Prague, Czechia, hosted Open Source Summit Europe 2026 (OSS EU 2026) from October 7 to 9. A highlight of the summit was the main-stage keynote delivered by Min Liu, Executive Director of openKylin, on October 8. The presentation, titled “From AI Runtime to System Agents: Reimagining Linux for the AI Era with openKylin,” introduced KART, a unified AI runtime, and an innovative system-level agent execution environment. Complementing this, Jianfeng Li, a Technical Committee member, provided a lightning talk on developing an AI-ready Linux desktop for RISC-V.
openKylin, an open-source operating system community originating from China, is dedicated to crafting an AI-integrated OS accessible to users globally. The project utilized the summit as a platform to exhibit its progress in blending artificial intelligence with operating system functionalities through talks, demonstrations, and engaging discussions with developers.
The Imperative for Operating System Evolution in the AI Age
The widespread adoption of open frameworks and freely available open-weight models is cultivating a strong foundation for advancements in AI. Tools like PyTorch, vLLM, and llama.cpp offer adaptable engineering components, while open-weight models, such as those from DeepSeek, simplify model customization. Furthermore, open-source communities consistently expand collaborative opportunities.
openKylin posits that the primary hurdle now extends beyond mere model performance to the essential system infrastructure required to seamlessly integrate AI with daily computing. Artificial intelligence capabilities often remain separate from typical desktop operations, AI stacks are fragmented, and the convergence of hardware and software continues to be intricate. The community views this as a universal challenge for all operating systems navigating the transition into the AI era, rather than an issue confined to a specific group or enterprise.
openKylin’s Architectural Blueprint: From Resource to Intent
To bridge this identified gap, openKylin advocates for a fundamental shift in the operating system's management philosophy—moving from conventional “resource management” to “intent management.” While traditional operating systems primarily manage components such as CPUs, memory, and files, AI-era operating systems must comprehend user objectives and translate them into actionable system processes. These processes are then coordinated and monitored, establishing a continuous cycle of understanding, planning, execution, and observation.
This forward-looking architecture is structured across three distinct layers:
- Runtime Layer: Features KART, the unified AI runtime, and the kytensor local inference engine, which together provide shared inference services.
- Agent Execution Layer: An environment specifically designed to support KylinBot, Kylin-CUA, and various third-party agents.
- Capability Layer: Offers API tools and desktop GUI skills, making existing Linux functionalities readily available, all while being governed by permission boundaries and execution monitoring.
openKylin 3.0: Empowering the Desktop with AI
The practical application of these capabilities is evident in three core desktop features introduced in openKylin 3.0:
- AI Assistant: Transforms user interaction from locating features to articulating intentions.
- Global Intelligent Search: Elevates search functionality from location-based queries to semantic discovery.
- Intelligent Space: Organizes information intuitively around specific tasks, moving beyond traditional folder structures.
Extending AI Capabilities to RISC-V and Open Hardware
openKylin is also actively engaged in expanding its AI stack to new frontiers, including RISC-V and other open hardware ecosystems. Key focus areas include hardware enablement, optimizing for RISC-V Vector (RVV) extensions, AI runtimes, and heterogeneous computing.
Noteworthy advancements include:
- Inference Acceleration: RVV kernel optimizations have demonstrated a 2–5x speed improvement for FP32 inference using ONNX Runtime.
- Large-Model Compatibility: The AI stack now supports CPU inference for large models with vLLM on the SpacemiT K3 platform.
- Diverse Distribution Methods: Software can be delivered via .deb packages, Python wheels, Conda, and Docker.
The community aims to integrate both RVV kernel optimizations and its inference framework enablement work into upstream projects through open contributions.
Community Engagement and Demonstrations
During the summit, the openKylin booth showcased AI PCs and various devices running openKylin 3.0, attracting numerous attendees who interacted with the systems and conversed with the technical team. Demonstrations included an interactive VR experience from Haihe Laboratory, a SpacemiT K3 laptop pre-installed with openKylin 3.0, and Singapore-based Nanyang Singtech’s SingNova series of AI edge computing solutions.
About openKylin
openKylin is a China-based open-source operating system community committed to developing an AI-integrated OS for users globally. The community actively contributes to both the Linux and RISC-V ecosystems, having contributed over a million lines of core code. Its UKUI desktop environment is integrated into the official repositories of 12 Linux distributions, including prominent ones like Debian, Ubuntu, and Fedora. As of October 2026, openKylin reports approximately 27 million deployments, 2.9 million active users, and a global contributor base exceeding 29,000.
The community has also cultivated an international network of user groups spanning more than 30 countries. Looking ahead, openKylin plans continued investment in three critical areas: strengthening the runtime foundation for easier AI development, enhancing security and control to ensure observable actions, and fostering a vibrant open ecosystem for improved collaboration among hardware, models, and applications.
This article is a rewritten summary based on publicly available reporting. For the original story, visit the source.
Source: MarkTechPost
