FAQ

Model downloads, GPU, failed updates, full offline use — the eight questions we hear most.

Model download fails or is slow?

Models come from Hugging Face by default; on a poor connection, switch the download source to ModelScope in Settings — both hosts serve identical content. Then restart the download.

The GPU isn't being used?

Built-in local inference enables the GPU (Vulkan) by default. If speed looks wrong: update your graphics driver; check the model size fits your VRAM (models that don't fit fall back to system memory). The diagnostics report (below) shows the current inference device.

In-app update fails?

Updates require fully exiting the app first (including the tray icon). If the installer reports "can't write", install manually: quit the app, run the new installer, and verify integrity against SHA256SUMS.txt (see Installation).

Can it run fully offline?

Yes. Once a local model is downloaded, core features (chat, files, commands, browser) all run locally with the network unplugged; only update checks need connectivity (ignorable). Cloud models and optional web search do need the network.

Where is my data stored?

Everything stays in the local data directory (%APPDATA%\ChengYoung): conversations, session history, settings, model files, and extension configuration. Deleting that directory erases it all. See Privacy & data.

How do I report a problem?

Settings → App → generate a diagnostics report (versions, system info, error summary), copy the text and send it with your description. The report contains none of your conversations.

What's the relationship between ChengYoung and the upstream open-source project?

ChengYoung is an independently branded distribution built on the upstream open-source project (Apache-2.0): the upstream code is rebranded and shipped as compiled artifacts, unaffiliated with the upstream project. Sources, attribution, and modifications: Provenance.

Which operating systems are supported?

Windows 10/11 (x64) installers are available today; other platforms are not provided yet.

This page is adapted from upstream open-source documentation (Apache-2.0) with modifications; see /legal/provenance for sources and changes.