🐾 NekoTask

NekoTask is a playful yet practical to-do list app with an AI-powered assistant 🐱. It was originally designed to support people with ADHD (adults or children, any age) by reducing cognitive load in task management — but of course, everyone can use it to stay organized.

⚠️ Important Note: At this stage, NekoTask is a functional prototype.

• Tasks are loaded from a local JSON file (app/src/main/assets/todolist_items.json).
• The AI chat logic is implemented, with both a mock and a real client.
• The real client requires an OpenAI API key to function.

---

✨ Features

• Dual Interaction Modes
  • 📝 Manual Mode: Add tasks via a clean, minimal bottom card.
  • 🤖 AI Chat Mode: A functional chat interface to add, update, and delete tasks using natural language.
• ADHD-Friendly UI Principles
  • Minimal, distraction-free interface.
  • Reduced decision fatigue — AI agent simplifies task entry.
  • Playful cat theme for motivation.
• AI Subtask Decomposition (Core Idea)
  • Ask the assistant to break a big task into small actionable subtasks.
  • Uses a strict JSON action contract to add tasks (optionally referencing a parentId).
• Due Date Picker (Two-step)
  • Material 3 DatePicker + TimePicker with a two-step flow (date → time).
  • Quick visibility of due-state; button indicates selection.
• Chat Overlay Modes
  • Peek bubbles over the list, or switch to fullscreen chat.
  • Pin/auto-dismiss behavior to reduce noise.
• AI Notepad (Working-Memory Notes)
  • Quick, lightweight notes for facts/clues/context that are not actionable tasks.
  • Pin important notes; optionally link a note to a task; convert note → task when needed.
  • Clean separation from the checklist to avoid clutter.

📸 Screenshots

(to be added — current UI demo with AI chat + manual add card)

🛠️ Tech Stack

• Android (Jetpack Compose)
• Kotlin
• Material 3 design
• Ktor for networking
• OpenAI GPT API integration
• kotlinx.serialization for robust JSON parsing
• kotlinx.datetime for time handling

🔑 Setup

To use the AI features, you will need to provide your own OpenAI API key.

1. Create a local.properties file in the root directory of the project if you don't already have one.
2. Add your OpenAI API key to the local.properties file:

   openai_key="YOUR_API_KEY"

3. Make sure local.properties is in your .gitignore file to prevent your API key from being committed to version control.

🚧 Current Status

• ✅ UI for task lists, AI chat bubbles, and manual add card.
• ✅ Sample cat-themed task data for testing.
• ✅ Tasks loaded from a local JSON file.
• ✅ AI backend integration with mock and real clients.
• ✅ Two-step due date selection (date → time) wired to task creation.
• ✅ Strict JSON contract for AI responses (say + actions) with parsing and safety checks.
• ❌ No persistence layer yet (no database; in-memory only).

---

🎯 Why NekoTask?

Many productivity tools are overwhelming — too many buttons, features, and settings. For users with ADHD, this can create friction instead of support.

NekoTask’s goal is to:

• Keep manual controls minimal.
• Provide AI-assisted task management.
• Build a friendly, motivating environment with playful design.
• Turn big, vague tasks into small, executable steps with AI subtask decomposition.

---

💡 Roadmap (Next 10 days focus + beyond)

Near-term (aimed for the current sprint)

• Subtasks S1 (in-memory):
  • Task model gains parentId (flat storage; UI groups by parent).
  • Expandable subtasks under each parent with inline add and quick complete.
  • Parent progress indicator (e.g., 2/5).
• AI contract update:
  • add_task supports optional parentId for AI-generated subtasks.
  • Keep strict single-JSON envelope (say + actions) for reliability.
• Time presets:
  • Quick presets like Tonight / Tomorrow / This week alongside the two-step picker.
• Configurable AI Persona:
  • Presets: Neutral (default), Coach, Cheerful Cat, Minimalist.
  • Persona influences only say tone, never actions schema.
• Energy level (Low/Medium/High):
  • Simple self-rating chip; drives “next small step” granularity and optional UI contraction (show only next actionable).
• Privacy toggle for AI snapshot:
  • Off by default. When on, send a minimal, non-sensitive task snapshot in a system message.
• AI Notepad MVP (in-memory):
  • New Note model and in-memory repository; CRUD + pin/unpin; optional link to taskId.
  • Notepad UI (sheet or screen) and a small pinned preview strip on Home.
  • Convert note → task (simple UI action); keep notes separate from the checklist.
• Chat + Notes integration:
  • “remember …/记一下 …” maps to add_note; retrieval answers appear in say.

