Run the work platform from the Rack
The engine now keeps your Rack, so it survives a relaunch, and one click runs the work platform on its Text and System One models.
The Rack is saved by the engine
What you put on each task in AI Gateway, Models, Rack is now saved by the engine on this Mac, not just in the window. Close the app, open it again, and every assignment is still there. Two windows editing the Rack at once never overwrite each other: the later one shows what was saved and says so.
Local models are now filed by what they do, so a speech or embedding model you install lands in its own category instead of under Text.
One click to put it to work
Run from the Rack turns your assignments into a work profile: the model on Text, Causal LM does each turn and the model on System One, Routing steers it. Tasks you leave empty keep the local defaults. Picking any other model opts out, and the Rack says so right away.
A change made while a turn is running is saved at once and takes effect when the turn ends, so nothing is swapped out from under work in progress.