Hey everyone,

I’ve been working on a small passion project for the past few weeks and wanted to share it here: Lumina AI.

It started as me reviving an old archived project called Shijima-Qt—a simple Qt desktop mascot. I always liked the idea, so I tried to push it further and turn it into something more useful: a lightweight, offline AI desktop agent.

I’m a 17-year-old self-taught developer from Indonesia, and this project has basically been my way of learning more about C++, desktop applications, and local AI systems by building something far beyond what I originally thought I could handle.

Some of the things I ended up implementing include:

• Local AI integration through Ollama
• A native tool-calling system written in C++
• Desktop and active-window awareness
• Basic file and system automation
• Lightweight web scraping using Qt networking
• Persistent JSON-based memory between sessions

The project is built with C++17, Qt6, Make, and Ollama, with a strong focus on staying lightweight and running entirely offline.

One thing I tried to be careful about was preserving the original project's history. The repository still contains the full commit history from Shijima-Qt, since I see this more as continuing an abandoned idea rather than replacing it.

The project is open source under GPLv3.

I'm still learning, so I'd be especially interested in hearing how more experienced C++ and AI developers would approach some of these problems differently. If you spot architectural mistakes, questionable design decisions, or areas that could be improved, I'd genuinely love to learn from the feedback.

Repository:
https://github.com/Lumina403/Lumina-AI