🖥️ AI in the Shell

AI in the Shell is an experimental project that gives a local AI model (via Ollama) unrestricted root access to a Unix shell. The AI takes natural language instructions, converts them into shell commands, and executes them on the system — no restrictions, no holding back.

⚠️ WARNING: This project is highly experimental and should only be run inside an isolated virtual machine (VM). Misuse can result in system damage, data loss, or unintended consequences.

---

🚀 Features

• Full root shell access for LLMs
• Natural language to shell command conversion
• Real-time command execution with output logging
• Local-only, runs fully offline via Ollama
• Lightweight Python Flask backend

---

🧠 How It Works

1. You send a prompt like:

Install nginx and start the service

2. The LLM (e.g., llama3) responds with:

sudo apt update && sudo apt install nginx -y && sudo systemctl start nginx

3. The shell executor runs the command and returns the output.

---

🛠️ Requirements

• Python 3.8+
• Ollama (installed and running)
• A virtual machine (e.g., Kali Linux, Ubuntu, etc.)

---

⚙️ Installation

1. Clone the repo:

   git clone https://github.com/yourusername/AI-in-the-Shell.git
   cd AI-in-the-Shell

2. Install dependencies:

   pip install flask requests

3. Install Ollama and run a model:

   curl -fsSL https://ollama.com/install.sh | sh
   ollama run llama3

4. Run the Flask server:

   python3 rootshell_combined.py

---

📡 Usage

Use curl or any HTTP client to interact with the API:

curl -X POST http://localhost:4224/run \
-H "Content-Type: application/json" \
-d '{"prompt": "Install nginx and start the service"}'

Response:

{
  "command": "sudo apt update && sudo apt install nginx -y && sudo systemctl start nginx",
  "output": "Hit:1 http://... nginx is already the newest version...",
  "user_input": "Install nginx and start the service"
}

---

📌 Notes

• You must run this in an isolated VM with sudo access.
• The AI has full control — treat this like giving a human root shell access.
• Designed for experimentation and research, not production.

---

🧪 Ideas for Expansion

• Web UI with terminal-like feedback
• Command history and approval mode
• Live output streaming
• Prompt-based session memory

---

⚠️ Disclaimer

This project is for educational and experimental purposes only. You are responsible for the consequences of running code generated by an AI with system-level permissions.

Use wisely — preferably in a VM that you can restore or wipe.

---

📄 License

MIT License. See LICENSE file for details.