---
license: mit
tags:
- offline-ai
- code-generation
- low-resource
- education
---

# 👻 Ghost Codex (Offline AI)

**Ghost Codex** is a lightweight, offline code assistant designed to run on legacy hardware (Windows XP/7, School Lab Computers) without internet access. 

It uses a **TF-IDF Neural Retrieval** mechanism instead of heavy Transformers, allowing it to run instantly with <50MB RAM usage.

## 📂 Project Structure
* **`ghost_brain.pkl`**: The pre-trained AI model (11,000+ coding patterns).
* **`ghost_coder.py`**: The inference app. Runs on any PC with Python/WinPython.
* **`rosetta_code_dataset.csv`**: The training dataset covering 22 major algorithms in Python, C++, and Java.

## 🚀 How to Run (Offline)
1. Download `ghost_coder.py` and `ghost_brain.pkl`.
2. Put them on a USB stick with Portable Python.
3. Run `python ghost_coder.py`.
4. Ask for code (e.g., "star pattern in C++").