A simpler Decisions tab, and screen picks that can see
Decisions now has three cards: how agents compact, which model decides, and each job it does with its trigger and cost. Screen picks can send Jev a screenshot.
Compaction is always on. One switch says who chooses what stays: the Decisions model, which keeps the parts that matter word for word, or the agent's own chat model, which writes a summary the way OpenCode does. Compact at sets when it happens.
Every job the Decisions model does while agents work now sits in the same tab, each with what it does, when it runs and what happens when you switch it off, next to the cost sheet and the screen-pick benchmark.
Screen picks can send a picture of the app window to a model that reads images, such as Jev 1.13 and 1.14. Models that read text only, like Laya, cannot be given one, and a picture the model refuses falls back to the same question as text.