dsh-vision-adapter
Run the following command in DeepSeek Harness:
dsh plugin install motongv/dsh-vision-adapter
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install motongv/dsh-vision-adapter from the DSH Web terminal; the source lives at https://github.com/motongv/dsh-vision-adapter , then restart DSH to see the Vision settings page.
About this plugin
DeepSeek Harness is a powerful reasoning workbench, but it is text-only. The moment you need to understand a screenshot, a whiteboard photo, or a data chart, the workflow breaks—you have to transcribe the image by hand and feed the text back in. dsh-vision-adapter closes that gap: after a one-line install and a restart, two vision models appear in the chat model picker. Drag an image into the conversation, a vision model describes it, and DeepSeek reasons over the combined output, all within the same interface.
Setup is nearly frictionless. Eight vision providers ship as complete presets—Kimi, Zhipu GLM, Tongyi Qianwen, OpenAI, Google Gemini, Doubao, MiniMax, and StepFun. Open Settings to the Vision page, pick a provider, paste your API key, and save. Endpoints, model names, and credential references are all pre-filled, and the key is written securely into the DSH credentials file rather than left in plain settings. Each provider pairs with two DeepSeek backends (Pro for depth, Flash for speed), and the model labels update automatically when you switch.
Under the hood the adapter runs an in-process HTTP bridge and registers the chosen vision model as an llm-pi-ai provider via incremental merge, so it never clobbers other plugins. Practical touches include per-image description caching, automatic 429 back-off and retry, and role/token normalisation fixes. If you live in DSH Web for coding, paper reading, or analysis and do not want to jump to a separate frontend just to look at an image, this plugin makes vision a first-class citizen in your daily workflow.
Use Cases
- Drag a screenshot or whiteboard photo into DSH chat for vision description followed by DeepSeek reasoning
- Understand figures and charts in papers without manually transcribing them into text
- Analyze data screenshots or UI mockups in a single vision-plus-reasoning step
Best For
- Developers who code and analyze documents daily in DSH Web
- Users with multimodal workflows that frequently involve screenshots, charts, or design mockups
- Harness users who want to add vision to DeepSeek without switching frontends
Related Plugins
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.
An armor-breaking plugin for DeepSeek: rewrites instruction priority, expands refusal-blocked phrases, persists across sessions, and shows a green active indicator.