Aion Instruct Preview Chat
A WinUI 3 sample chat app from Microsoft that runs Windows' on-device Aion Instruct Preview language model on Copilot+ PCs, streaming tokens with first-token latency and tokens-per-second shown under each reply.
Aion Instruct Preview Chat is a WinUI 3 desktop sample app demonstrating on-device language model inference with the AionInstructPreview framework on Windows Copilot+ PCs. It streams tokens into the chat bubble as they generate and shows first-token latency and tokens-per-second under each reply, though the README notes these preview numbers are not final performance.
A Bootstrap.ps1 script downloads and installs the model framework from the repository's public GitHub releases, prepares the QNN execution provider, and builds and launches the app; the first launch compiles the model for the device's NPU and takes a few minutes, with later launches near-instant.
This preview runs only on ARM64 Copilot+ PCs (Snapdragon, QNN NPU) with Windows 11 Developer Mode enabled; there is no CPU fallback, and x64 support is planned but not yet available.