Peekr

Peekr is an offline cross-platform OCR desktop application that can extract text from screen and images. Recognition runs on local PaddleOCR models (PP-OCRv6 and PP-OCRv5) via ONNX Runtime.

Preview

Installation

Download the latest release for your platform from the Releases page.

Windows

• peekr-windows-x86_64-Setup.exe
• peekr-windows-x86_64-Portable.zip

Linux

• peekr-<platform>.AppImage (linux-x86_64 or linux-aarch64, needs libfuse2)

  chmod +x peekr-linux-x86_64.AppImage
  ./peekr-linux-x86_64.AppImage

• peekr-<version>-<platform>.tar.gz (this distribution doesn't support autoupdate)

Usage

You can use it in 2 modes - default tray app that runs in background and called by shortcut or CLI.

Tray app

Note

On Wayland apps cannot register a global hotkey themselves so you should bind peekr --capture to a shortcut in your settings instead.

The default mode. Peekr starts in it whenever it is run without arguments, and stays in the system tray until it is quit.

CLI

peekr                             # Start the tray app (the default)
peekr --capture                   # Select a region once, copy the text and exit
peekr --image image.png           # Print the text of an image file
peekr --list-models               # Show the models and which are installed

# Model selection, for any of the above
peekr --det pp-ocrv5-mobile       # Use this detector instead of the first installed one
peekr --rec pp-ocrv5-eslav        # Use only this recognizer
peekr --rec auto                  # Run every installed recognizer (the default)

With --rec auto each line keeps its most confident reading. pp-ocrv5-eslav reads Cyrillic, pp-ocrv6-small reads Chinese, Japanese, English and Latin-script languages.

License

This project is licensed under the MIT license.

The PaddleOCR models are provided by PaddlePaddle under the Apache License 2.0. Release archives include the licenses of the models, ONNX Runtime and all dependencies.