Infrastructure

• Local persistence with Room (entities/DAOs, v1 schema; replace JSON-only).
• Compose Navigation + Task Details screen (edit title/notes/due/priority; manage subtasks).
• Room for notes (separate notes table) and repository abstraction.

Assistant & Privacy

• Privacy toggle: choose whether to send a current-task snapshot to the AI (with clear copy; default off).
• Confirmation gates for destructive actions (bulk delete/complete) and guardrails around overdue bulk ops.
• Separate privacy toggle for notes snapshot (default off) with simple redaction.

Reminders & Input

• Gentle reminders with WorkManager (due and overdue notifications; rate limited, quiet hours friendly).
• Focus 20-min mode (Pomodoro-like) with gentle start/stop cues.
• In-app voice input for quick add; App Shortcuts for “Quick Add”.

Later

• Accessibility improvements (large touch targets, high contrast, single-hand mode).
• Optional light gamification (streaks/stickers) — low-arousal, can be turned off.
• Advanced search/filter/sort; calendar views.
• Wearable and assistant integrations (Android first).

---

🤖 AI Contract (Brief)

The real client sends a system prompt requiring the assistant to reply with a single compact JSON object:

{
  "say": "string",
  "actions": [
    { "type": "add_task", "title": "string", "notes": "string?", "dueAt": "ISO-8601?", "priority": "LOW|MEDIUM|HIGH|DEFAULT?", "parentId": 123 },
    { "type": "complete_task", "id": 123 },
    { "type": "delete_task", "id": 123 },
    { "type": "update_task", "id": 123, "title": "string?", "notes": "string?", "dueAt": "ISO-8601?", "priority": "LOW|MEDIUM|HIGH|DEFAULT?", "parentId": 123 }
  ]
}

Responses are parsed and mapped to app actions. Mock and real clients are both supported.

CURRENT_TODO_STATE (Nested)

The app also sends a nested snapshot of current tasks as a system message. Structure:

{
  "now": "ISO-8601",
  "unfinished": [
    {
      "id": 1,
      "title": "Parent Task",
      "status": "OPEN",
      "priority": "HIGH",
      "dueAt": "2025-08-25T10:00:00Z",
      "children": [
        { "id": 2, "title": "Child A", "status": "DONE" },
        { "id": 3, "title": "Child B", "status": "OPEN" }
      ],
      "totalChildren": 2,
      "doneChildren": 1,
      "progress": 0.5
    }
  ],
  "finished": [
    { "id": 4, "title": "Completed Parent", "status": "DONE", "children": [] }
  ],
  "finished_count": 1
}

Rules for the assistant (enforced via the system prompt):

• Use ids from CURRENT_TODO_STATE at any depth (top-level or nested children).
• complete_task: only for items in the "unfinished" tree (any depth).
• delete_task: allowed for any item in either tree.
• add_task: provide parentId to create a subtask; omit to create a top-level task.
• update_task: include parentId to reparent under another existing taskd.
• If ambiguous, ask a clarifying question in say and return "actions": [].

Note actions (for working-memory notes) extend the same actions array without changing the envelope:

{
  "say": "string",
  "actions": [
    { "type": "add_note", "text": "string", "taskId": 123, "pinned": true },
    { "type": "update_note", "id": 1, "text": "string", "pinned": false, "taskId": 123 },
    { "type": "delete_note", "id": 1 },
    { "type": "pin_note", "id": 1, "pinned": true }
  ]
}

---

🔬 Research Notes (for the dissertation)

• Primary research questions (examples):
  • Does AI-assisted subtasking reduce “time-to-start” (time to complete the first subtask)?
  • Do subtasks + progress visualization improve completion rate and reduce overdue rate?
  • Does a strict JSON contract increase action reliability (parse success and execution rate)?
• Suggested metrics:
  • Parse success rate; action execution success; undo rate.
  • Start time; completion/overdue rates; clicks per operation.
  • Short subjective scales (SUS, simplified NASA-TLX).
• Ethics & privacy:
  • No medical claims. Use “support level” language, not diagnosis.
  • Snapshot-to-AI toggle defaults to off; users can opt in.

---

🎓 This project is currently developed as part of a Master's dissertation — and because I also live with ADHD and depression, I've taken it personally.