Sorwcyra/ds-vision-plugin 预览 preview

Sorwcyra/ds-vision-plugin

将图片粘贴到DeepSeek Harness中,支持四模型视觉竞赛、OCR及自动文本桥接功能。

Project Overview项目介绍

This is a native plugin built exclusively for DeepSeek Harness that adds vision support to text-only DeepSeek models. It requires Node.js 22.19 or 24 to build and run, and works without any forks or modifications to the core DeepSeek Harness source code. After installation, users can paste or drop images directly into the Harness web composer in PNG, JPEG, WebP, or GIF format, and the plugin will automatically convert images to grounded text using a multi-model race approach. It also exposes two tool endpoints, vision_analyze for workspace image processing and vision_status for diagnostics without exposing private API keys.

The default configuration runs a four-model race where Agnes 2.5, Agnes 2.0, GLM-4V-Flash, and GLM-4.1V-Thinking-Flash all start processing the same image at the same time. The first valid result returned is used, which maximizes availability and minimizes latency for end users. Setup is made easy by a guided command-line interface that handles configuration, API key storage, status checks, and adding custom models, so users never have to manually edit YAML files. This plugin is ideal for any DeepSeek Harness user that regularly works with image input like screenshots, diagrams, or photos.

This plugin is released under the open-source MIT license, so it is free to use, modify, and distribute. It supports adding custom OpenAI-compatible vision models, either as part of the concurrent race or as an ordered fallback, and users can choose between two failure modes: annotate, which marks failed conversions visibly, or error, which stops the request on failure. For privacy-sensitive images, users can disable automatic conversion or use local VLM or OCR runtimes to avoid sending image data to third-party cloud providers. To build and verify the plugin locally, you can run pnpm install followed by pnpm run build, pnpm run check, and pnpm pack to generate the installable bundle.

这是一款专为 DeepSeek Harness 开发的原生视觉插件,能够让纯文本版的 DeepSeek 模型原生支持图片输入与 OCR 识别。用户只需在 Harness 的网页编辑器中粘贴或拖入图片,插件就会自动调用配置好的多模型并发竞态识别,将图片转换为带上下文的文本输出给 DeepSeek,无需修改 Harness 源码即可使用,同时还暴露了接口支持分析工作区图片和诊断状态,也支持用户自定义添加私有模型或本地运行时,满足不同隐私需求。

对于普通用户来说,使用流程非常简单:启动网页配置文件,选择 DeepSeek 提供商,粘贴或拖入一张或多张 PNG/JPEG/WebP/GIF 格式图片,按需添加问题后发送即可。用户还可以通过引导式 CLI 完成配置、密钥设置、状态检查和模型自定义,全程无需手动编辑 YAML 文件。插件默认采用四模型竞态模式,能最大程度保证低延迟和高可用性,适合关注识别速度和稳定性的 DeepSeek Harness 用户。

本插件基于 TypeScript 开发,要求 Node.js 版本为 22.19 或 24,采用 MIT 许可证开源,完全免费使用。隐私方面,网页上传的图片仅通过 Harness 验证的私有附件服务读取,配置的云渠道只会获得图片字节,用户如果处理敏感图片,可以使用本地 VLM 或关闭自动转换。失败处理默认采用标注模式,不会静默丢弃图片,保障问题可追溯。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 1 warning1 项注意
  • Only 4 stars - very few users, little community feedback星标只有 4,几乎没人在用,遇到问题缺少社区反馈
DSH walks through these 9 checksDSH 会逐条核对这 9 项

Compatibility兼容性

  • DSH, Node, OS and profile requirementsDSH 版本 / Node 版本 / 操作系统 / profile 是否满足要求
  • External dependencies and runtimes (Electron / Python / Docker, ...)外部依赖与运行时(Electron / Python / Docker 等)是否齐备
  • Conflicts with installed plugins: command names, skill / tool names, ports, duplicate MCP registration与已装插件是否冲突:命令名、skill / tool 重名、端口占用、重复 MCP 注册

Security安全性

  • Repo matches the facts registered here; archived or abandoned?仓库是否与页面登记一致,是否归档或长期停更
  • Safety of preinstall / install / postinstall and install.sh / setup.ps1preinstall / install / postinstall 与 install.sh、setup.ps1 是否安全
  • curl|bash, download-then-execute, obfuscation, unrelated domains → stop immediatelycurl|bash、下载即执行、混淆代码、无关域名 → 立刻停止
  • Typosquatting or unmaintained packages among the new dependencies新增依赖里有没有 typosquatting 或无人维护的包
  • Requested permissions vs. what the feature actually needs申请了哪些权限、是否超出功能所需(filesystem / network / shell / clipboard)
  • Any sudo / admin requirement, plus uninstall and rollback是否要求 sudo / 管理员权限,以及卸载与回滚方式

Anything uncertain must be marked unknown with a note on how to confirm it. This site's signal screen is a static snapshot, not a security audit.拿不准的必须标「未知」并说明要我怎么确认。本站的信号筛查是静态快照,不能替代安全审计。

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

CLI Install命令行安装

npx -y @deepseek-ai/dsh plugin --profile web add "github:Sorwcyra/ds-vision-plugin"

把 Sorwcyra/ds-vision-plugin 加入你的 DSH 配置(web profile)即可启用。

READMEREADME

A black whale carrying image data through a four-model vision race into text

ds-vision-plugin

Paste an image. Let four vision models race. Keep DeepSeek text-only.

简体中文 · Quick start · Routing · Verification · License

CI Stars Forks Last commit Issues License

Version 0.4.0 DeepSeek Harness plugin Four-model race Node.js 22.19 or 24

paste → attachment → race ×4 → grounded text → DeepSeek

An installable DeepSeek Harness bundle that gives a text-only DeepSeek model a natural image-input experience. Paste or drop an image into the Web composer; the plugin reads Harness's verified attachment, races configured vision models or OCR, replaces the image with grounded text at agent/pre-step, and lets DeepSeek continue normally.

[!NOTE] No Harness source fork is required. The plugin also exposes vision_analyze and vision_status for workspace files and diagnostics.

Why it exists

Showing the opening section of the README — the full document lives in the repository以上为 README 开头摘要,完整文档在仓库内 · View the full README on GitHub →在 GitHub 查看完整 README →

← 上一个 Prev dsh-thincoder-suite 下一个 Next dsh-reasoning-effort-hdbzq →