A small passion project by a 17-year-old self-taught developer has breathed new life into an abandoned desktop mascot, turning it into a lightweight, offline AI companion. The project, named Lumina AI, started as a revival of Shijima-Qt, a simple Qt6-based desktop companion, but quickly evolved into something far more ambitious.
Lumina AI now integrates a local AI core powered by Ollama’s Qwen model, enabling users to interact with an autonomous agent that operates entirely offline. The developer sought to learn modern C++ practices, desktop application development, and local AI integration by building a system that combines nostalgia with cutting-edge functionality. The result? A project that balances lightweight performance with powerful capabilities.
From desktop mascot to offline AI assistant
The original Shijima-Qt project was a minimalist desktop mascot—a common trope in software history—designed to sit idle on a user’s screen. While nostalgic, it lacked practical utility beyond aesthetics. The developer saw an opportunity to repurpose the codebase, transforming it into Lumina AI, an offline AI agent that responds to user commands and automates tasks without relying on cloud services.
The transition required several key upgrades:
- Replacing the static mascot with an active AI agent
- Integrating Ollama’s Qwen model for local inference
- Adding system and window awareness to contextualize interactions
- Implementing persistent memory using JSON files for continuity between sessions
The project maintains the original repository’s full commit history, ensuring transparency and acknowledging the project’s roots as a continuation rather than a replacement.
Technical architecture and offline-first design
Lumina AI is built entirely with offline capabilities in mind, prioritizing privacy and reliability. The core stack includes C++17 for performance-critical components, Qt6 for cross-platform desktop integration, and Make for streamlined compilation. Ollama serves as the local inference engine, enabling the AI agent to process natural language queries without external dependencies.
Key technical features include:
- A native C++ tool-calling system for executing system commands
- Desktop and active-window monitoring to tailor responses to the user’s context
- Lightweight web scraping via Qt’s networking stack for dynamic data retrieval
- Persistent JSON-based memory to retain context across sessions
The developer emphasized maintaining a lightweight footprint, ensuring the agent remains responsive even on modest hardware. By avoiding cloud services, Lumina AI guarantees data privacy and offline functionality, aligning with modern expectations for secure, autonomous tools.
Open source philosophy and community engagement
Lumina AI is released under the GPLv3 license, reflecting the developer’s commitment to open collaboration. The project is hosted on GitHub, where the full commit history from Shijima-Qt is preserved, offering transparency into its evolution. This approach not only honors the original project but also invites developers to explore the architectural decisions behind the transformation.
The developer has explicitly invited feedback from experienced C++ and AI practitioners, seeking critiques on design choices, potential pitfalls, and opportunities for optimization. This openness underscores the project’s educational intent—not just as a functional tool, but as a learning platform for aspiring developers.
Lumina AI stands as a testament to how legacy code can be reimagined with modern technology. By merging nostalgia with practicality, it demonstrates that even simple desktop companions can evolve into powerful, private, and autonomous AI agents. For developers exploring local AI integration or C++ desktop applications, Lumina AI offers a compelling case study in repurposing existing codebases for new challenges.
As the project matures, it could inspire further innovations in offline AI assistants, proving that innovation isn’t always about building from scratch but about revitalizing what already exists.
AI summary
Shijima-Qt’nin yeniden canlandırılmasıyla ortaya çıkan Lumina AI, tamamen yerel ve hafif bir masaüstü AI ajanı. C++ ve Qt6 kullanarak nasıl geliştirileceğini öğrenin.
