A modern desktop assistant that watches your screen and guides you step-by-step to accomplish tasks. Think of it as having a helpful friend looking over your shoulder, pointing exactly where to click.

Ever struggled to find a setting on your computer? Or wished someone could just show you where to click? This AI Assistant does exactly that:

1. You tell it what you want to do - For example: "How do I test my speakers?"
2. It figures out the steps - The AI thinks: "First, open Settings, then System, then Sound..."
3. It looks at your screen - Takes a quick snapshot to see where you are
4. It draws a rectangle - Points exactly where you need to click
5. You click, then say "next" - And it guides you to the next step

It's like GPS navigation, but for your computer screen!

Just type naturally like you're talking to a friend:

• "How do I change my wallpaper?"
• "I want to connect to WiFi"
• "Help me test my audio"

The assistant uses an efficient process:

• Red rectangles highlight exactly where to click
• Stays on top of all windows
• Clear and easy to see

• Modern "glassmorphism" aesthetic
• Semi-transparent frosted glass effect
• Sleek dark theme that's easy on the eyes

┌─────────────────────────────────────────────────────────────┐
│  YOU: "I want to test my speakers"                          │
└─────────────────────────────────────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────────┐
│  STEP 1: AI thinks "First step is to Open Settings"         │
│  (No screenshot needed - just logical thinking)             │
└─────────────────────────────────────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────────┐
│  STEP 2: Takes screenshot, AI says "You're on Desktop"      │
│  (Quick check - just identifies the screen)                 │
└─────────────────────────────────────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────────┐
│  STEP 3: AI decides "Click on Settings icon"                │
│  (Adjusts based on where you actually are)                  │
└─────────────────────────────────────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────────┐
│  STEP 4-6: Finds "Settings" text on screen using OCR,       │
│  draws a RED RECTANGLE around it, then STOPS and WAITS      │
└─────────────────────────────────────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────────┐
│  YOU: Click on Settings, then type "next"                   │
│  → The cycle repeats for the next step!                     │
└─────────────────────────────────────────────────────────────┘

• What it is: A smart AI from Google that can understand text and images
• What it does here: Figures out what steps you need and identifies your screen
• Why it's good: Very accurate at understanding context and giving helpful answers

• What it is: Technology that reads text from images
• What it does here: Finds where buttons and labels are on your screen
• Tool used: Tesseract OCR (free and open source)
• Why it's good: Fast and works offline once the AI decides what to look for

• What it is: A toolkit for building desktop applications
• What it does here: Creates the beautiful glass-like chat window
• Why it's good: Works on Windows, looks modern, very customizable

• What it is: An invisible window that sits on top of everything
• What it does here: Draws the red rectangles pointing to buttons
• Why it's good: Doesn't interfere with your apps, just highlights things

• What it is: A Python library for working with images
• What it does here: Takes screenshots of your desktop
• Why it's good: Fast, reliable, works with all screens

1. Python 3.8 or higher

   • Download from python.org

2. Tesseract OCR

   • Download from GitHub
   • Install to C:\Program Files\Tesseract-OCR\

3. Google Gemini API Key

   • Get one free at Google AI Studio

pip install -r requirements.txt

python circular_window.py

1. Launch the app - A small circular window appears
2. Click the window - It expands into the chat interface
3. Enter your API key - Click the settings icon (gear) to add your Gemini API key
4. Start chatting! - Type something like "how do I check my wifi?"
5. Follow the rectangles - Click where it points, then type "next"
6. Say "done" when finished - Or "cancel" to stop anytime

You: "I want to test my speakers"
AI: 🎯 Goal: test my speakers
    → Click on Settings
    [Red rectangle appears on Settings icon]

You: "next"
AI: → Click on System
    [Red rectangle appears on System]

You: "next"  
AI: → Click on Sound
    [Red rectangle appears on Sound]

You: "next"
AI: → Click on Test
    [Red rectangle appears on Test button]

You: "done"
AI: Great! Goal completed! 🎉

You: "help me change my wallpaper"
AI: 🎯 Goal: change my wallpaper
    → Click on Settings
    ...

AI-assistant/
├── circular_window.py    # Main application (all the code)
├── task_graph.json       # Predefined task templates (optional)
├── guided_task.log       # Debug log file
└── README.md             # This documentation

• Screenshots stay local - They're only sent to Google's AI API for analysis
• No data stored - Screenshots are discarded after each analysis
• API key in memory only - Not saved to disk (you enter it each session)

• UI Framework: PyQt5
• AI Model: Google Gemini
• OCR Engine: Tesseract
• Design Style: Glassmorphism (frosted glass aesthetic)

This project is licensed under the MIT License - see the LICENSE file for details.

Found a bug or have an idea? Please read our CONTRIBUTING guide before opening an issue or submitting a pull request.

Made with ❤️ to make computers easier for everyone.

• Audio Chat: Voice interaction for hands-free guidance.
• Cursor Control: Allow the AI to perform clicks for you.
• MCP Server: Integrate with Model Context Protocol to use external tools.
• Cross-Platform Support: Mac and Linux support.