| name | ocr |
| description | Read text out of images, screenshots, and scanned PDFs with local Apple Vision OCR. Use when a task references an image or scanned document whose text needs to be read or quoted exactly. |
OCR
Extract text from images and scanned PDFs, fully local (Apple's Vision
framework — no API, nothing leaves this machine). The ocrit binary is
already installed and on PATH.
Usage
Any image (png, jpg, heic, tiff, …) or PDF — text prints to stdout:
ocrit photo.png
ocrit scanned-document.pdf
Multiple files in one call:
ocrit page-1.png page-2.png
Non-English documents — pass language codes (repeatable):
ocrit facture.pdf -l fr-FR
When to use which lane
- Scanned/image-only PDFs:
ocrit reads PDFs directly — you usually do
NOT need to render pages first. bakin_exec_pdf_read tells you when pages
look scanned; run ocrit <path> on the PDF it flagged.
- Exact text vs. understanding: OCR gives you machine-exact text
(serial numbers, amounts, codes) for free. If you instead need layout,
handwriting judgment, or "what IS this?", render with
bakin_exec_pdf_render and view the PNGs with your own eyes — vision
handles what OCR can't.
- Screenshots and photos: point
ocrit at the file directly.
Honest failure
If output is empty or garbled, the page may be handwritten, rotated, or
very low quality — fall back to rendering + viewing the image yourself,
and say which pages OCR could not read. Never present partial OCR as the
full document.
