dsh-meal-picker

Scrolled through a delivery app for fifteen minutes only to order the same dish as yesterday? That is the exact problem dsh-meal-picker solves. A draggable bubble sits in the bottom-right corner of your screen; one left-click pulls three dishes from a local library of 4,423 entries using weighted randomness. The whole pick runs offline, costs zero tokens, and never calls the model. The first two slots are guaranteed mainstream favourites; the third leaves a little room for discovery.

Need more than a list? Three AI buttons sit on each dish card: generate a home-cooking recipe, estimate the calorie count (explicitly labelled as an estimate), or hear a review from Da Fei Yu, a cute food-obsessed whale who has her own take on everything. Those three are the only actions that spend tokens.

Who is it for? Anyone who runs DSH Desktop for daily model work and does not want a separate meal-planning app. The dish library works fully offline, favourites and dietary exclusions persist in local JSON files that survive browser data resets, and once you set your city, cuisine, and meal window the bubble gets more personal with every pick. Zero runtime dependencies, pure ESM, no build step-restart and it is ready.