AI Agent Hub
Back to plugins
🧰

dsh-baidu-ocr

Web Tools Updated 2026.08.26

Run the following command in DeepSeek Harness:

dsh plugin install pipiwolve/dsh-baidu-ocr

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install pipiwolve/dsh-baidu-ocr in the DeepSeek Harness CLI to install the plugin; the full source is hosted at https://github.com/pipiwolve/dsh-baidu-ocr . After installation, restart the Web UI and refresh your browser.

About this plugin

Turning a screenshot or a PDF into Markdown inside the DeepSeek Harness web UI is not as seamless as it should be. The browser only hands you a File object on drop—no real local path—while the OCR engine absolutely needs that path to read the file. dsh-baidu-ocr closes that gap: drag a file onto the page, the host process resolves the real path, calls the Baidu Cloud API, and writes the Markdown result right next to the source, without opening a terminal or stitching paths by hand.

One BCE IAM key powers two engines. PaddleOCR-VL (synchronous, Qianfan platform) handles images and generic PDFs, returning Markdown with layout analysis, chart recognition, and orientation/direction correction. Unlimited-OCR (asynchronous, aip platform) goes deeper into document parsing: formulas become LaTeX, tables become HTML, and multi-column layouts are intelligently merged. File extensions auto-route to the correct engine, or you can override the choice in the panel. Results land as .md and .json files in an ocr_output/ folder beside the source, easy to audit and re-ingest.

If you regularly convert scans, contracts, or slide decks into editable Markdown, or you want your Agent to call OCR directly in a conversation instead of you copy-pasting text back and forth, this plugin offers a lightweight dual-channel workflow: a drag-and-drop GUI panel plus a model tool. One key, two engines, instant recognition on drop, and results written to disk—no heavy local PaddleOCR deployment and no juggling multiple API credentials.

Use Cases

  • Drag a scanned contract or slide deck onto the page and get editable Markdown in one step
  • Call the baidu_ocr tool directly in an Agent conversation to recognize a local image path
  • Batch-parse PDFs and office documents into structured text with LaTeX formulas and HTML tables

Best For

  • Document handlers who frequently convert images or PDFs to Markdown
  • Developers who want their Agent to invoke OCR automatically without manual copy-paste
  • Lightweight-focused users who prefer cloud APIs over local PaddleOCR deployment