Tether open‑sources QVAC SDK

Most artificial intelligence apps still work like a remote control: your phone sends the hard part to a data center, waits for an answer, and hopes the connection stays up. Tether’s new QVAC software development kit is trying to flip that so the model runs on the device in your hand or on a nearby machine instead. (tether.io) That idea is called on-device artificial intelligence, and the pitch is simple: keep the data local, cut the round trip to the cloud, and keep working when the internet is weak or absent. Tether says QVAC is open source and built for Linux, macOS, Windows, Android, and iOS from one cross-platform stack. (docs.qvac.tether.io, github.com) A software development kit is a box of prebuilt parts, like buying a kitchen with the cabinets already assembled instead of cutting every board yourself. QVAC’s kit includes tools for large language models, speech, retrieval-augmented generation, and other local artificial intelligence tasks. (github.com, docs.qvac.tether.io) Retrieval-augmented generation is the trick that lets a model look up your own files before it answers, like an open-book exam instead of pure memory. QVAC’s documentation says developers can wire that into apps that run across phones, desktops, and servers without swapping to a different cloud provider’s application programming interface for each platform. (docs.qvac.tether.io) The unusual part is that Tether is not pitching this as a chatbot website. The company says QVAC can also delegate inference, meaning one device can hand off a model run to another peer, using built-in peer-to-peer features instead of a central server. (github.com, qvac.tether.io) Peer-to-peer means the devices talk directly, like laptops sharing files on the same local network instead of uploading everything to one company’s warehouse first. Tether’s project page compares that design to systems like BitTorrent and the InterPlanetary File System, but aimed at artificial intelligence workloads. (github.com) Under the hood, outside reporting says QVAC Fabric is based on llama.cpp, the widely used open-source engine for running compact language models on local hardware. That matters because it places QVAC closer to the “run smaller models anywhere” camp than to the “rent giant cloud clusters” camp. (pulse24.ai, decrypt.co) Tether announced the release on April 9, 2026, and tied it to a bigger push beyond stablecoins into artificial intelligence infrastructure. The company’s own release says it plans to keep funding the QVAC ecosystem in areas including robotics and brain-computer interfaces, which is a much wider ambition than a single developer tool. (tether.io) For developers, the practical appeal is less about slogans and more about demos that survive outside a notebook. If one codebase can run the same artificial intelligence feature on an iPhone, a Windows laptop, and a Linux server, that is the difference between a prototype on your screen and a product someone else can actually install. (docs.qvac.tether.io, github.com) The open question is whether QVAC becomes a real developer standard or just another toolkit in a crowded local-artificial-intelligence pile. But the launch makes one thing concrete: Tether wants a place in the layer where models are packaged, moved, and run across everyday devices, not just in the business of issuing digital dollars. (siliconangle.com, tether.io)