Tether unveils QVAC SDK, an open-source framework for on-device AI across devices and operating systems

Tether’s QVAC team has announced the release of QVAC SDK, an open-source software development kit positioned as a “universal foundation for artificial intelligence.” The company frames the initiative around a future environment—described as roughly 2100–2150—where 10 billion people live alongside 10 billion autonomous machines and a trillion AI agents, under what it calls the “Stable Intelligence Era.”

What QVAC SDK is designed to do

In a blog post, Tether describes QVAC SDK as a modular, high-performance, device-centric AI environment intended to work reliably across “virtually any hardware or software setup.” The platform is designed to scale with hardware advances, remaining adaptable for decades or even centuries as silicon technology progresses.

The SDK is built to let developers construct, deploy, and refine AI models directly on end-user hardware while maintaining consistency. Tether says applications built with QVAC SDK can run AI tasks—such as large language models and other core components—across consumer devices including smartphones, laptops, and desktops, as well as enterprise-level systems.

Cross-platform operation and local execution

Tether says the same code runs without modification on iOS, Android, Windows, macOS, and Linux, aiming to reduce the need for platform-specific rewrites or workarounds. The company also emphasizes that the resulting ecosystem supports private, user-owned intelligence that operates independently of external servers.

According to Tether, routine AI-powered tools—such as writing support, real-time translation, voice-to-text, image creation, financial tracking, document summarization, and intelligent search—can execute instantly on personal devices. It adds that keeping data local is intended to reduce risks associated with cloud transmission and help apps remain functional during internet outages or server disruptions.

Core technology and capabilities

Tether says the SDK’s interface is based on “QVAC Fabric,” described as a specialized adaptation of the widely used llama.cpp engine. It also integrates local tools for speech recognition, including whisper.cpp and Parakeet, and on-device language translation via Bergamot.

The update states that capabilities include text generation, embeddings, visual analysis, optical character recognition, speech synthesis, and more, all accessible through a single API.

Decentralization and peer-to-peer features

Decentralization is presented as another cornerstone of QVAC SDK. Tether says the SDK leverages the Holepunch framework to embed peer-to-peer features for sharing models and distributing computational tasks. It also points to upcoming updates that would enable collaborative swarms for training and inference without centralized servers.

Tether adds that these functions are intended to work transparently across supported platforms, supporting resilient applications that can operate even in disconnected environments.

Leadership perspective and next steps

Paolo Ardoino, Tether’s CEO, is quoted emphasizing that traditional reliance on central servers cannot handle the scale of an AI-saturated world, citing physical limits such as signal latency and system vulnerabilities. He argues that QVAC is purpose-built for a decentralized reality ahead.

Tether also reports plans to dedicate major resources to expanding the open-source community around QVAC, including adding specialized modules for areas such as robotics and brain-computer interfaces. The company says the focus on on-device processing is intended to meet demands for speed, security, and autonomy as AI becomes increasingly integrated into physical infrastructure and everyday activities.