mochgolf/dsh-deepseek-vision-router

Name: mochgolf/dsh-deepseek-vision-router
Author: mochgolf

插件Plugin 原生Native ⭐ 2 MIT cat_label.ai-agentscat_label.ai-agents cat_label.cordiscat_label.cordis cat_label.deepseekcat_label.deepseek cat_label.deepseek-harnesscat_label.deepseek-harness cat_label.dshcat_label.dsh cat_label.multimodalcat_label.multimodal

DeepSeek Harness 的透明图像预处理路线

项目介绍Project Overview

由知微Agent(zhiwei)编辑Edited by Zhiwei Agent (zhiwei)

这是实验性 DSH 插件，为纯文本 DeepSeek 主代理增加图片接收能力：聊天附件先经 Models 中配置的视觉模型生成描述，再交给官方 DeepSeek 适配器继续对话、推理、流式输出、工具与重试。需要在 DSH 中发送图片、又保留 DeepSeek 主会话时使用。注意它只是预处理桥接，并非 DeepSeek 原生多模态；图片字节会发往所选视觉提供商，且仅兼容指定版本。

An experimental DeepSeek Harness plugin that lets a text-only DeepSeek main agent accept images from the DSH composer. Image attachments are routed to a configured vision provider for description, then the text result is passed to the official DeepSeek adapter, which keeps handling conversation, reasoning, streaming, tools, and retries. Use it when image input is needed while retaining DeepSeek as the main agent. It is a preprocessing bridge, not native DeepSeek multimodality; image bytes go to the selected vision provider.

在 GitHub 查看View on GitHub

发给你的 DSH 安装Send to your DSH

请帮我了解并安装插件：【dsh-deepseek-vision-router】【https://github.com/mochgolf/dsh-deepseek-vision-router】

把上面这条消息直接发给当前会话里的 DSH，让它帮你了解并安装。安装命令不一定准确，发给 DSH 更稳。Send this message to DSH in your current session. CLI install commands may not be accurate across systems — DSH will figure it out for you.

或使用命令行安装（适合开发者）Or use CLI install (for developers)

命令行安装CLI Install

dsh plugin --profile web add github:mochgolf/dsh-deepseek-vision-router

把 mochgolf/dsh-deepseek-vision-router 加入你的 DSH 配置（web profile）即可启用。

READMEREADME

2.3 KB · 缓存于 2026-08-27 05:42:15 · 2.3 KB

dsh-deepseek-vision-router

English | 简体中文

An experimental DeepSeek Harness plugin that lets a text-only DeepSeek main agent accept images directly from the DSH chat composer.

DSH image attachment -> vision model description -> deepseek-official

The plugin adds a separate deepseek-vision route. Image analysis runs through an existing provider from DSH's Models settings; the official DeepSeek adapter still owns the main conversation, reasoning, streaming, tools, and retries.

Compatibility

DSH 0.1.0-rc.6
Node.js 24+
mcp-vision-bridge 0.2.7

Other versions may work but have not been tested.

Install

dsh plugin --profile web add github:mochgolf/dsh-deepseek-vision-router

Open Settings → Models and add or reuse an image-capable provider. Its endpoint and credential stay owned by the Models page.
Open Settings → Plugins → DeepSeek Vision, select that provider, and save the vision model. The model field defaults to mimo-v2.5 but accepts any model ID exposed by the selected provider.

Provider additions, edits, and removals remain in Settings → Models. The plugin stores only visionProvider and visionModel in DSH's native settings namespace and applies changes to the next image request without a restart.

Select DeepSeek + Vision in DSH's model menu. New sessions can use it as their default provider; existing sessions retain their recorded provider until switched.

Security and privacy

Provider endpoints and credentials remain managed by DSH's Models settings; this plugin never copies or stores them.
Image bytes are sent to the configured vision provider; DeepSeek receives the resulting text description, not the pixels.
Descriptions are marked as untrusted image content before reaching DeepSeek, so text visible inside an image is not promoted to system instructions.
Successful descriptions use a bounded, process-local cache.

Limitations

The cache is not persisted across DSH restarts.
The analysis prompt comes from an internal mcp-vision-bridge module; an upstream directory change may require a compatibility update.
This is a preprocessing bridge, not native DeepSeek multimodality.

Development

npm ci
npm test

渲染基于仓库 mochgolf/dsh-deepseek-vision-router 的 README.md。在 GitHub 查看原文 →

命令行安装CLI Install 复制安装命令Copy Install