dsh-pdf

Most DSH agent plugins hit a hard wall when reading PDFs: a 64 KB file cap that rejects real-world documents, and no way to extract text from scanned or image-heavy pages. dsh-pdf removes both constraints in one pass. It lifts the single-file limit to 200 MB by default and pulls the full Unicode text layer through pdfjs-dist, so Chinese, English, and other scripts work out of the box.

The standout feature is its automatic OCR fallback. When a page's text layer falls below a configurable character threshold, the plugin renders that page to an image and runs OCR. On Windows it prefers the built-in WinRT recognizers (Simplified Chinese and English, zero-install, zero-download); in headless or non-Windows environments you can drop tesseract.js trained-data into the expected folder as an alternative. Every page in the result is tagged as text, ocr, or mixed so the agent can gauge confidence. All knobs—OCR engine, render scale, per-page timeout, cache size—are adjustable live from the DSH settings UI, with changes taking effect immediately.

If your workflow involves feeding multi-hundred-page contracts, tens-of-megabytes whitepapers, or handwritten lecture notes to the agent, dsh-pdf is a lightweight, service-free plugin that just works.