techjarves/Portable-AI-USB

[!IMPORTANT]

This repository (Portable Uncensored AI ) is deprecated and is no longer actively maintained.

All active development, features, and bug fixes (including the Windows crash/access violation fixes) have moved to:

### 👉 Uncensored-Local-Studio

Please visit the new repository to download the latest version, report issues, and follow the project's development.

🔒 Portable Uncensored AI — Runs Entirely from a USB Drive

A fully private, portable, uncensored AI assistant that runs 100% from a USB flash drive. No internet needed after setup. No data leaves the USB. Works on Windows, Mac, and Linux.

Now with multi-model support! Choose from 6 curated AI models or bring your own.

📺 Watch the Tutorial

⚡ Available Models

During installation, you'll choose which model(s) to download:

| # | Model | Size | Label | Best For | |---|-------|------|-------|----------| | 1 | NemoMix Unleashed 12B | ~7.0 GB | 🔓 UNCENSORED | ⭐ Recommended — best quality uncensored | | 2 | Dolphin 2.9 Llama 3 8B | ~4.9 GB | 🔓 UNCENSORED | Classic uncensored all-rounder | | 3 | Mistral 7B Instruct v0.3 | ~4.1 GB | 🔒 STANDARD | Strong reasoning & coding | | 4 | Qwen 2.5 7B Instruct | ~4.7 GB | 🔒 STANDARD | Great multilingual support | | 5 | Llama 3.2 3B Instruct | ~2.0 GB | 🔒 STANDARD | Lightweight — fast on old PCs | | 6 | Phi-3.5 Mini 3.8B | ~2.2 GB | 🔒 STANDARD | Lightweight — good reasoning | | C | Custom GGUF | Varies | 🎨 CUSTOM | Bring your own HuggingFace model |

🔓 UNCENSORED = No content filters, answers everything

🔒 STANDARD = Normal safety guidelines apply

🚀 Setup (One Time Only)

What You Need

• A USB flash drive with at least 16 GB of free space (32 GB recommended for multiple models)
• Format the USB as exFAT (works on Windows, Mac, and Linux)
• An internet connection for the initial download

Steps

1. Download this repo and copy ALL files to your USB drive 2. Double-click install.bat on the USB drive 3. Choose your model(s) from the interactive menu 4. Interactive AnythingLLM Setup:

• The AnythingLLM installer will open automatically.
• IMPORTANT: When asked for the "Install Location", click Browse and select the anythingllm folder on your USB drive.
• Wait for it to finish, then close the installer window.

⚠️ If a Model Download Fails

The installer automatically retries failed downloads. If it still fails:

1. Download the model manually from the HuggingFace resolved link shown in the console error. 2. Place the .gguf file into the models\ folder on your USB. 3. Re-run install.bat — it will detect the file and skip the download.

🔄 Adding More Models Later

Just re-run install.bat and select additional models. Already-downloaded models are automatically skipped.

🎨 Custom Models

Want a model not on the list? During install, choose option C and paste any direct .gguf download link from HuggingFace. The installer handles the rest!

🗄️ Configuring Ollama Token Limit / Context Size

By default, the installer configures Ollama with a 4K token limit for optimal performance on most PCs. If you want to adjust this, follow these steps after install script has finished running:

1. Open this folder on your USB drive. anythingllm_data/storage 2. Open the .env file in a text editor. 3. Find the line that says OLLAMA_MODEL_TOKEN_LIMIT=4096 and change 4096 to your desired token limit (e.g., 8192 for 8K tokens). 4. Save the file and restart the AI using the launcher script start-windows.bat or start-mac.command or start-linux.sh. 5. Incase if you re-run the installer script install.bat or install.sh it will reset the token limit back to 4096, so you will need to change it again in the .env file. 6. For mac, you may re-run the start-mac.command script and it will not overwrite the token limit in the .env file, so you can just restart the AI using this script after changing the token limit in the .env file.

▶️ How to Use

On Windows

• Double-click start-windows.bat on the USB drive.
• Improved Portability: The launcher now automatically clears old path caches. This allows you to move between different computers without "JavaScript errors."
• The AnythingLLM chat window will open automatically.
• Switch between models in AnythingLLM: Settings → LLM → select your model.
• Keep the black terminal window open while chatting.
• Press any key in the terminal to safely shut down.

On Mac

• Double-click start-mac.command on the USB drive.
• First time: It will automatically download the Mac engine (~2 min).
• The AnythingLLM window will open automatically.
• Press ENTER in the terminal to safely shut down.

🐧 On Linux

1. Open a terminal on your USB drive. 2. Run the launcher: make it executable first:

   chmod +x start-linux.sh preflight-check.sh install.sh install-core.sh 
   

   bash preflight-check.sh
   

3. The preflight script checks for everything and executes the install script. 4. open ANYTHING LLM folder and open the appImage there 4. Switch between models: Settings → LLM → select your model. 5. Keep the terminal open while chatting. 6. Press Enter in the terminal to safely shut down.

---

🔐 Privacy

• All chats & settings stay on the USB — never saved to the host PC.
• No registry keys or local files are left behind.
• Works completely offline after initial setup.
• No telemetry, no cloud, no tracking.

📁 USB Drive Structure (After Setup) - WINDOWS & MAC

USB Drive/
├── install.bat             ← Run this first (one time only)
├── install-core.ps1        ← Setup script (called by install.bat)
├── start-windows.bat       ← Windows launcher (with auto-cache clearing)
├── start-mac.command       ← Mac launcher
├── ollama/                 ← AI engine (Windows)
├── models/                 ← AI model files (.gguf) & configs
│   ├── installed-models.txt    ← List of installed models
│   └── *.gguf                  ← Model weights
├── anythingllm/            ← Your AI Interface (installed here)
├── installer_data/         ← Temporary installer files (auto-cleaned)
└── anythingllm_data/       ← Your chats & settings (100% portable!)

📁 USB Drive Structure (After Setup) - LINUX

USB Drive/
├── install.sh             ← Linux / Mac installer
├── preflight-check.sh     ← Linux USB drive health check [ run this first ] 
├── install-core.sh        ← Core setup logic (called by install.sh)
└── start-linux.sh         ← Linux launcher
├── ollama/                 ← AI engine (Windows)
├── models/                 ← AI model files (.gguf) & configs
│   ├── installed-models.txt    ← List of installed models
│   └── *.gguf                  ← Model weights
├── anythingllm/            ← Your AI Interface (installed here)
├── installer_data/         ← Temporary installer files (auto-cleaned)
└── anythingllm_data/       ← Your chats & settings (100% portable!)

---

💾 USB Size Guide

| Models | Minimum USB | Recommended USB | |--------|-------------|-----------------| | 1 lightweight (3B/3.8B) | 16 GB | 16 GB | | 1 recommended (NemoMix 12B) | 16 GB | 32 GB | | 2-3 models | 32 GB | 64 GB | | All 6 presets (~25 GB) | 64 GB | 64 GB |

⚠️ Important Notes

• Manual Path Selection: When installing AnythingLLM, you must manually select the anythingllm folder on the USB to keep it portable.
• Moving between PCs: If you see a "JavaScript Error" on a new PC, just close it and run start-windows.bat again. The script will automatically wipe the old PC's cached paths and fix the run.
• Performance: The AI runs on your CPU — responses take 10–30 seconds depending on hardware.
• RAM: 12B models (NemoMix) need at least 8 GB RAM. 7B models need at least 6 GB RAM.
• Always safely eject the USB before unplugging.

📜 License

MIT License — See LICENSE for details.