Tether releases QVAC SDK
Tether announced QVAC SDK, pitching it as a universal AI building block for running, training and evolving intelligence across devices and platforms. (x.com) The company frames the SDK as cross-device foundational tooling, signalling competition in the lower-level software stack that supports AI workflows. (x.com)
Most artificial intelligence software still assumes your data will leave your device, travel to a company server, and come back as an answer a second later. Tether’s new QVAC software development kit is built around the opposite idea: keep the model and the work on the phone, laptop, or server you already control. (tether.io) A software development kit is the box of parts developers use to build an app, the way a plumbing kit gives a builder the pipes and joints before the house exists. Tether says QVAC is that box of parts for local artificial intelligence, and released it on April 9, 2026 as fully open-source software. (tether.io) The basic bet is that one codebase should run across Linux, macOS, Windows, Android, and iPhone without being rewritten for each machine. QVAC’s documentation says developers can run language models, speech tools, and retrieval systems across all five operating systems from the same JavaScript and TypeScript stack. (docs.qvac.tether.io) Tether is also pushing peer-to-peer networking, which means devices talk directly to each other instead of routing every task through one central company server. QVAC’s main site says that cross-platform and peer-to-peer design is the core of the system, not an extra feature bolted on later. (qvac.tether.io) That pitch did not start this week. Tether introduced QVAC in May 2025 as a platform for running artificial intelligence agents on user devices instead of in big data centers, and the software development kit is the piece that turns that idea into something outside developers can actually build with. (tether.io) Since then, Tether has been filling in the stack underneath the new kit. In December 2025 it launched QVAC Fabric large language model tools for inference and fine-tuning on consumer graphics cards, laptops, smartphones, and servers, and in March 2026 it added a framework for training Microsoft BitNet models with low-rank adaptation on those same kinds of devices. (tether.io 1) (tether.io 2) Low-rank adaptation is a way to customize a model by changing a small add-on layer instead of retraining the whole thing, like swapping a bike gear instead of rebuilding the bicycle. BitNet is Microsoft’s line of one-bit language models, designed to use much less memory than standard models, which is why Tether keeps pairing it with phones and everyday graphics hardware. (tether.io) The new software development kit pulls those pieces behind one interface. QVAC’s addon documentation says the kit “unifies all addon capabilities under a single interface,” adds built-in peer-to-peer connectivity, and lets developers use QVAC beyond Tether’s own Bare runtime. (docs.qvac.tether.io) There is also a compatibility play in the launch. Tether says the command-line tool can expose an OpenAI-compatible application programming interface, which means developers with software already built around OpenAI-style calls may be able to swap in local QVAC components without redesigning the whole app. (docs.qvac.tether.io) Tether is best known for the USD₮ stablecoin, not developer tools, which is why this launch stands out. The company’s own site now lists more than 20,000 dedicated graphics processing units in its artificial intelligence operations, and QVAC shows Tether trying to turn that infrastructure push into a software platform that competes lower in the artificial intelligence stack, where developers choose the runtime before users ever see the app. (tether.io)