Tether has introduced the QVAC SDK, a toolkit designed for developing and deploying artificial intelligence directly on user devices.
Tether Launches QVAC SDK as the AI Universal Building Block that Runs, Trains, and Evolves Intelligence Across any Device and Platform
Learn more: https://t.co/n7mkjkD8cz— Tether (@tether) April 9, 2026
According to the company, the open-source software kit allows for the creation and further training of AI models on devices ranging from smartphones to servers, independent of cloud infrastructure.
Solutions developed with the SDK can operate on iOS, Android, Windows, macOS, and Linux with a single codebase, eliminating the need for adaptation to specific operating environments.
Moving Away from Centralized AI
The company introduced the QVAC platform for decentralized artificial intelligence in May 2025. Tether noted that the current centralized AI model is not scalable due to latency, infrastructure vulnerabilities, and concentrated control.
As an alternative, the company proposed a concept of local deployment, where computations occur on the user’s device, and data is not transmitted to external servers.
This approach aims to ensure:
- higher privacy levels;
- low latency;
- resilience to failures and internet outages.
“Applications based on QVAC continue to function even with low connectivity speeds, making AI more practical in real-world scenarios. If the internet goes down or a server farm fails, nothing changes for the user,” the developers explained.
Technical Architecture
The SDK is built on QVAC Fabric—a modification of the minimalist llama.cpp engine for running local models. It supports full compatibility with the Llama ecosystem for text generation, embedding, and multimodal workloads.
The platform also integrates several standalone mechanisms such as Parakeet (speech recognition) and Bergamot (translation). These are unified by a single API, allowing developers to combine functions without altering application logic.
QVAC SDK includes built-in P2P interaction mechanisms based on the Holepunch stack, enabling:
- distribution of models without centralized servers;
- use and future training of AI in a distributed network.
Tether positions QVAC SDK as a foundational element for a new class of applications—from personal AI assistants to autonomous agents integrated with Web3 services.
“The world is approaching a point where billions of people will share the planet with billions of autonomous machines and trillions of AI agents. The current structure, where every decision is made through a centralized server, will not scale to meet this reality,” stated Tether CEO Paolo Ardoino.
Back in December 2025, the QVAC team introduced an open dataset for AI training.