Tether open‑sources QVAC SDK
Tether released QVAC SDK, an open-source, cross-platform AI framework designed to let developers run and even train models entirely on local devices rather than central servers. The project promises a local-first path for building privacy-preserving or offline-capable AI apps by avoiding vendor lock-in. That makes it a practical option for side projects that need to keep data on-device or run on heterogeneous client hardware. (x.com)
Most artificial intelligence apps work like a vending machine in the cloud: your phone sends data to someone else’s server, the server runs the model, and the answer comes back over the internet. Tether’s new QVAC software development kit is trying to flip that so the model runs on the device you already own. (tether.io) That matters because “on-device” changes where the data lives. If a voice note is transcribed on your laptop instead of a remote server, the raw audio never has to leave the machine in the first place. (qvac.tether.io) A software development kit is the box of parts developers use to build an app without starting from bare metal. QVAC’s pitch is that one kit can target Linux, Mac operating system, Windows, Android, and iPhone operating system instead of forcing a team to build five separate stacks. (github.com, docs.qvac.tether.io) Tether says QVAC is fully open source, which means the code is published for anyone to inspect, modify, and reuse under its repository terms. The public GitHub repository describes it as a local-first, peer-to-peer system for language models, speech tools, retrieval systems, and other artificial intelligence tasks. (tether.io, github.com) “Peer-to-peer” is the part that makes this more than an offline app. QVAC says a device can hand work to other devices on the network, which is a way to spread computation around instead of pushing everything into one company’s data center. (github.com) The unusual claim is not just local inference, which means running a model, but local training too, which means updating a model with new examples. Tether said on April 9, 2026 that QVAC is designed to run, train, and evolve intelligence across devices and platforms. (tether.io) The framework is also aimed at mixed hardware, not just top-end graphics cards from Nvidia. Cointelegraph reported that Tether pitched it for smartphones, consumer hardware, and non-Nvidia graphics processors, which is a direct shot at the idea that useful artificial intelligence must live inside expensive server clusters. (cointelegraph.com) Tether has been building toward this for months. In May 2025 it announced QVAC as a platform for artificial intelligence agents on user devices, and in December 2025 it launched QVAC Health, a health dashboard that processes wearable data locally and offline. (tether.io, cointelegraph.com) The developer-facing pieces are already visible. The documentation says the main software development kit exposes QVAC capabilities through one interface, and the repository says it can run on Node.js, Bare runtime, and Expo for cross-platform app development. (docs.qvac.tether.io, github.com) This does not mean every phone will suddenly train giant frontier models. It does mean smaller models for transcription, translation, search over your own files, or private assistants can be built without renting an application programming interface from a cloud vendor every time a user taps a button. (github.com, qvac.tether.io) That is the bet inside this launch: if developers can ship one app that works across laptops and phones, keeps data on the device, and still borrows compute from nearby peers when needed, then local artificial intelligence stops being a demo and starts looking like normal software. (tether.io, github.com, qvac.tether.io)
