Repository navigation
The model knows where it is
Chat opens with a few lines from Chaty (on by default, Settings → Chat): which model is
running and on which engine, the computer it runs on, what Chaty can do and where, and
how its reply will be shown. Asked who it is, a model gives the name of what is running;
it points to the + menu instead of offering to look up what it cannot reach.
A thought can be cut short
While a chat reply is thinking, Skip appears at the end of its Thinking line when the
pointer is over the thought: the thought ends where it stands, kept as it was written,
and the model answers from what it had thought so far — on both engines. A limit in
Settings → Chat, off by default, does the same by itself once a thought runs that long;
an experimental floor, the other way round, keeps a model thinking at least that long.
A thought going round in circles is closed the same way, in chat and in Code, and
Code's think limit now closes the thought and lets the step act, as chat's does.
More image models, ready from the store
Most popular text-to-image models now download with what they need beside them, as
Qwen-Image and Z-Image did: FLUX.3 Image Klein, MiniMax-H3-Image, P-H3-Image, Krea 2,
MOSS-Image, FLUX.2 [dev], ERNIE-Image, Anima, Ovis-Image, LongCat-Image, Iris-3B,
Qwen-Image Edit, Stable Diffusion 3.5 and more, each set up as its makers run it. The
store reads what a model is from the file itself, so fine-tunes under names of their own
get their family's files too.
One thing at a time
A Code session at work can be left to read other sessions or chat; nothing can be sent,
and the model can be neither switched nor unloaded, until it is done — a line over the
box says where it is. Each session's row says where it stands: a light over the running
one's title, an accent bar by one waiting on you, an unread title on one that finished
while you were away. Downloads go on with the store or Set up for me closed, show at the
top of the store wherever they were started, and fill a hairline under the model chip.
A model load can be stopped at any point, from beside the chip, and holds chat's send as
it holds Code's.
Code mode
The first Qwen3 models keep thinking through a long task, and a few steps gone round
with nothing changing are pointed out, then paused. A screenshot no longer breaks a
session; a small model that sends the same plan again is handed the look at the
workspace its plan began with; one reading a file again for something not in it is
told where that is. With tool calls folded, the view keeps following a run to its end.
Nothing stutters, nothing jumps
A page's picture is made out of sight and kept across restarts; long code, the first
diagram or formula, and badly broken diagrams no longer hitch the window. Loading a
model names it on the chip as it goes, and an MLX model's bar keeps moving to the end
instead of standing at half; the sidebar slides; folded code, thoughts and
Code mode's cards open and close smoothly, by a click or by themselves; Code mode's
thought fades in word by word, and a long one shows under its run as it is written.
Code mode's welcome holds still while a model loads.
Install
| Platform | File |
|---|---|
| Windows x64 | Chaty_*_x64-setup.exe — per-user installer, no admin |
| macOS (Apple Silicon) | Chaty_*_aarch64.dmg |
| Linux x64 (beta) | Chaty_*_amd64.AppImage — chmod +x and run |
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.appThen open Chaty normally. (Alternatively: try to open it once, then go to
System Settings → Privacy & Security → Open Anyway.)
Full Changelog: v2.4.2...v2.4.3