linenxi-ctrl/dsh-vision
111Last commit Aug 15, 2026
dsh-vision DSH plugin
dsh-vision is a plugin for DeepSeek Harness that integrates external vision models (e.g., GPT-4o, Claude, Gemini) via a configurable API. It provides a floating button and panel for users to send images for recognition, and injects screenshot and recognize_image tools for agent-driven vision tasks. The plugin automatically adapts to multiple API protocols and supports custom endpoints.
How to install the dsh-vision DSH plugin
dsh plugin --profile web add @linenxi-ctrl/dsh-visionCopying does not run this command. Review the repository and version before installing the dsh-vision DSH plugin.
dsh-vision DSH plugin data source
dsh-vision DSH plugin snapshot date: Aug 16, 2026
discovered
What the dsh-vision DSH plugin can do
- Provides a draggable whale button and configuration panel for setting API address, key, model, prompt, proxy, and timeout.
- Allows users to send images for recognition via the panel; recognition text is automatically posted back to the current chat session.
- Injects screenshot and recognize_image tools into the agent, enabling the model to capture and analyze screenshots autonomously.
- Automatically detects and adapts to OpenAI Chat/Responses, Anthropic, Gemini protocols, with a custom template for arbitrary APIs.
Where the dsh-vision DSH plugin fits
- Send an image to a chat and have the model describe its contents based on external vision model analysis.
- Ask the model to read an error message on your screen; the model will take a screenshot and recognize the text.
- Use the configuration panel to switch between different vision models (e.g., GPT-4o, Claude, Gemini) without modifying code.
Who the dsh-vision DSH plugin is for
- Users of DeepSeek Harness who need vision capabilities but prefer to keep their primary model text-only.
- Developers and power users who want to extend DSH with custom external vision models.
dsh-vision DSH plugin limitations
- Requires an external vision model API key, base URL, and model name; no built-in vision model is included.
- Recognition quality depends entirely on the external model; the plugin only passes images and returns text.
- The plugin is in early development (v0.2.6) and may have edge cases; the custom protocol feature requires manual JSON template configuration.
- Only supports image input; video or real-time streaming is not supported.
dsh-vision DSH plugin: from the repository README
Quoted from the linenxi-ctrl/dsh-vision README, the upstream source of the dsh-vision DSH plugin. Copyright remains with the original authors.
为 [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) 增加「外挂识图模型」能力：让本来不具备视觉能力的模型，通过一个可自定义地址/密钥/提示词的外部视觉模型来「看懂」图片与屏幕。 ## 功能 1. **网页配置按钮与面板**：页面右下角出现一个DeepSeek 鲸鱼圆形按钮（可拖动），点击即可配置外挂识图模型的 API 地址、密钥、模型名、识图提示词（skill）、代理与超时。 2. **发送图片识图并自动回传**：点鲸鱼按钮打开面板，点「📤 发送图片」选图，插件会先把它发给外挂识图模型，等识别完成后把识别文本**自动作为消息发回当前会话**（无需手动复制粘贴），DeepSeek 基于识别文本作答。 3. **模型自己截图 + 识图**：插件为 agent 注入 `screenshot`（截屏）与 `recognize_image`（识图）两个工具，并注入提示词，模型可自行「截图 → 识图 → 等待结果」。 4. **自动适配识图 API 协议**：内置 OpenAI Chat Completions、OpenAI Responses、Anthropic Messages、Google Gemini 四种协议，并按 `apiBase` 自动探测；另有 `custom` 模板协议适配任意长尾接口。 ## 文件结构 ``` dsh-vision/ ├── install.bat # Windows 一键安装（双击） ├── install.sh # macOS/Linux 一键安装 ├── install.mjs # 安装脚本本体（npm 场景只做 agent 工具平面；目录场景全自动） ├── bootstrap-node.ps1 # Windows 引导脚本：未装 Node.js 时从国内镜像自动下载免安装版 ├── package.json # 包定义（dsh.bundle + dsh.client 声明；tool 为独立子路径） ├── cordis.patch.yml # 插件挂载声明（dsh.bundle.patch 自动应用到 profile layer） ├── lib/ │ ├── index.js # host 平面插件：识图服务 + 协议适配 + settings 配置 + HTTP 路由 │ ├── tool.js # agent 工具插件：recognize_image / screenshot + 提示词注入 │ └── client.js # 客户端插件：鲸鱼按钮 / 配置面板 / 发送图片识图 / 自动回传 └── README.
Read the full READMERepository license: MIT
dsh-vision DSH plugin questions
How do I install dsh-vision?
Run `dsh plugin --profile web add @linenxi-ctrl/dsh-vision` in your terminal. This requires Node.js 18+ and pnpm. Alternatively, you can download the zip from Releases and run the install script (no Node.js required). After installation, restart DSH with `dsh web`.
Why is my image recognition failing with a 401 error?
A 401 error means your API key is missing or incorrect. Open the configuration panel by clicking the whale button in the bottom-right corner, then re-enter the correct API key. Make sure the key matches the external vision model service you are using.
How do I make the model automatically take a screenshot and analyze it?
First, ensure the agent tools are injected by running `node ~/.dsh/profiles/web/node_modules/@linenxi-ctrl/dsh-vision/install.mjs` after npm installation. Then, in a chat session, simply ask something like "Look at my screen and tell me what error I'm seeing". The model will call the `screenshot` tool, then `recognize_image`, and respond.
Can I use a custom vision model that is not OpenAI, Anthropic, or Gemini?
Yes, set the protocol to `custom` in the configuration panel. You need to provide a `requestTemplate` JSON with placeholders like `{{model}}`, `{{prompt}}`, `{{image}}`, and a `responsePath` to extract the text from the response. The plugin will send the request with `Authorization: Bearer <apiKey>` header. For special authentication, you may need to open an issue.
How do I uninstall dsh-vision?
If you installed via npm, first run `dsh plugin --profile web remove @linenxi-ctrl/dsh-vision`. Then run the uninstall script: on Windows, double-click `uninstall.bat`; on macOS/Linux, run `bash uninstall.sh`. This will remove the plugin files, the profile entry, and the agent preset.