Portable Uncensored AI — Runs Entirely from a USB Drive

This project delivers a fully private, portable, and uncensored AI assistant that operates 100% from a USB flash drive on both Windows and Mac. It targets users requiring AI capabilities without internet access or data persistence on host machines, offering a zero-trace solution for on-the-go AI tasks.

How It Works

The system integrates Ollama as the AI engine, pre-configured with various models like Llama 3 and NemoMix, alongside the AnythingLLM interface for chat. It's designed to run exclusively from a USB drive, ensuring no data persists on the host computer. Installation involves copying files to a USB, running a setup script, and selecting AI models, with AnythingLLM being manually pointed to its USB directory for complete portability.

Quick Start & Requirements

• Primary install / run command: Copy all files to a USB drive (formatted as exFAT). Run install.bat (Windows) or install.command (Mac) for one-time setup. Launch using start-windows.bat or start-mac.command.
• Non-default prerequisites: USB flash drive (16GB minimum, 32GB recommended), internet connection for initial download.
• Links: A tutorial video is mentioned but no URL is provided.

Highlighted Details

• Model Variety: Offers 6 curated models (NemoMix Unleashed 12B, Dolphin 2.9 Llama 3 8B, Mistral 7B, Qwen 2.5 7B, Llama 3.2 3B, Phi-3.5 Mini 3.8B) with options for uncensored or standard content filters, plus custom GGUF model support.
• Zero-Trace & Offline: Operates entirely offline after setup, leaving no data, registry keys, or local files on the host PC.
• Cross-Platform Portability: Works on both Windows and Mac, with launchers that automatically clear old path caches to prevent JavaScript errors when moving between computers.
• Storage Guide: Provides clear USB size recommendations based on the number and size of selected models.

Maintenance & Community

No specific details on contributors, sponsorships, or community channels (like Discord/Slack) are provided in the README.

Licensing & Compatibility

• License type: MIT License.
• Compatibility notes: The MIT license generally permits commercial use and linking with closed-source projects, but users should consult the full LICENSE file for specifics.

Limitations & Caveats

AI responses are CPU-bound, with typical generation times of 10–30 seconds. Larger models (e.g., 12B) require at least 8GB of RAM, and 7B models need at least 6GB. Users must remember to safely eject the USB drive before unplugging.