Argonaut790/dsh-deepseek-vision

This is a native open-source vision plugin built exclusively for DeepSeek Harness (DSH), designed to add image understanding, full-screen OCR, and persistent visual evidence capabilities to text-only DeepSeek models without replacing the parent model. To install the plugin, you first clone the GitHub repository to your local machine, run corepack yarn install and corepack yarn build to compile it, then add the local package to your DSH web profile using the dsh plugin command. It adds a global vision model selector next to the main model picker in DSH’s UI, so you can change the vision analysis route without altering your main conversation model.

When a user pastes an image into a DSH conversation, the text-only parent model can call the see_image tool to request visual analysis from the plugin, which supports selecting the latest image, all images, or specific image IDs by attachment ID. The plugin maintains a conversation-scoped vision analyst with follow-up memory, so it reuses existing context for new questions instead of re-analyzing the same images multiple times. Each analysis result is stored as a readable evidence card in the conversation history, making all visual work reviewable later, and it fits users who need to add reliable vision capabilities to a text-only DSH setup.

The plugin requires Node.js version 22.19.0 or higher, DSH version 0.1.0-rc.6 or newer, and an image-capable model already registered in your DSH catalog, along with the DSH spawn subagent provider. It is released under the permissive MIT open-source license, but you should note that selected images are sent to your configured vision provider, so you need to review that provider’s privacy and pricing terms before use. You also must disable any existing built-in DSH vision tools before installing this plugin, to avoid conflicts from duplicate services and tools that break DSH functionality.