Living Dream DSH 🚀
Living Dream DSH — The Ultimate DeepSeek Harness Desktop Configuration
A battle-tested DSH configuration framework with 8+ MCP servers, custom plugins, free model access, mobile remote control, and more.
🆚 Living Dream DSH vs. Mainstream AI Coding Frameworks
Data as of 2026-08. Prices from official pages. Living-Dream-DSH is an open-source DSH config pack (MIT).
Overview
Why Choose Living-Dream-DSH?
-
Zero cost: The only "out-of-box and fully free" option. Claude Code costs $100–200/mo for heavy use; Copilot/Cursor start at $10–20/mo. This repo integrates multiple free model channels (AMD Radeon Cloud, DeepSeek signup bonus, Zhipu GLM-4V-Flash free vision), lowering the barrier to $0 while keeping BYOK for upgrades.
-
Full-stack capability: 8+ MCP servers out of the box — desktop automation, browser control, screen OCR, doc conversion, semantic memory, code execution. Others either don't support these or require manual per-server setup. One install = full toolkit, every component battle-tested (pitfalls documented in the troubleshooting guide).
-
Chinese ecosystem: The only full-Chinese one-stop solution. Complete Chinese docs, troubleshooting manual, and anti-pitfall guide. Commercial products and OSS projects are predominantly English, which is a real barrier for Chinese users.
-
Open source + self-built: MIT license — configs, scripts, custom plugin source all public. Local-first, API keys stay with you. No vendor lock-in vs. closed-source Claude Code/Copilot/Cursor; vs. OSS trio (OpenHands/Aider/Cline), adds the "pre-integrated config" layer.
-
Mobile office: Tailscale-based remote access from phone browser — a capability most CLI/IDE competitors (Aider, Cline, Cursor) lack.
Caveats
- Free channels depend on third-party policies; stability and quotas are subject to change. Heavy/commercial use recommended with BYOK or commercial subscriptions.
- This repo is a DSH config pack, not an independent agent kernel; the "framework ceiling" depends on official DSH capabilities.
- Comparison based on public info; some features (Claude Code computer use, Cline Browser Use) are still in Beta.
Quick Decision Guide
✨ Features
📁 Directory Structure
Living-Dream-DSH/
├── README.md # English
├── README.zh.md # Chinese
├── LICENSE # MIT License
├── .gitignore
├── setup.ps1 # Headless setup script (run by installer)
├── installer-v2.iss # Inno Setup script (online installer)
├── installer-v2-offline.iss # Inno Setup script (offline installer)
├── install-gui.ps1 # GUI installer (MPC-HC style wizard)
├── install-gui-offline.ps1 # Offline GUI installer (uses bundled deps)
├── install.ps1 # CLI installer (fallback)
├── install.bat # Installer launcher
├── package.json # repo scripts deps (http-proxy for proxy.js)
├── configs/
│ ├── cordis.patch.yml.template # MCP config template
│ ├── package.json.template # Plugin list template
│ ├── settings.yaml.template # Global settings template
│ ├── AGENTS.md # AI instructions
│ ├── .credentials.yaml.template # API Key template
│ └── README.md # Config file guide
├── scripts/
│ ├── start-dsh.bat.template # DSH launcher
│ ├── create-dsh-shortcut.ps1 # Desktop shortcut helper (post-install)
│ ├── proxy.js # Phone remote proxy (CORS rewrite)
│ ├── secrets.ps1 # DPAPI key decryption (reads secrets.json)
│ ├── os-copilot-mcp-server.py # OS-Copilot MCP (legacy location)
│ ├── os-copilot-mcp-README.md # OS-Copilot MCP docs
│ ├── os-copilot-mcp-LICENSE # OS-Copilot MIT license
│ └── mcp/ # MCP server scripts (all 6 bundled)
│ ├── dsh-history-server.py
│ ├── dsh-vision-server.py
│ ├── dsh-memory-server.py
│ ├── store_engine.py # Memory engine (SQLite + vector search)
│ ├── dsh-markitdown-server.py
│ ├── dsh-ocr-server.py
│ └── os-copilot-server.py
├── plugins/
│ ├── README.md # Plugin installation guide
│ └── dsh-paste-input/ # Custom file drag & paste plugin
│ ├── package.json
│ ├── lib/
│ │ ├── index.js
│ │ └── client.js
│ ├── cordis.patch.yml
│ ├── README.md
│ └── LICENSE
└── docs/
├── comparison.md # EN comparison table (7 frameworks)
├── comparison.zh.md # ZH comparison table
├── phone-remote.md # Mobile remote access tutorial
├── vision-patch.md # Image recognition patch
└── troubleshooting.md # Troubleshooting manual
🚀 Quick Start
Prerequisites (Must Install First)
💡 After installing Python, run
pip install mcp markitdown zstandardto enable all MCP servers. The GUI installer does this automatically.
