OpenMandriva Linux User Deploys Local LLMs to Build Missing Graphical Updater and Package Manager

Two years ago, IT professional and long-time Linux enthusiast Richard Sezov made the switch to OpenMandriva Linux. The transition was driven by a need to balance a rolling release model with his company’s requirement for RPM-based security software. Having previously enjoyed a seamless experience as an Arch Linux user, Sezov sought to avoid the traditional upgrade and reinstallation cycles associated with non-rolling distributions. Leveraging his professional seniority, he successfully standardized his workflow around OpenMandriva.

However, adopting a new distribution for personal and family use introduced unexpected challenges. Sezov adheres to a philosophy of consistency across all household machines to streamline technical support and maintenance. While his own familiarity with the command line interface makes managing raw system updates straightforward, other members of his household require a more accessible user experience. Sezov needed a graphical updater and installer simple enough for non-technical family members to operate independently, a feature set that OpenMandriva lacked out of the box.

The Grass Isn’t Always Greener: Recognizing Papercuts in Volunteer-Run Distros

Sezov’s history with the ecosystem began with the original Mandriva Linux, a distribution that predated Ubuntu’s mainstream popularity and served as a pioneer for user-friendly Linux adoption. Mandriva was widely celebrated for offering intuitive graphical updates and package management tools that predated and inspired software like Synaptic. Expecting a similar level of mainstream polish from OpenMandriva, its spiritual successor, Sezov quickly discovered the realities of a platform driven entirely by volunteer contributions.

The default update mechanism provided by OpenMandriva relied on a toolbar shortcut executing a shell script via the command line, completely omitting visual notifications for pending updates and requiring manual intervention. Community forums overwhelmingly pointed users toward terminal commands for routine maintenance—an approach entirely unsuitable for users who rely on graphical interfaces, such as Sezov’s mother, who depends on a streamlined graphical environment to keep her systems current.

Further exploration of the system menus uncovered DnfDrake, a legacy graphical package manager included with the distribution. While the utility performed basic package management functions, it suffered from severe usability hurdles. The application window lacked resizability, rendering interface buttons entirely inaccessible on budget laptop screens with a 1366×768 resolution. Additionally, the software proved sluggish, frequently reloading entire package lists between basic operations. Because the tool was not enabled by default and received little community traction among new users, it became evident that the distribution required a modern alternative.

Leveraging Local LLMs to Bridge Technical Gaps

Motivated to contribute to a smaller, volunteer-run open-source community, Sezov recognized an opportunity to build the missing tools himself. However, his professional expertise lay primarily in back-end development rather than desktop graphical user interface design. Writing a cross-platform desktop application from scratch manually would have demanded a significant investment of time, threatening to stall the project before it even began.

Fortunately, Sezov had been experimenting with running local Large Language Models on his Framework 13 laptop, which features an AMD Ryzen AI 9 HX 370 processor, a Radeon 890M GPU, and 64GB of RAM partitioned evenly between system memory and video memory. Utilizing tools like Ollama and OpenCode, he had been successfully debugging cross-platform code for personal projects, realizing that local AI infrastructure could bridge the gap in his desktop development knowledge.

Running local LLMs efficiently, however, presented its own set of technical hurdles. Default configurations for Ollama failed to utilize hardware acceleration, resulting in sluggish performance that ignored the integrated graphics capabilities. To resolve this, Sezov had to manually configure the Ollama system service, embedding specific environment variables to leverage Vulkan support, integrated GPU processing, and optimized context length caching.

Selecting the right model proved equally critical. While dense models offered deep data comprehension, their execution speeds were impractical for rapid development workflows. Sezov ultimately utilized mixture-of-experts models, which selectively activate parameters per token, achieving token generation speeds comparable to paid cloud-based AI services while maintaining the technical depth required for software engineering.

Designing and Developing Native Applications for KDE Plasma

With his local AI infrastructure optimized, Sezov utilized the plan mode in OpenCode to conceptualize the software architecture alongside an AI model. Because OpenMandriva is fundamentally built around the KDE Plasma desktop environment, the development team selected Python paired with PyQt 6 to ensure a native look and feel that integrated seamlessly with the system’s aesthetic.

The project encompassed two distinct applications: a system tray utility designed to monitor and execute system updates, and a modern graphical package manager intended to replace DnfDrake. To maintain system security without compromising usability, both applications were engineered to launch with standard user permissions, dynamically re-launching as privileged worker processes only when system modifications or installations were strictly required.

The development process involved iterative bug testing and code refinement. While initial iterations required manual oversight—particularly when handling complex tasks like streaming terminal command outputs directly into a graphical interface—the integration of newer, more powerful open-source models accelerated development. The resulting package manager was designed to manage multiple package formats supported by OpenMandriva, including RPMs, Flatpaks, and Snaps, with automatic detection for snapd availability.

Continuous Improvement and Community Integration

After packaging the newly developed applications and releasing them into the OpenMandriva ecosystem, Sezov received early feedback from the community regarding edge-case bugs that had not appeared during initial local testing. By utilizing his local LLM setup, he was able to rapidly diagnose these error reports, craft targeted patches, and push updated software releases to the distribution repositories.

Whether these custom graphical tools ultimately become the official standard for OpenMandriva remains to be seen, but the initial releases have proven stable and user-friendly enough for everyday family deployment. More importantly, the project demonstrated that modern local artificial intelligence infrastructure has matured to a point where a developer without prior desktop GUI experience can conceptualize, build, and deploy functional system applications. By handling complex development workloads locally, users can successfully navigate software gaps in open-source ecosystems while retaining absolute control over their computing environments.