User flow
- Copy an image or screenshot in any macOS app.
- Hold cmd and press c twice quickly.
- Quick reads image data from the system pasteboard.
- Quick runs OCR locally.
- Quick shows recognized text in the floating popup.
Routing
| Clipboard content | Behavior |
|---|---|
| Text | Sent to the configured OpenAI-compatible provider for translation. |
| Image data | Processed locally with OCR. |
| Image data plus text or file path | Image data takes priority. |
| Empty or unsupported content | Quick shows a lightweight error message. |
Privacy
Image OCR runs on device. Copied images are not sent to OpenAI or third-party API providers. Only text translation uses the configured API key, Base URL, model, and System Prompt.
Implementation
OCR is implemented in Swift with ONNX Runtime and bundled PP-OCRv6 tiny models.
| Asset | Purpose |
|---|---|
ppocrv6_tiny_det.onnx |
Text detection model. |
ppocrv6_tiny_rec.onnx |
Text recognition model. |
ppocrv6_tiny_det.yml |
Detection metadata. |
ppocrv6_tiny_rec.yml |
Recognition metadata and character dictionary. |
Verification
swift test swift run QuickOCRInspect inspect AppBundle/Resources/OCR swift run QuickOCRInspect recognize AppBundle/Resources/OCR /path/to/image.png
Current limits
- OCR quality is aimed at clear screenshots, UI text, and document text.
- Image OCR does not automatically translate recognized text in v0.2.0.
- The first OCR release uses a lightweight Swift DB postprocessor with axis-aligned crops, not full PaddleOCR rotated box postprocessing.