💡 The launcher auto-detects DSH Desktop from common install paths. If it fails, set
$env:DSH_DESKTOP_PATH = "D:\Tools\DeepSeekHarness-Desktop"(your actual path), or add it as a system environment variable for persistence.
Option 1: Offline Install ⭐ Recommended
No internet required during install. Node.js, Python, Git are bundled in the package.
- Download
Living-Dream-DSH-v2.9.0-Offline-Setup.exe(~122 MB) from Releases - Double-click to run — professional Inno Setup wizard launches
- Follow the wizard: Language → License → Choose Install Location → Install
- After extraction, the headless setup script runs automatically (in a visible PowerShell window):
- Installs Node.js, Python, Git from bundled files (no download)
- Copies configs to
~/.dsh - Installs plugin dependencies
- Creates desktop shortcut
- Finish page: optional "Add DeepSeek Harness to desktop" checkbox
- After install: fill in API Keys in DSH Settings
💡 There is also an online installer
Living-Dream-DSH-v2.9.0-Setup.exe(~2 MB) that downloads dependencies via winget.
Option 2: GUI Installer (Source Build)
# 1. Clone repository
git clone https://github.com/alllllllllli/Living-Dream-DSH.git
cd Living-Dream-DSH
# 2. Double-click install.bat (launches GUI wizard)
# Or run in PowerShell: .\install-gui.ps1
The GUI installer walks you through:
- ✅ Welcome → License Agreement → Choose Install Location
- ✅ Auto-check environment (Node.js, Python, pnpm, Git)
- ✅ Copy config files to
~/.dsh - ✅ Install plugin dependencies
- ✅ Create desktop shortcut
- ✅ After install: fill in API Keys in DSH Settings
Option 3: Manual Install
Step-by-Step
# 1. Install prerequisites (if not installed)
winget install OpenJS.NodeJS.LTS # Node.js
winget install Python.Python.3.13 # Python
npm install -g pnpm # pnpm
# 2. Clone repository
git clone https://github.com/alllllllllli/Living-Dream-DSH.git
cd Living-Dream-DSH
# 3. Copy config files to DSH directory
Copy-Item configs\cordis.patch.yml.template $env:USERPROFILE\.dsh\profiles\web\cordis.patch.yml
Copy-Item configs\package.json.template $env:USERPROFILE\.dsh\profiles\web\package.json
Copy-Item configs\settings.yaml.template $env:USERPROFILE\.dsh\settings.yaml
Copy-Item configs\AGENTS.md $env:USERPROFILE\.dsh\AGENTS.md
Copy-Item configs\.credentials.yaml.template $env:USERPROFILE\.dsh\.credentials.yaml
# 4. Edit config files (fill in your API Keys)
notepad $env:USERPROFILE\.dsh\.credentials.yaml
# 5. Install plugins (dsh-paste-input is a file: dependency - bundled in repo)
Copy-Item plugins\dsh-paste-input $env:USERPROFILE\.dsh\profiles\plugins\dsh-paste-input -Recurse
cd $env:USERPROFILE\.dsh\profiles\web
pnpm install
# 6. Restart DSH Desktop
🔌 MCP Server List
All server scripts are bundled in
scripts/mcp/. GUI installer configures paths automatically.
