[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-29":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":14,"stars7d":16,"stars30d":17,"stars90d":15,"forks30d":15,"starsTrendScore":18,"compositeScore":19,"rankGlobal":9,"rankLanguage":9,"license":20,"archived":21,"fork":21,"defaultBranch":22,"hasWiki":23,"hasPages":21,"topics":24,"createdAt":9,"pushedAt":9,"updatedAt":25,"readmeContent":26,"aiSummary":27,"trendingCount":15,"starSnapshotCount":15,"syncStatus":28,"lastSyncTime":29,"discoverSource":30},29,"OpenClaude-Portable","techjarves\u002FOpenClaude-Portable","techjarves","Run Claude Code from a USB drive on any PC — no installation required.",null,"HTML",861,330,20,15,0,32,220,45,11.56,"MIT License",false,"main",true,[],"2026-06-12 02:00:06","# OpenClaude — Portable AI Coding Agent\n\n> **Run a full-featured AI coding agent from a USB drive or any folder — no installation required.**\n> Plug in. Launch. Code. Take it anywhere.\n\n[![License: MIT](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-blue.svg)](LICENSE)\n[![Platform](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fplatform-Windows%20%7C%20Linux%20%7C%20macOS-lightgrey.svg)]()\n\n\n**🎥 Watch the Setup & Demo Video:** [https:\u002F\u002Fyoutu.be\u002F9Dh3kKWFFjg](https:\u002F\u002Fyoutu.be\u002F9Dh3kKWFFjg)\n\n[![OpenClaude Portable Demo](https:\u002F\u002Fimg.youtube.com\u002Fvi\u002F9Dh3kKWFFjg\u002Fmaxresdefault.jpg)](https:\u002F\u002Fyoutu.be\u002F9Dh3kKWFFjg)\n\n---\n\n## What Is This?\n\n**OpenClaude Multi-Platform** is a fully portable AI coding agent powered by the open-source [OpenClaude](https:\u002F\u002Fgithub.com\u002Fgitlawb\u002Fopenclaude) engine. It bundles a self-contained Node.js runtime, a smart system-prompt proxy for local models, and a web-based dashboard — all configurable from a single `START.bat` (Windows) or `start.sh` (Linux\u002FmacOS).\n\nEverything runs strictly inside the project folder. No files are written to the host machine.\n\n---\n\n## Key Features\n\n| Feature | Details |\n|---|---|\n| **7 AI Providers** | NVIDIA NIM · DeepSeek · OpenRouter · Google Gemini · Anthropic Claude · OpenAI · Ollama (offline) |\n| **Zero Footprint** | All data, keys, and logs stay inside `data\u002F` — nothing touches the host system |\n| **Local Speed Proxy** | Trims system prompts by up to 90% before sending to Ollama, dramatically improving response time on CPU-only hardware |\n| **Auto-Update Cache** | Checks for engine updates once per day (skips the network call on repeat launches) |\n| **Session Resume** | Resume any interrupted session with `RESUME.bat \u003Csession-id>` |\n| **Web Dashboard** | ChatGPT-style browser UI with agent mode, tool cards, and thinking visualisation |\n| **Limitless Mode** | Optional full-autonomy mode — the agent runs without asking for approval |\n| **Cross-Platform** | Shared `data\u002F` folder works across Windows, Linux, and macOS |\n\n---\n\n## Quick Start\n\n### Windows\n```\n.\\START.bat\n```\nOn first run it automatically downloads Node.js (~25 MB) and the OpenClaude engine (~5 MB), then walks you through provider selection. Every subsequent launch skips setup and goes straight to the menu.\n\n### Linux \u002F macOS\n```bash\nchmod +x start.sh\n.\u002Fstart.sh\n```\n\n> **First-time setup requires internet.** After that, only API calls need a connection (or none at all if you use Ollama offline mode).\n\n---\n\n## Project Structure\n\n```\nOpenClaude-Multi-Platform\u002F\n│\n├── START.bat                  Windows entry point — handles everything\n├── start.sh                   Linux\u002FmacOS entry point\n├── RESUME.bat                 Resume a previous session by ID (Windows)\n│\n├── data\u002F                      All persistent data (shared across platforms)\n│   ├── ai_settings.env        Active provider, model, and API key\n│   ├── openclaude\u002F            Session history and agent memory\n│   ├── ollama\u002F                Local Ollama binary and model storage\n│   └── proxy.log              Speed proxy activity log (silent background)\n│\n├── engine\u002F                    Node.js runtime + OpenClaude npm package\n│   ├── node-win-x64\u002F          Bundled Node.js (Windows)\n│   └── node_modules\u002F\n│       └── @gitlawb\u002Fopenclaude\u002F\n│\n├── tools\u002F                     Helper scripts\n│   ├── local-proxy.js         System-prompt trimming proxy for local models\n│   ├── setup_local_models.ps1 Ollama model downloader (Windows)\n│   ├── setup_local_models.sh  Ollama model downloader (Linux\u002FmacOS)\n│   ├── Change_Provider.bat    Switch AI provider or API key (Windows)\n│   ├── change_provider.sh     Switch AI provider or API key (Linux\u002FmacOS)\n│   ├── Open_Dashboard.bat     Launch web dashboard (Windows)\n│   ├── open_dashboard.sh      Launch web dashboard (Linux\u002FmacOS)\n│   └── Setup_Local_Models.bat Wrapper launcher for local model setup\n│\n└── dashboard\u002F                 Web dashboard UI\n    ├── server.mjs             Dashboard Node.js server\n    └── index.html             Chat interface\n```\n\n---\n\n## Main Menu Options\n\nWhen you run `START.bat`, you are presented with:\n\n```\n1) Launch AI       — Normal Mode      (asks before writing files or running commands)\n2) Limitless Mode  — Auto-executes    (fully autonomous, no approval prompts)\n3) Open Dashboard  — Web UI at http:\u002F\u002Flocalhost:3000\n4) Change Provider — Switch model or API key\n5) Setup Offline   — Download local Ollama models\n```\n\nThe menu auto-selects **Normal Mode** after 10 seconds if no key is pressed.