Introduction

When developing a DeepSeek Harness (DSH) plugin or using a DSH agent, handling unstructured documents (such as PDFs and scanned documents) is a common need. Traditional approaches often rely on external APIs, which may introduce privacy risks or network latency. dsh-docs provides a localized solution that can directly complete document reading, OCR recognition, and text extraction on a local computer.

What Is This

protoctistmoses143/dsh-docs is a model inference plugin focused on local document processing. It supports converting PDFs, Office documents, images, and scanned documents into Markdown, JSON, or plain text. It includes built-in offline OCR capabilities and requires no server support, ensuring data privacy.

Core Features

This plugin has the following verified capabilities:

  • Multi-format reading: Supports PDFs, Office documents, and image files.
  • Offline OCR: Recognizes text in scanned documents without network access and extracts text from images into editable text.
  • Data extraction: Supports full-text extraction, converting document content into Markdown, JSON, or plain text formats.
  • In-document search: Supports searching within extracted text.
  • Batch processing: Supports processing multiple files in batches.
  • Privacy-first: All processing is completed locally, and data is not uploaded to servers.

Installation and Dependencies

This plugin is primarily designed for Windows computers and requires local environment support.

  • Operating system: Windows (mainly supports Windows 10/11).
  • Runtime dependencies:
    • Node.js version: ^22.19.0 or >=24.0.0
    • Package manager: pnpm version >=10.0.0

Typical Usage

This plugin is primarily operated through a graphical interface, without requiring additional code.

  1. Start the application: Run the installed dsh-docs application.
  2. Drag and drop files: Directly drag PDFs, Office documents, or image files into the main window.
  3. Export results: The application automatically parses the document and displays the extracted text content in the interface. Users can copy the extracted text to other programs or use it as a data source.

Example scenario:
Users can take a photo of a restaurant menu with their mobile phone, transfer the image to a PC, and then drag it directly into the dsh-docs window. The system will use the local OCR engine to recognize text in the image and generate editable menu text.

Notes

  • Platform limitations: The plugin is designed mainly for Windows computers; other systems may not be able to run it directly.
  • Fully offline: OCR features and all document processing do not rely on a network connection.
  • Privacy and security: Because it uses a fully local processing mode, it is suitable for handling sensitive documents. However, before using it, we recommend reviewing the GitHub source code and license (MIT).

Summary

dsh-docs provides DSH users with a reliable tool for local document processing, especially in scenarios where data privacy protection or offline environments are required, effectively solving document recognition and extraction issues.