DSH Desktop

A local-first, cross-platform desktop shell for DeepSeek Harness.

English · 简体中文

DSH Desktop packages the local DeepSeek Harness web experience as a desktop application. Choose a workspace and the app launches a local Harness instance, manages a random loopback port, persists profiles, plugins, and sessions, and opens the full interface as soon as Harness is ready.

[!IMPORTANT] DSH Desktop is currently an early preview and depends on the rapidly evolving @deepseek-ai/dsh@0.1.0-rc.6. Current builds are not code-signed or notarized by Apple and are not recommended for production use.

Download

Platform	Package	Download
macOS Apple Silicon	DMG installer	Download for Apple Silicon
macOS Intel	DMG installer	Download for Intel Mac
Windows x64	Setup installer	Download Windows installer
Windows x64	Portable executable	Download portable version

All current and historical packages are available on the GitHub Releases page.

Why this project exists

DeepSeek Harness already provides a complete agent runtime and Web UI. DSH Desktop does not reimplement Harness; it supplies the host capabilities needed for a desktop product:

• Run without manually starting a CLI or managing local ports
• Open workspaces with the native system directory picker and remember recent directories
• Manage the Harness child process, readiness checks, logs, and shutdown in one place
• Store profiles, plugins, and sessions outside the application installation directory so upgrades do not remove user data
• Provide packaging entry points for macOS and Windows

Features

• Opens directly into Harness without an additional landing page
• Prompts for a workspace on first launch and automatically restores the most recent workspace afterward
• Offers retry, workspace switching, log viewing, and exit actions when Harness fails to start
• Provides Workspace menu actions for opening a workspace, selecting a recent workspace, and restarting Harness
• Gracefully terminates the Harness child process when the desktop app exits
• Listens only on a random 127.0.0.1 port for each launch
• Removes Node.js privileges from the renderer and enables contextIsolation, sandboxing, and navigation restrictions
• Uses the DSH brand logo consistently in the desktop window and Harness sidebar
• Includes a production DSH app icon in macOS ICNS and Windows ICO formats

Model providers

During initial setup, you can choose a model provider and enter its API key directly. DSH Desktop uses the real Harness Settings and Credentials APIs: the key is written only to the credential store, the corresponding provider route is created automatically, and its built-in model catalog is inherited without requiring model IDs to be entered manually.

The initial setup currently includes:

Type	Providers
Model vendors	DeepSeek, OpenAI, Anthropic, Google Gemini, xAI, Moonshot/Kimi, MiniMax, Zhipu GLM, Mistral AI
Model aggregation	OpenRouter
Inference platforms	Groq, Together AI

Additional built-in or custom providers can be added from Settings → Models in Harness.

Quick start

Requirements

• Node.js 22 or later
• npm
• macOS on Apple Silicon or Intel, or Windows x64

Local development

git clone https://github.com/dataelement/dsh-desktop.git
cd dsh-desktop
npm install
npm run dev

npm install runs patch-package to reapply DSH Desktop's model-provider onboarding and sidebar branding, installs the brand asset, and then installs the Electron runtime.

Quality checks

npm test
npm run typecheck
npm run build

Packaging

# Generate unsigned DMG and ZIP artifacts for the current Mac architecture
npm run package:mac

# Run each command on a Mac or CI runner with the matching architecture
npm run package:mac:arm64
npm run package:mac:x64

# Generate NSIS and Portable artifacts on a Windows x64 machine or runner
npm run package:win

Harness includes architecture-specific native modules. Dependencies must be reinstalled and built on the matching platform for macOS ARM64, macOS Intel, and Windows x64. The architecture-specific scripts validate the current platform/arch before packaging to prevent artifacts that appear successful but are missing native dependencies.

Runtime architecture

DSH Desktop (Electron Main)
├── Native workspace picker and recent workspaces
├── Harness child-process lifecycle
├── Random loopback port and readiness checks
├── Native logging and recovery actions
└── Hardened BrowserWindow
     └── http://127.0.0.1:<random>  DeepSeek Harness Web UI

Electron userData
├── desktop-settings.json
├── logs/harness.log
└── harness/
    ├── profiles/
    ├── sessions/
    └── Plugins and user data

Harness runs in a separate Electron Node child process. The --expose-internals permission required by Cordis HMR is granted only to that child process and never to the web renderer.

Project structure

src/main/             Electron main process, windows, and Harness lifecycle
src/shared/           Shared runtime types
patches/              Reproducible UI customizations for the pinned DSH version
scripts/              Brand-asset installation and target-platform packaging checks
test/                 Settings, runtime, security, and provider coverage tests
build/                Application icon assets

Current validation status

• macOS Apple Silicon: development workflow, real Harness startup, DMG/ZIP packaging, and mounted artifacts verified
• macOS Intel: packaging configuration and platform checks provided; runtime verification still requires an Intel Mac or runner
• Windows x64: NSIS/Portable configuration and platform checks provided; runtime verification still requires a Windows runner
• Windows ARM64: not currently supported
• Code signing, Apple notarization, and automatic updates: not yet integrated

Upstream version and patches

The project currently pins @deepseek-ai/dsh@0.1.0-rc.6. The initial provider list is captured with patch-package under patches/ rather than relying on untracked changes in node_modules.

When upgrading DSH:

1. Verify the upstream Settings, Credentials, and Provider Directory contracts.
2. Reapply or rewrite the customized onboarding interface.
3. Regenerate the patch.
4. Run regression checks against a real Harness startup and provider configuration flow.

Contributing

Issues and pull requests are welcome. Before submitting a change, run at least:

npm test
npm run typecheck
npm run build

Never include real API keys in issues, logs, screenshots, or test data.

License

This project is open source under the MIT License.

DeepSeek Harness and its dependencies remain subject to their respective upstream licenses and trademark policies. DSH Desktop is an independent community desktop wrapper.