DeepSeek Cowork

English | 中文文档

---

DEMO

deepseek-cowork-1.mp4

Why This Project?

On January 13, 2026, Anthropic released Claude Cowork:

"Introducing Cowork: Claude Code for the rest of your work."

Great product, but:

	Claude Cowork	DeepSeek Cowork
Price	💰 Expensive	✅ Ultra-low cost
Accessibility	🔒 Complex setup, regional restrictions	✅ Ready to use
Open Source	❌ Proprietary	✅ Fully open source
Self-hosting	❌ Not supported	✅ Supports private deployment

We want everyone to have access to a great AI assistant — so we built this.

Why DeepSeek?

Solid Baseline	Ultra Affordable	Fully Open
Reliable performance among open-source LLMs	Most competitive API pricing	Supports local deployment & customization

Core Philosophy

Open-source models will eventually catch up with closed-source.

We believe it's only a matter of time. Rather than wait, we're building the infrastructure now.

When open-source models reach parity, DeepSeek Cowork will be ready.

Why Now?

This would have been impossible before. But two things changed:

1. AI Coding explosion - Dramatically reduced development costs, enabling individuals to build complex applications
2. Engineering bridges the gap - Prompt engineering, skill systems, and context management can enhance the experience on existing models

What Can It Do?

Use natural language to have AI help you with:

• 🤖 Task Coordination - Turn natural language requests into multi-step AI workflows
• 📁 File Management - Browse, organize, and preview your workspace files
• 🧠 Persistent Memory - AI remembers conversation context, understands your habits

Typical Scenarios

Scenario	Example
Task Delegation	"Review this module, summarize the risks, and suggest a safe refactor plan"
Project Setup	"Create the project structure and prepare the initial configuration files"
Content Organization	"Sort files in my downloads folder by type"
Session Follow-up	"Summarize what changed today and remind me what to do next"

💡 Like having a 24/7 digital assistant at your command

---

Technical Documentation

Architecture Highlights

DeepSeek Cowork adopts a unique Hybrid SaaS architecture, combining the best of cloud-based SaaS and local desktop applications:

┌─────────────────────────────────────────────────────────────────┐
│                        User's Computer                          │
│  ┌──────────────┐    ┌──────────────┐    ┌──────────────────┐  │
│  │   Electron   │    │  Web Browser │    │   CLI Tool       │  │
│  │   Desktop    │    │  (Chrome,    │    │ deepseek-cowork  │  │
│  │     App      │    │   Edge...)   │    │                  │  │
│  └──────┬───────┘    └──────┬───────┘    └────────┬─────────┘  │
│         │ IPC               │ HTTP/WS             │ manage     │
│         └───────────────────┼─────────────────────┘            │
│                             ▼                                   │
│                    ┌────────────────┐                          │
│                    │  LocalService  │◄── All data stays local  │
│                    │  (Node.js)     │                          │
│                    └────────┬───────┘                          │
└─────────────────────────────┼───────────────────────────────────┘
                              │ Encrypted
                              ▼
                    ┌────────────────┐
                    │   Happy AI     │
                    │   (Cloud)      │
                    └────────────────┘

Feature	Benefit
Zero Server Cost	Static frontend hosted on GitHub Pages, no backend infrastructure needed
Data Privacy	All user data, settings, and files remain on your local machine
Unified Experience	Same UI/UX whether using Desktop app or Web browser

How It Works

1. Desktop Mode: Electron app communicates with LocalService via IPC
2. Web Mode: Browser connects to LocalService via HTTP/WebSocket on localhost:3333
3. CLI Mode: Manage LocalService directly from terminal

The ApiAdapter layer automatically detects the environment and routes API calls appropriately.

Happy Integration

DeepSeek Cowork integrates with Happy, an open-source mobile and web client for AI coding agents.

