Local AI: Moderator

~150 ms per message on desktop GPU / CPU (WebGL is slower).
23 languages.
HateCheck F1 0.87.
Runs locally: CUDA → Vulkan → CPU. A GPU is faster, not required.
Editor-only, Windows x86_64.
Unity 6 (6000.0) or higher.
About 12 GB VRAM. 10 GB free disk for the download.
Model files are downloaded from the editor tool (not packed in the asset).