MIT 许可的 Electron 桌面客户端，在本机托管 DeepSeek Harness（自动启动 `dsh web`），提供零代码会话、插件发现与一键安装、Preset 广场和视觉路由，支持 macOS arm64 与 Windows x64。
DSH 适配
原生运行时
作者声明
安全审计
未审计
最后核验
2026-08-21
许可证
MIT
01它能帮你完成什么？
Run DeepSeek Harness as a zero-code desktop app on macOS and Windows, without touching the command line
A local desktop workspace keeping DeepSeek Harness sessions, workspaces, models, tools, Skills and plugin capabilities, hosted by a `dsh web` service the app starts and manages
Users who want a DeepSeek Harness desktop experience with no Node.js setup or terminal commands
Discover, install and manage DSH plugins and Agent Presets from a visual plugin center
Verified plugin installs from the public npm registry with enable/disable/update/uninstall management, plus one-click Agent Preset installs usable in new sessions
DeepSeek Harness users who want to browse, verify and install ecosystem plugins without memorizing package names or CLI flags
Give DeepSeek image understanding in desktop conversations via vision enhancement
Images routed to the exact model's native image input, or to a verified Bailian/OpenRouter-compatible vision path, covering screenshots, error dialogs, design mocks and charts
Users who paste product screenshots, error screens or design files into DeepSeek Harness chats and want the agent to actually see them
02如何接入 DeepSeek Harness？
前置条件
- macOS arm64 or Windows x64 machine — preview installers are published only for these platforms
- No Node.js or pnpm needed for the packaged desktop app
- For source development only: Node.js `^22.19.0 || >=24.0.0` and pnpm `11.7.0`
安装步骤
- 01
Download the macOS arm64 preview ZIP or the Windows x64 Setup.exe from this repo's GitHub Releases (desktop installers are not distributed via third-party download sites)
- 02
macOS: unzip the preview, drag `DeepSeek Harness.app` into Applications, then run `xattr -dr com.apple.quarantine "/Applications/DeepSeek Harness.app"` and `open "/Applications/DeepSeek Harness.app"` because the preview is not Apple-notarized yet
- 03
Windows: run `DeepSeek-Harness-Desktop-Windows-x64-0.1.0-rc.14-Setup.exe` directly to install
- 04
From source instead: `git clone https://github.com/fufankeji/deepseek-harness-studio.git`, then `pnpm install` and `pnpm run dev:desktop` to build the required modules and launch the desktop dev environment
$ git clone https://github.com/fufankeji/deepseek-harness-studio.git
验证接入成功
- After the first successful open on macOS, the app can be launched like a normal application from Finder or the Dock
03DSH 适配与能力边界
Native desktop host: the Electron main process starts and manages a local `dsh web` service, wrapping the DeepSeek Harness core and Web workspace in a zero-code desktop app
Local Harness Host (`dsh web` lifecycle)
the bundled DeepSeek Harness core and Web workspace→a local `dsh web` service started by the desktop main process, kept ready in the app window and shut down when the app exits
starts and manages a local `dsh web` host process on your machinePlugin discovery & public plugin center
the online public `dsh-plugin` catalog and published npm packages (short name, full npm name, or mapped GitHub repo)→curated/recent/trending plugin feeds, natural-language `/find-plugins` search, verified installs, and enable/disable/update/uninstall management
reads the online plugin catalog and downloads packages from the public npm registryinstalls plugins locally and restarts the Harness Host after installuninstall keeps config and plugin data by default; data cleanup requires separate confirmationVision enhancement with auto routing
images attached in the chat input (screenshots, error screens, design mocks, charts, photos)→the original image sent to the model's native image input, or a verified Bailian/OpenRouter-compatible vision path — one path per image
when enabled, images enter the model-visible context and are sent to the selected model/vision provider; compatible vision requires a verified providerPreset Square & Application Center
Fufan-official and community Agent Presets (7 built-in real-scenario workflows) and built-in AI applications such as FF–LLM Wiki→one-click preset installs usable directly in new sessions, and full AI apps with their own UI, data directory and runtime flow
preset installs are verified (source, size, digest, archive path) and written to the local user Preset directoryFF–LLM Wiki writes documents, Wiki, knowledge graph and SQLite data to an app-specific directory under DSH Home
04适合谁？何时不该用？
适合
- Users who want a DeepSeek Harness desktop experience with no Node.js setup or terminal commands
- DeepSeek Harness users who want to browse, verify and install ecosystem plugins without memorizing package names or CLI flags
- Users who paste product screenshots, error screens or design files into DeepSeek Harness chats and want the agent to actually see them
不适合
- The macOS preview package is not Apple-notarized yet, so first launch requires removing the quarantine attribute via `xattr`; previews are only macOS arm64 and Windows x64, distributed exclusively through this repo's GitHub Releases.
