Neolithic OCR

A Java desktop tool that captures a region of the screen, extracts the text with Tesseract and sends it to an OpenAI model with your instructions.

Neolithic runs in the background and listens for a global hotkey. When triggered, you drag a box around any text on screen (a worksheet, a PDF, a web page), proofread the OCR result, type an instruction such as "explain this" or "solve this math problem", and get the answer in a chat window. It was written to help students with learning disabilities work with text they find hard to read.

Features

• Global hotkey (Ctrl + Shift) that works from any application
• Click-and-drag area selection with a live preview
• OCR with Tess4J, using the bundled English model in tessdata/
• Text cleanup with LanguageTool spelling and grammar correction
• Editable OCR result and instruction box before anything is sent
• Chat window for follow-up questions, using gpt-4-turbo-preview

Requirements

• JDK 21
• Maven
• An OpenAI API key
• A desktop session (the app uses Swing and a native keyboard hook)

Building

mvn clean package -DskipTests

This is the same command CI runs. mvn test runs the unit tests in src/test.

Configuration

The API key is read from the OPENAI_API_KEY environment variable:

export OPENAI_API_KEY=your-key        # Linux/macOS
set OPENAI_API_KEY=your-key           # Windows cmd

If the variable is not set, the app falls back to config.json on the classpath. Copy the example and add your key:

cp src/main/resources/config.example.json src/main/resources/config.json

{
  "api": {
    "openai": {
      "token": "your-key"
    }
  }
}

config.json is listed in .gitignore.

Usage

Run from the repository root, since the OCR model is loaded from ./tessdata:

mvn compile exec:java -Dexec.mainClass="org.codered.neolithic.Neolithic"

1. Press Ctrl + Shift to start a capture. A preview window opens.
2. Click and drag over the area you want. The preview updates when you release the mouse.
3. Click Accept (or press the hotkey again). Press Esc or close the preview to cancel.
4. In the "Converted Text" dialog, correct the OCR text if needed and enter an instruction. Refresh Text runs OCR on the image again.
5. Click Send to AI. The response appears in a chat window, where you can ask follow-up questions.

Closing the chat window exits the application.

Project structure

src/main/java/org/codered/neolithic/
  Neolithic.java          entry point
  screenshot/             global hotkey and screen area selection
  images/conversion/      Tesseract OCR
  images/processing/      text cleanup, spelling and grammar correction
  images/ui/              converted text dialog
  openai/                 OpenAI request and chat window
  utils/                  API key lookup (environment variable or config.json)
tessdata/                 English Tesseract model