How it works

User Workflow: Step-by-Step

Here’s how you typically interact with Ghost for interview preparation:

1. Launch Ghost: Start the application. A discreet, minimal window will appear on your screen.

2. Configure Settings: The first time you run Ghost (or if you need to change settings), use Ctrl + Shift + I (Windows/Linux) or Cmd + Shift + I (macOS) to open the settings menu. Here you will:

   • Enter your personal API key (either OpenAI or Google Gemini).

   • Select your preferred AI models for different tasks (e.g., problem extraction, solution generation).

3. Capture the Problem: When you encounter a coding problem you want to analyze (e.g., on a web page or in an editor), position the Ghost window if needed and use Ctrl + H / Cmd + H. Ghost takes screenshots, designed to capture both problem text and code snippets effectively.

4. Process with AI: Once screenshots are captured, press Ctrl + Enter / Cmd + Enter. This action initiates the AI processing sequence using your API key.

5. AI Analysis & Solution:

   • Ghost sends the captured screenshot data directly to the selected AI provider (OpenAI or Google Gemini).

   • The AI first extracts and understands the coding problem from the visual data.

   • It then generates an optimized solution, often including detailed explanations and time/space complexity analysis.

6. Review Solution: The AI-generated solution appears within the Ghost window for you to study and learn from.

7. Manage the Window: While using Ghost, you can:

   • Move the window: Ctrl + Arrow keys / Cmd + Arrow keys

   • Toggle visibility: Ctrl + B / Cmd + B

   • Adjust opacity: Ctrl + [ / Cmd + [ (Decrease) and Ctrl + ] / Cmd + ] (Increase)

   • (Resize and Zoom functionality may also be available)

8. Start New Problem: Press Ctrl + R / Cmd + R to clear the current context and prepare Ghost for a new problem-solving session.

Core Mechanisms: Under the Hood

Understanding how Ghost operates can help you use it effectively:

• Stealth Interface:

  • Ghost is designed with a "Stealth Mode" – an interface intended to be minimally intrusive and bypass many screen capture methods during practice sessions.

  • Invisibility Details (Based on Testing):

    • Hidden From: Zoom (versions older than 6.1.6), browser-based screen recording, Discord (all versions), standard macOS screenshots (Cmd+Shift+3/4).

    • Visible To: Zoom (versions 6.1.6 and newer), native macOS screen recording (Cmd+Shift+5).

  • The exact technical methods aim to prevent the window from being included in the video feed or recording of specific applications, but it's not universally undetectable.

• Smart Capture & AI Processing:

  • The screenshot function captures visual information from your screen.

  • When you trigger processing, Ghost sends this visual data directly to the external AI's API endpoint (like GPT-4o Vision or Gemini).

  • Crucially, this uses your personal API key. The AI interprets the image, extracts the problem, and generates the solution. Ghost acts as the interface and orchestrator for these API calls.

• AI Model Integration:

  • You can choose between supported AI models (currently OpenAI and Google Gemini families) within the settings.

  • This allows flexibility based on cost, performance, or preference for different tasks (extraction vs. solution generation).

• Debugging Assistance:

  • Ghost leverages the AI to provide structured feedback and assistance for debugging your own code snippets, helping you identify and fix errors during practice.

• Privacy:

  • Your API key is stored locally on your machine and is only used to make direct calls to the AI provider (OpenAI/Gemini) when you initiate processing.

  • Ghost does not send your key or problem data to any intermediary servers.

• Extensibility:

  • The application is designed to be extensible. Developers can integrate support for other Large Language Models (LLMs) like Claude, Deepseek, etc., by modifying the API handling code found in /src/main/processor.

Tech Stack

Ghost is built using modern web technologies packaged as a desktop application:

• Framework: Electron

• UI: React, TypeScript, Tailwind CSS, Radix UI Components

• AI Integration: Direct API calls to OpenAI and Google Gemini.