Tether Data AI Research has released an update to QVAC SDK 0.13.0, which adds support for local video generation, brain-computer interfaces (BCI), and robotics models. A key innovation was the integration of the Wan 2.1 model to create video from images without transmitting data to third-party services, and the pi0.5 model, which is called the "robot brain" by the company - it allows multimodal control systems to be run on one consumer video card.

The update also includes a new whisper.cpp-based BCI module for local translation of neural signals into text, support for OpenAI-compatible APIs and a tool to create keyboard applications for macOS, Windows and Linux in one team. Other improvements include the accelerated GPU transcription of Whisper on Android, improved Parakeet voice models for mobile devices and full-featured GPU acceleration of the Supertonic TTS speech synthesis system via Metal, Vulkan and OpenCL.

Release continues Tether’s strategy to develop a local AI infrastructure without relying on cloud services. Previously, the company had already introduced a local QVAC-fabric-llm Large Language Model Advanced Learning system, QVAC Health’s platform for personal health data analysis and the QVAC Genesis II dataset covering 148 billion tokens in 19 educational domains.