Updated the app to be model-first: the prompt now tells the model to choose activities from user context rather than list order, with a larger generation budget so it can complete valid plans. The fallback planner now ranks activities by constraint/preference fit, removes long-horizon options, and avoids screen time in recommended plans by default.
Updated the app to focus the results by removing the redundant Decision Guide, keeping per-plan effort/risk/cleanup guidance, and simplifying the footer to emphasize the current tiny model. Also cleaned up the Gradio theme warning by moving theme configuration to launch().
Made the Gradio Space more reliable and demo-ready by fixing the model load crash, switching to a faster default local model, adding progress/fallback behavior, and tightening the deployment config.