QV.AC
by Tether Data
Tether's flagship local AI assistant — runs open models on your hardware, extends via agents and skills, syncs peer-to-peer with nothing in the middle.
Download
QV.AC is the reference implementation of what QVAC is for. It began life as QVAC Workbench, released in October 2025 alongside the Genesis I dataset, and was rebuilt into a consumer assistant with a new design, an agent system, and a skills layer.
The app runs state-of-the-art open models directly on the device with no internet connection required. It carries a large skills library — calendar, tasks, weather, notes, music, among others — so the assistant can act rather than just answer. Vision and voice transcription run on-device, and your own files and notes can be used as private knowledge the model reads from.
Its most distinctive feature is delegated inference. Your phone can borrow your laptop’s compute over a peer-to-peer connection, letting a mobile device run models it could not host on its own. There is no server involved in the handoff. Two devices find each other over Hyperswarm and one does the work for the other. Nobody else in this space ships that.
There are no accounts. Sync between your devices is encrypted and peer-to-peer.
Note on naming
The Google Play package is io.tether.qvac.workbench, and Tether’s developer documentation still refers to the desktop application as QVAC Workbench. QV.AC and Workbench are the same product under two names.
What it uses
Text generation, transcription, multimodal inference over text and images, and retrieval over your own documents. Delegated inference and peer-to-peer sync run on the Holepunch stack.
Status
Labeled Alpha on Tether’s own site.