Introduction¶

In the DSH workflow, if you need to process image descriptions, text extraction, or table recognition without letting images leave your local machine, niyongsheng/free-vision-skill is a DSH plugin for macOS. It utilizes the macOS Vision Framework to complete image understanding and OCR locally on your Mac.

What is this¶

niyongsheng/free-vision-skill is maintained by niyongsheng and is licensed under MIT. It is positioned as a DSH-Plugin: fully local image understanding and OCR for DeepSeek Harness, with capabilities derived from macOS Vision.

Core Features¶

The verified plugin capabilities are as follows:

Fully local image understanding: Provides OCR, table extraction, and image description via macOS Vision.
view_image: Describes image content, including scene, people, QR codes, and composition.
ocr_image: Extracts text; when layout=true, you obtain table structures and coordinates.
Input support: Supports http(s) URLs, base64, and local paths.
Web UI paste to path: After pasting an image into the DSH web input box (Cmd+V), the local absolute path of that image is inserted.

Installation and Activation¶

Install the plugin:

dsh plugin add @niyongsheng/free-vision-skill

Add the plugin configuration in cordis.patch.yml. The following example inserts the plugin into cordis and configures timeout to 120000:

- insert:
    - id: free-vision-skill
      name: '@niyongsheng/free-vision-skill'
      config:
        timeout: 120000

Typical Usage¶

If you want to verify the local image recognition flow first, you can run the following Swift script example:

swift scripts/ocr.swift image.png # OCR
swift scripts/ocr.swift --layout image.png # table + coordinates
swift scripts/ocr.swift --describe image.png # describe image

image.png: Execute OCR.
--layout: Used for table structures and coordinates.
--describe: Execute image description.

In DSH, the plugin provides two tools: view_image and ocr_image. Inputs can be http(s) URLs, base64, or local paths; ocr_image supports layout=true.

Applicable Scenarios and Notes¶

System requirements: macOS 11+ and Xcode Command Line Tools installed.
Runtime requirements: node >=18.
First run: Compiles for approximately 5–10s, after which it is cached.
License: MIT.
Security: Images do not leave the Mac.
Web UI upload: Restricted to loopback and undergoes magic-byte checks. Supported formats include PNG / JPEG / GIF / WebP / HEIC / HEIF.
Permissions: The plugin runs with the permissions of the current dsh process. It is recommended to check the source code and license before installation to confirm that the call method and input processing align with your environment requirements.

Summary¶

The value of niyongsheng/free-vision-skill lies in performing image description, OCR, and table recognition locally on macOS, and integrating into DSH via view_image and ocr_image. Before using, confirm the macOS version, Xcode Command Line Tools, and the license.

Community directory page: https://www.skillhub.cn/plugins/niyongsheng/free-vision-skill
GitHub: https://github.com/niyongsheng/free-vision-skill

niyongsheng/free-vision-skill: DSH Local Image Understanding and OCR Plugin