🤖 Free Model Access
AMD Radeon Cloud
- Endpoint:
https://developer.amd.com.cn/radeon/api/v1 - Model: DeepSeek-V4-Flash
- Register: developer.amd.com.cn/radeon
TokenRhythm
- Endpoint:
https://tokenrhythm.studio/v1 - Protocol: OpenAI / Claude compatible, multi-model aggregation (DeepSeek-V4 Pro, GPT 5.5, Claude Opus 4.7, GLM 5.2, etc.)
- Free: join the "体验官计划" (pilot program) after signup for free token quota
- Register: https://tokenrhythm.studio/i/rf_tr_ZNVLkmDci5hWbRquWa6KP_fW
📱 Mobile Remote Access
Access DSH from phone browser via Tailscale:
# 1. Install Tailscale
winget install tailscale.tailscale
# 2. Login to same account
tailscale up
# 3. Start the rewrite proxy (must go through 8090 - DSH Web UI has CORS
# checks, connecting directly to 3080 fails)
npm install http-proxy # one-time dependency
node scripts/proxy.js # listens on 127.0.0.1:8090
# 4. Configure serve (point to the 8090 proxy, NOT 3080)
tailscale serve --https=443 --bg http://127.0.0.1:8090
# 5. Access from phone browser
# https://<your-device-name>.<your-domain>.ts.net
🖼️ Image Recognition Patch
Make DSH Desktop auto-call GLM-4V-Flash for image recognition:
# 1. Backup original file
$dshPath = (Get-Process DeepSeekHarness* -ErrorAction SilentlyContinue).Path
if (-not $dshPath) { $dshPath = "D:\Tools\DeepSeek-Harness-Desktop" }
Copy-Item "$dshPath\resources\runtime\node_modules\@deepseek-ai\dsh-host-apiproxy\lib\index.js" `
"$env:USERPROFILE\dsh-host-apiproxy-index.js.bak"
# 2. Apply patch (manually modify describeImagesLocally function around line 911)
# See docs/vision-patch.md
⚠️ Note: DSH upgrades will overwrite this patch, re-apply after upgrade.
🛡️ Troubleshooting
1. Plugin Installation
# ✅ Correct way
dsh plugin --profile web add <package>@<version>
# ❌ Wrong way (will clear unlisted bundles)
cd $env:USERPROFILE\.dsh\profiles\web
pnpm add <package>
2. Desktop vs Dev Version Conflict
Both share ~/.dsh, causing concurrent write conflicts. Close Dev version (port 3080) when using Desktop.
3. Image Patch
DSH upgrades will overwrite dsh-host-apiproxy modifications, re-apply patch.
4. billion-context-dsh Version
Must lock version 0.2.1, default will misparse as 0.1.7 (no dsh.bundle).
📝 Configuration Files
cordis.patch.yml
MCP server config file at ~/.dsh/profiles/web/cordis.patch.yml
# Example: Add an MCP server
- insert:
- id: mcp-dsh-example
name: '@deepseek-ai/dsh-mcp-client'
config:
transport: stdio
serverName: dsh-example
command: python
args:
- path/to/server.py
env: {}
cwd: ''
toolCallTimeoutMs: 120000
failOnStartupError: false
settings.yaml
Global settings file at ~/.dsh/settings.yaml
# Configure model providers
llm-pi-ai:
providers:
my-provider:
apiKeyEnv: MY_API_KEY
api: openai-completions
baseURL: https://api.example.com/v1
models:
- id: my-model
name: My Model
contextWindow: 1000000
.credentials.yaml
API Key storage at ~/.dsh/.credentials.yaml
MY_API_KEY: sk-xxxxxxxxxxxx
🔧 Custom Extensions
Add New MCP Server
- Create MCP server (reference existing implementations)
- Add config to
cordis.patch.yml - Add plugin to
package.jsonbundles (if needed) - Restart DSH Desktop
Create Custom Plugin
Reference dsh-file-uploads plugin structure.
📚 Resources
🤝 Contributing
Issues and Pull Requests are welcome!
📄 License
This project is licensed under the MIT License.
👤 Author
🙏 Acknowledgments
Living Dream DSH stands on the shoulders of these amazing open-source projects and services:
Core Platform
MCP Servers & Engines
AI Models & Services
DSH Plugins
Infrastructure & Tools
Last Updated: 2026-08-18 (Inno Setup installer)