\n\n\n\n\n## Supported AI Providers\n\n| Provider | Cost | API Key |\n|---|---|---|\n| **NVIDIA NIM** | Free tier (1 000 credits\u002Fmonth) | [build.nvidia.com](https:\u002F\u002Fbuild.nvidia.com) |\n| **DeepSeek** | Paid API | [platform.deepseek.com](https:\u002F\u002Fplatform.deepseek.com) |\n| **OpenRouter** | Free + paid models | [openrouter.ai](https:\u002F\u002Fopenrouter.ai) |\n| **Google Gemini** | Free tier available | [aistudio.google.com](https:\u002F\u002Faistudio.google.com) |\n| **Anthropic Claude** | Paid | [console.anthropic.com](https:\u002F\u002Fconsole.anthropic.com) |\n| **OpenAI** | Paid | [platform.openai.com](https:\u002F\u002Fplatform.openai.com) |\n| **Ollama** | Free, fully offline | [ollama.com](https:\u002F\u002Follama.com) |\n\n---\n\n## Local Model Performance (Ollama)\n\nRunning a local model on CPU or USB 2.0 is inherently slower than a cloud API. The built-in **speed proxy** (`tools\u002Flocal-proxy.js`) intercepts every request and trims the OpenClaude system prompt from ~10 000 tokens down to ~300 tokens before it reaches Ollama.\n\n**Typical result:** first-token latency drops from 60–120 s to 5–20 s on CPU-only hardware.\n\nProxy activity is logged silently to `data\u002Fproxy.log` — it never writes to the terminal.\n\n**Recommended models for CPU inference:**\n\n| Model | Size | Speed |\n|---|---|---|\n| `gemma3:1b` | ~800 MB | Fastest |\n| `qwen2.5:1.5b` | ~1 GB | Fast |\n| `phi3:mini` | ~2.3 GB | Moderate |\n\n> For best performance, copy `data\u002Follama\u002F` to your local SSD if USB 2.0 read speeds are the bottleneck.\n\n---\n\n## Security & Privacy\n\n- **Zero Footprint** — `XDG_CONFIG_HOME`, `XDG_DATA_HOME`, and `CLAUDE_CONFIG_DIR` are all redirected to `data\u002F`, keeping the host system clean.\n- **No Telemetry** — Nothing is sent anywhere except your chosen AI provider.\n- **API Key Safety** — Keys are stored only in `data\u002Fai_settings.env` on your drive.\n- **Approval Mode** — In Normal Mode the agent asks before any file write or shell command.\n\n---\n\n## System Requirements\n\n| Platform | Requirement |\n|---|---|\n| **Windows** | Windows 10 or later — Node.js is bundled, nothing else needed |\n| **Linux** | `curl` (pre-installed on most distros) |\n| **macOS** | `curl` (pre-installed) |\n\n**Disk space:** ~150 MB for Node.js + engine. Local Ollama models require additional space (800 MB–8 GB depending on model).\n\n---\n\n## Troubleshooting\n\n| Symptom | Fix |\n|---|---|\n| `Node.js not found` | Run `START.bat` first — it downloads Node automatically |\n| `EADDRINUSE: port 11435` | The speed proxy from a previous session is still running. Restart `START.bat` — it kills it automatically |\n| `openclaude: dist\u002Fcli.mjs not found` | The engine install was interrupted. Pull the latest launcher and run `START.bat` again; it will repair incomplete installs automatically |\n| `npm error could not determine executable to run` | Pull the latest launcher. The app now runs the verified bundled OpenClaude binary instead of falling back to `npx` |\n| `Claude Code on Windows requires git-bash` | Pull the latest launcher and run `START.bat` again; it installs\u002Frepairs bundled GitPortable and adds Git Bash to the launch environment |\n| `'D_ARGS' is not recognized` | Old version of START.bat with nested if-blocks. Pull the latest version |\n| Ollama response is very slow | Use a smaller model (`gemma3:1b`), or copy models to a local SSD |\n| API key rejected | Verify your key at the provider's website; re-run option 4 to update it |\n| Port 3000 already in use | The dashboard is already running — open `http:\u002F\u002Flocalhost:3000` directly |\n| `openclaude` not found in PowerShell | Use `.\\RESUME.bat \u003Csession-id>` instead of calling `openclaude` directly |\n\n---\n\n## License\n\nMIT — use it, fork it, ship it.\n","OpenClaude-Portable 是一个可以从USB驱动器或任何文件夹运行的全功能AI编码助手，无需安装。该项目集成了Node.js运行环境、智能系统提示代理和基于Web的控制面板，支持NVIDIA NIM、DeepSeek等七种AI提供商，并且具有零足迹特性，所有数据、密钥和日志都存储在项目文件夹内，不写入主机系统。特别适合需要跨平台（Windows、Linux、macOS）使用的开发者，尤其当工作环境频繁变化时，仅需插入USB设备即可快速启动编码辅助服务。此外，通过本地速度代理优化了CPU硬件上的响应时间，提供自动更新缓存和会话恢复等功能，确保开发工作的连续性和高效性。",2,"2026-06-11 02:30:31","CREATED_QUERY"]