- Current downloads are development preview assets (rc.14); signed official releases (macOS `.dmg`, Windows `.exe`) will only arrive later, so preview behavior may change.
05兼容性、维护与安全提示
- The macOS preview package is not Apple-notarized yet, so first launch requires removing the quarantine attribute via `xattr`; previews are only macOS arm64 and Windows x64, distributed exclusively through this repo's GitHub Releases.
- Current downloads are development preview assets (rc.14); signed official releases (macOS `.dmg`, Windows `.exe`) will only arrive later, so preview behavior may change.
- Source development requires pinned runtimes — Node.js `^22.19.0 || >=24.0.0` and pnpm `11.7.0`; actual model calls additionally need user-configured credentials in settings.
MIT · actively maintained (latest release desktop-preview-v0.1.0-rc.14, 2026-08-20)
06常见问题
Studio 是怎么接入 DeepSeek Harness 的？
它是原生桌面宿主而非 MCP 插件：项目基于 deepseek-ai/deepseek-harness 的 Harness 核心、Cordis 插件体系和 Web 界面开发，Electron 桌面主进程启动本地 `dsh web`，等待服务就绪后加载工作区，并在应用退出时关闭 Host 进程。Studio 0.1.0-rc.14 整合了 DeepSeek Harness 0.1.0-rc.8 的核心与 Web 能力。
支持哪些平台？从哪里下载？
目前提供 macOS arm64 预览 ZIP 与 Windows x64 安装程序，只通过本仓库 GitHub Releases 发布，不使用第三方下载站。macOS 预览包尚未 Apple 公证，首次打开前需在终端执行 `xattr -dr com.apple.quarantine` 去除隔离属性。
需要提前安装 Node.js 或配置 API Key 吗？
不需要。运行打包后的桌面端无需另行安装 Node.js 或 pnpm，下载与启动也无需填写 API 密钥；只有在应用中实际调用模型时，才在设置里配置所选模型服务与凭证（FF–LLM Wiki 的生成式 RAG 另需自己的 `DEEPSEEK_API_KEY`）。
在线安装插件的流程和校验是怎样的？
插件中心支持用短包名、完整 npm 包名或映射的 GitHub 仓库查找发布到 npm 公共 Registry 的插件，下载确定版本并校验包身份、完整性和 Bundle 声明，确认后自动安装并重启 Harness Host 验证运行状态；Studio 不会直接安装仓库源码。卸载默认保留配置与数据，清理数据需另行确认。
开启视觉增强后，图片会发送到哪里？
按当前确切模型的能力自动选路：模型声明支持图片时直接发送原图；否则走已完成配置与验证的百炼或 OpenRouter 兼容视觉。两条路径互斥，同一张图片只处理一次；关闭开关后图片不进入模型上下文。自定义背景图片仅在本机裁切，不会上传原图。
07相关的 DSH 工作流
open-design
作者 nexu-io
DeepSeek Harness 的开源 AI 设计插件，本地优先桌面应用，让编码智能体变身设计引擎，生成原型、落地页、幻灯片、图片与视频，支持多种格式导出。
picgo
作者 molunerfinn
PicGo：面向高效创作者的图片上传神器，支持 Obsidian、Typora、VS Code 等编辑器，以及 S3、GitHub、Cloudflare R2、Imgur、阿里云 OSS 等 60+ 图床，粘贴即上传。
dsh-desktop
作者 anywhere-labs
为 DeepSeek Harness (DSH) 插件生态打造的现代化桌面端解决方案。万物皆「插件」，桌面本身也是「插件」。
openbiliclaw
作者 whiteguo233
本地私有、开源的自进化跨平台 AI 内容发现 Agent：先理解你，再主动从 B站、小红书、抖音、YouTube、X、知乎、Reddit、微博等平台与开放 Web 寻找内容。（支持 deepseek harness 插件） | Local-first open-source cross-platform AI content discovery agent: understands you, then proactively finds content across Bilibili, Xiaohongshu, Douyin, YouTube, X, Zhihu, Reddit, Weibo and the open web.（support deepseek harness plugin）
08数据与来源
DeepSeek Harness Studio 使用 Electron 承载 DeepSeek Harness 的 Web 工作区，并由桌面主进程启动和管理本地 `dsh web` 服务。
**本地 Harness Host**：桌面主进程启动 `dsh web`，等待本地服务就绪，并在应用退出时关闭 Host 进程。
本项目基于 [deepseek-ai/deepseek-harness](https://github.com/deepseek-ai/deepseek-harness) 的 Harness 核心、Cordis 插件体系和 Web 界面继续…
页面基于项目公开文档、仓库元数据和 DSH Plugins 的结构化解析生成；最后核验于 2026-08-21。发现错误？提交更正。
最佳 DeepSeek Harness 插件
从全目录挑出的 12 个值得优先安装的插件，覆盖各个分类。