In September 2026 TypeSafe shipped Jev, a model that doesn't write text: it just points a finger and picks. Answers land in 70 to 500ms, at a price that changes what's worth automating. I built these prototypes to find out how far that goes.
text to color palette
Describe a scene or a mood and Jev weighs in color by color, building the palette and deciding how many colors it needs
sentiment analysis on the fly
Type anything and Jev measures sentiment, intensity and irony as you go, with no model training and no dataset
semantic icon search
Say what you need and Jev picks the tags that make sense, ranking 3,478 icons into a short list in three calls