Tether releases QVAC SDK for running AI models on personal devices
Tether has put tooling behind QVAC, its AI effort centred on running models locally on everyday hardware, according to a report by Crypto Briefing.
Toolkit and runtime for developers
Tether launched the QVAC SDK on April 9, 2026. A software development kit gives developers ready-made parts for building apps on a platform, so they do not have to write the underlying machinery themselves. Its core is a cross-platform inference runtime, the engine that generates a model's response on a given device.
The aim, as described, is for one model to operate across a laptop, a desktop and a phone without being rebuilt separately for each. Tether has named personalization and on-device execution as its two principal design priorities.
Fine-tuning on consumer hardware
A second component, QVAC Fabric, handles fine-tuning, the further training of an existing model toward a narrower task. According to the report, it works across a range of consumer hardware, including mobile GPUs, chips designed for many simultaneous calculations. Fabric builds on advances in LoRA fine-tuning, a method that modifies a small portion of a model rather than retraining all of it.
Tether has also released core components as open source under permissive licences, letting outside developers examine, alter and extend the code.
Smaller model with benchmark claims
On November 11, 2026, Tether released QVAC BotPsy, a model with 800 million parameters tuned for agentic tasks, where an AI invokes tools, carries out actions and completes multi-step jobs for a user. Initial benchmarks put BotPsy at 44.0% on the BFCL v4 test, while a 2-billion-parameter model reached 37.7% on the same test. These figures come from the initial benchmarks as reported and have not been independently verified.
Open questions
- How well do BotPsy's benchmark results hold up under independent testing?
- Which consumer devices and models will the SDK support in practice?