| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| CaptureOCR source code.tar.gz | 2026-10-21 | 583.2 kB | |
| CaptureOCR source code.zip | 2026-10-21 | 588.1 kB | |
| README.md | 2026-10-21 | 1.6 kB | |
| Totals: 3 Items | 1.2 MB | 3 | |
CaptureOCR is a screenshot, annotation and OCR productivity tool distributed as an MSIX-packaged Win32 desktop application. The executable is built with Python/PySide6 and ships with a bundled Python runtime; it is a full-trust Win32 process by architecture and cannot operate inside the UWP AppContainer.
runFullTrust is declared because the app performs the following operations that are only available to a full-trust desktop process:
Screen capture and DPI-correct cropping. The app captures monitors using Win32/mss APIs and crops selections by physical pixels. This is necessary to keep selection coordinates accurate across monitors with mixed DPI scaling (100%/150%/200%), which a sandboxed process cannot reliably achieve.
Local OCR inference. PP-OCRv6 ONNX models are executed in-process via ONNX Runtime. This is a local, offline-capable workload; no screenshot pixels or OCR text leave the device.
Global hotkey registration. The app uses the standard Windows RegisterHotKey API so the user can trigger capture from any foreground application. A tray menu is always available as a fallback if the hotkey is unavailable.
User-initiated file and model I/O. The app reads and writes files the user explicitly selects or configures: saving screenshots, exporting Markdown/CSV/TXT, and caching downloaded model files under %LOCALAPPDATA%.
No broad filesystem or registry access is used. The app does not collect or transmit screenshot content or OCR results; network requests are limited to model download, Store license/update checks and optional anonymous crash reports, as described in the privacy policy.
