The Open Source Claude Code is here, running on Windows, Linux, and macOS.
Run powerful AI coding agents from any computer — no installation required.
Plug in. Launch. Code.
Claude Code (Open Source) is the professional-grade, multi-platform alternative...
By bundling a pre-configured Node.js runtime, high-performance AI engines, and an advanced web dashboard, it delivers a seamless multi-platform experience across Windows, Linux, and macOS without ever touching the host system's global configuration.
- 🤖 6 Elite AI Providers — Integrated support for NVIDIA NIM (Free), OpenRouter, Google Gemini, Anthropic Claude, OpenAI, and Ollama (Local/Offline).
- 🌐 Advanced Web Dashboard — A premium, ChatGPT-style interface featuring autonomous agent modes, tool execution, and thinking process visualization.
- 🔧 Autonomous Coding Agent — A powerful open source Claude Code engine that performs file operations, executes shell commands, and analyzes codebases independently.
- 🔒 Privacy-First (Zero Footprint) — Absolute portability. No host system leaks; all session data, logs, and API keys remain encrypted on your portable drive.
- 🔄 Unified Cross-Platform Config — Configure your settings once and deploy instantly on any operating system — it just works.
- ⚡ Limitless Agentic Mode — Optional high-autonomy mode for rapid, hands-free development and complex task execution.
1. Download/clone this repo to a USB drive or folder
2. Double-click: Windows\Setup_First_Time.bat
3. Follow the prompts (picks provider, model, API key)
4. Done! Use Start_AI.bat or Open_Dashboard.bat anytime
cd Portable_AI_USB/Linux
chmod +x *.sh
./setup_first_time.shcd Portable_AI_USB/Mac
bash setup.shFirst-time setup requires internet to download Node.js (~25MB) and the AI engine (~5MB).
After that, everything runs offline (except API calls to your AI provider).
Portable_AI_USB/
│
├── 📂 dashboard/ ← Web Dashboard (shared across all platforms)
│ ├── server.mjs Node.js server with agent system
│ └── index.html Full-featured chat UI
│
├── 📂 data/ ← All portable data (shared across platforms)
│ ├── ai_settings.env Your API keys and model config
│ └── chats/ Saved conversations
│
├── 📂 Windows/ ← Windows scripts + Node.js binary
│ ├── Setup_First_Time.bat One-time setup
│ ├── Start_AI.bat Main launcher (CLI mode)
│ ├── Open_Dashboard.bat Web dashboard launcher
│ ├── Change_Model_or_Provider.bat
│ ├── Uninstall.bat
│ └── bin/ Node.js + engine (created by setup)
│
├── 📂 Linux/ ← Linux scripts
│ ├── setup_first_time.sh
│ ├── start_ai.sh
│ ├── open_dashboard.sh
│ ├── change_model_or_provider.sh
│ ├── uninstall.sh
│ └── bin/ (created by setup)
│
├── 📂 Mac/ ← macOS scripts
│ ├── setup.sh
│ ├── start_ai.sh
│ ├── open_dashboard.sh
│ ├── change_model_or_provider.sh
│ ├── uninstall.sh
│ └── bin/ (created by setup)
│
└── README.md
| Script | What it does |
|---|---|
| Setup_First_Time | Downloads Node.js, installs AI engine. Run once. |
| Start_AI | Configure provider → select model → launch CLI agent. Use --quick for instant limitless mode. |
| Open_Dashboard | Starts the web dashboard at http://localhost:3000 |
| Change_Model_or_Provider | Menu to change model, change API key, or full reset config |
| Uninstall | Removes engine (bin/) and optionally all data (settings + chats) |
| Provider | Free? | Setup |
|---|---|---|
| NVIDIA NIM | ✅ Free tier (1000 credits) | Get API key at build.nvidia.com |
| OpenRouter | ✅ Free + Paid models | Get API key at openrouter.ai |
| Google Gemini | ✅ Free tier available | Get API key at aistudio.google.com |
| Anthropic Claude | ❌ Paid only | Get API key at console.anthropic.com |
| OpenAI | ❌ Paid only | Get API key at platform.openai.com |
| Ollama | ✅ Fully free & offline | Install ollama.com, run ollama pull llama3.2:3b |
The web dashboard provides a ChatGPT-style interface with agent superpowers:
- Chat Mode — Simple conversation with streaming responses
- Agent Mode — AI can create files, run commands, search code
- Thinking Cards — See the AI's reasoning process (expand/collapse with ▼)
- Tool Execution — Visual cards showing what the agent is doing
- Normal Mode — Asks for approval before writing files or running commands
- Limitless Mode — Full autonomous execution
Launch it:
Windows: Open_Dashboard.bat
Linux: ./open_dashboard.sh
macOS: bash open_dashboard.sh
Then open http://localhost:3000 in your browser.
The data/ folder is shared across all platforms. This means:
- ✅ Set up your API key on Windows
- ✅ Plug the USB into a Linux machine — your settings are already there
- ✅ Move to a Mac — same thing, zero reconfiguration
Each platform only needs its own bin/ folder (created by running setup on that platform).
- Zero Footprint — No files are written outside the USB/project folder
- Portable Data —
CLAUDE_CONFIG_DIR,XDG_CONFIG_HOME, andXDG_DATA_HOMEare all redirected todata/ - API Key Masking — Keys are masked in all display output (e.g.,
sk-abc1****xyz9) - Approval System — In Normal mode, write operations require your explicit OK
- No Telemetry — Nothing is sent anywhere except your chosen AI provider
| Platform | Requirements |
|---|---|
| Windows | Windows 10+ (nothing else needed — Node.js is bundled) |
| Linux | curl (pre-installed on most distros). Git/Python recommended. |
| macOS | curl (pre-installed). For git: xcode-select --install |
Disk Space: ~150 MB for Node.js + engine per platform
| Problem | Solution |
|---|---|
| "Node.js not found" | Run Setup_First_Time first |
| "Port 3000 in use" | Dashboard is already running, or another app uses port 3000 |
| API key rejected | Double-check your key at the provider's website |
| "No internet" | Setup requires internet. After setup, only API calls need internet. |
| Models not loading | OpenRouter API may be slow. Try again, or enter a model name manually. |
Data leaking to ~/.openclaude |
Make sure you launch via start_ai.sh / Start_AI.bat — they set CLAUDE_CONFIG_DIR to keep data portable. |
MIT — Use it however you want.