Fangyuan025/Chaty v2.3.9

on GitHub

Engines that update without a Chaty release

Settings → Runtimes lists llama.cpp, stable-diffusion.cpp, audio.cpp and MLX
with the build each is on, and a newer one to switch to at a click — or by
itself, once you turn that on. Every day each engine's newest upstream is
built into this release's own wrappers and published only if it starts; the
list is signed, and a newer engine is checked and started once on your
machine before it is used. One that will not start is set aside and the
bundled one takes over by itself.

llama.cpp in a process of its own

GGUF models now run in chaty-llama, a helper beside the app, as the image,
music and MLX engines already did. That is what lets it be replaced, and a
crash inside llama.cpp ends that process instead of the window. Loading shows
llama.cpp's own progress.

MLX on upstream

MLX runs on upstream mlx-swift-lm and mlx-swift 0.33 instead of our fork:
upstream's own Qwen3.5 MTP support made it unnecessary. Speculative decoding
stays as fast — Qwen3.5-4B MTPLX at 69 tokens a second against 40 without it
on an M4 Pro.

Install

macOS first launch — "Apple could not verify…" / "damaged"

Chaty is ad-hoc signed but not notarized (no paid Apple Developer
account), so Gatekeeper flags it on first launch. The app is safe —
everything runs locally. Clear the download quarantine once, in Terminal:

xattr -dr com.apple.quarantine /Applications/Chaty.app

Then open Chaty normally. (Alternatively: try to open it once, then go to
System Settings → Privacy & Security → Open Anyway.)

Full Changelog: v2.3.8...v2.3.9