, we're still here - just handling a lot at once. You can join our Discord to stay in the loop. https://discord.gg/voxta
After installing the desktop version, the last step prompts "Connection refused, unable to connect to 127.0.0.1:5001". What should I do?
how can i download the vam voxta scene
On our Discord server. We can't host them on Patreon, since our Patreon page is SFW.
Hi I tried various Qwen3.5 models locally with Llamacpp (ex Qwen3.5-9B.Q8_0.gguf) but can't run them (error 500) with the new version, even after update. Is it normal? Is there a workaround? I've tried without Voxta with llama-server.exe and it runs fine.
Most likely the LlamaSharp module is still lagging behind in supporting the new Qwen 3.5 models. There's not much we can do on our side, you'll need to follow their updates. If they release a new version with Qwen 3.5 support, we will update the module accordingly.
Using your installation package, with a speed of 10M, do you live in Africa? Was it still in the last century? I directly downloaded the 100M version of the HUGGINFACE model. However, when I placed the complete model with the best quality in your directory, an error occurred. Could it be that the model downloaded from your installation package contains some special code?
Voxta downloads models directly from Hugging Face's servers. Download speeds depend entirely on Hugging Face's CDN and your internet connection. we don't host or proxy the files ourselves, so we have no way to control or improve that. About replacing the model manually: Voxta expects a specific model file (exact format, quantization, and filename). If you downloaded a different version from Hugging Face (e.g., a different size, quantization level, or file format), it won't be compatible even if it's from the same model family. There's no "special code" in the files, it's simply that the wrong model variant was used.
Damn fellas, ED integration?! You are making dreams come true for this pilot. 07
Elite and Dangerous integration is a nice surprise.