Preface

DeepSeek Harness (DSH) adopts a plugin architecture, aiming to encapsulate specific capabilities into workflows. When performing local document recognition (OCR), directly calling the official cloud model or setting up a standalone environment is often not efficient enough. This plugin provides local PaddleOCR 3.x capabilities for the DSH Web profile, invokable through the official DSH interface without extra configuration.

Plugin Introduction

Name: qiuchunhuimax/dsh-plugin (project identifier is @dsh-external/dsh-paddle-ocr)
Maintainer: qiuchunhuimax
Category: Model inference
Purpose: Workspace-confined PaddleOCR tool and /ocr command for DeepSeek Harness.

Core Features

The plugin provides two invocation methods:
1. Model tool: paddle_ocr. The DSH Agent can directly call this tool within the current workspace.
2. Human command: /ocr <file path> [ch|en]. You can enter the command directly in the conversation.

Supported input file formats include: PDF, PNG, JPG, WEBP, BMP, TIF/TIFF.
For output, the plugin generates a unique Markdown file in the artifacts/paddle-ocr/ directory of the current workspace and does not overwrite the original files.

Runtime Requirements

  1. Python environment: The system must have Python 3 installed.
  2. PaddleOCR library: The paddleocr 3.x version must be installed.
  3. Environment variables: By default, it runs python3. If PaddleOCR is installed in another Python environment, set the PADDLEOCR_PYTHON environment variable before starting DSH.

Installation and Enablement

Use the official plugin installation command to add the plugin:

dsh plugin --profile web add github:qiuchunhuimax/dsh-plugin

After installation, restart or refresh the dsh web service to use it in conversations.

Typical Usage

Enter the following command in the conversation to perform recognition:

/ocr scans/invoice.pdf ch

Alternatively, issue a direct instruction to the Agent, for example: “Recognize scans/invoice.pdf and organize the results”.

Use Cases and Notes

This plugin is suitable for developers who need local document recognition in DSH workflows.

Notes:
* The plugin uses the official DSH bundle/Tool Registry interface, calls the official PaddleOCR Python API, and does not use the DeepSeek-OCR model.
* The plugin resolves symbolic links, but the resolved path does not leave the current workspace.
* It is recommended to check the source code and license (MIT) before installation.

Ecosystem context: The DSH philosophy is “everything is a plugin”; the community directory is an independent site and has no official affiliation with DeepSeek / High-Flyer.

View Plugin Directory | Access GitHub Repository