Feature	Description
End-to-End Encryption	All messages are encrypted locally before transmission - your data never leaves your device unencrypted
Mobile Access	Use the Happy App (iOS / Android) to monitor and control AI tasks on the go
Push Notifications	Get alerted when AI needs permission or encounters errors
Open Source	Fully auditable code with no telemetry or tracking

DeepSeek Cowork uses Happy's account server for session management and encrypted sync across devices.

Core Components

Component	Description
Claude Code	Original Claude Code integrated as Agent kernel with all features and capabilities
Happy	Open-source AI session management with E2E encryption and mobile app support
LocalService	Local runtime that powers the desktop app, web UI, and CLI workflows
Electron App	Cross-platform desktop interface integrating all components

Quick Start

git clone https://github.com/imjszhang/Deepseek-Cowork.git
cd deepseek-cowork
npm install
npm start

Development mode: npm run dev

Runtime Entry Points

• npm start launches the Electron desktop app.
• npm run local-service starts the supported LocalService runtime directly.
• npm run server is kept as a compatibility alias for npm run local-service.
• server/index.js is deprecated and is no longer a supported standalone entry point.

Web Version (Hybrid SaaS)

Use DeepSeek Cowork directly in your browser without installing the desktop app.

Online Demo

Visit deepseek-cowork.com to try the web interface.

Prerequisites

• Node.js 18+
• npm or yarn

Setup Local Service

# Install CLI tool globally (latest version: 0.3.0)
npm install -g deepseek-cowork@0.3.0

# Or start the same LocalService from the repository
npm run local-service

# Start local service (background mode)
deepseek-cowork start --daemon

# Open web interface in browser
deepseek-cowork open

CLI Commands Reference

CLI Version: deepseek-cowork@0.3.0

Command	Description
deepseek-cowork start	Start local service (foreground)
deepseek-cowork start --daemon	Start local service (background)
deepseek-cowork stop	Stop local service
deepseek-cowork status	Check service status
deepseek-cowork open	Open web interface in browser
deepseek-cowork config	View/edit configuration
deepseek-cowork deploy	Deploy skills to work directories
deepseek-cowork module	Manage server modules

Deploy Skills

# Deploy built-in skills to work directories
deepseek-cowork deploy

# Deploy with Chinese templates
deepseek-cowork deploy --lang zh

# Deploy custom skill from any path
deepseek-cowork deploy --from ./my-skill --target my-project

# Check deployment status
deepseek-cowork deploy status

Manage Server Modules

# List available modules
deepseek-cowork module list

# Deploy a module
deepseek-cowork module deploy demo-module

# Deploy custom module from any path
deepseek-cowork module deploy my-module --from ./my-module-source

# Check deployed modules status
deepseek-cowork module status

Build Web Version

# Build static files for web deployment
npm run build:web

# Verify docs/app is still in sync with renderer sources
npm run check:web-sync

# Output: docs/app/

The web frontend is deployed to GitHub Pages automatically.

Building Desktop Clients

Build standalone installers for Windows, macOS, and Linux:

# Build for current platform
npm run build

# Build for specific platform
npm run build:win    # Windows (NSIS installer + portable)
npm run build:mac    # macOS (DMG, Intel + Apple Silicon)
npm run build:linux  # Linux (AppImage, deb, rpm)

# Build for all platforms
npm run build:all

Built packages will be output to the dist/ directory.

Version Management

The project uses semantic versioning (SemVer). Current version: V0.3.0

Update version numbers:

npm run version:patch   # 0.1.0 → 0.1.1 (bug fixes)
npm run version:minor   # 0.1.0 → 0.2.0 (new features)
npm run version:major   # 0.1.0 → 1.0.0 (breaking changes)

The version number is automatically synchronized to:

• package.json - Source of truth
• renderer/index.html - UI display (auto-updated during build)
• Application runtime - Dynamically loaded from package.json

Contributing

PRs welcome! Fork → Change → Submit.

License

MIT

Acknowledgments

This project is built upon:

• Happy - AI session management client
• Electron - Cross-platform desktop framework
• DeepSeek - Open-source LLM

---

Making great AI assistants accessible to everyone