dAIly: AI Planning Agent for My Friend.

This is a project submitted for the Hacktoberfest Weekend Challenge: Build for a Friend. The creator, Kushagra, is a third-year computer science student who wants to improve his machine learning skills and competitive programming abilities. He struggles to balance his academic responsibilities, assignments, mid-semester exams, and travel. To help him manage his time and energy, the developer built dAIly, a local desktop planning agent.

dAIly is designed to discuss goals, break them into smaller tasks, and create a day plan using the user's class timetable, deadlines, free time, and energy levels. Users can add new assignments or adjust priorities, and the agent will generate a new plan accordingly. After a focus session, users can record their actual progress.

The app uses Gemma 3 4B running locally through Ollama and is built with Electron, React, and TypeScript. SQLite stores all the necessary data like goals, subtasks, plans, conversations, and focus-session history. dAIly allows users to control the suggestions and does not silently apply them, ensuring they remain in charge of their schedule.

The hardest part of developing dAIly was making the agent understand when and why it should modify a plan, goal, or subtask. The developers improved Gemma's model instructions, response formats, checks, saved records, and scheduling code to address this issue.

To maintain privacy and avoid incurring API charges, the app runs the model locally. This also means that only a small amount of hardware and electricity is required. All of the app's code is MIT licensed, and the developers can easily inspect and modify the model instructions, checks, and planning behavior. This open innovation approach enables the creator to customize the agent to suit Kushagra's specific needs.