openKylin Unveils KART AI Runtime and Agent Execution Environment at Open Source Summit Europe 2026

The keynote was titled “From AI Runtime to System Agents: Reimagining Linux for the AI Era with openKylin.” Jianfeng Li, a member of the community’s Technical Committee, also delivered a lightning talk titled “Building an AI-Ready Linux Desktop on RISC-V.” Through presentations, demonstrations, and developer exchanges, openKylin showcased its latest work on AI and operating system integration.

openKylin says the rapid adoption of open frameworks and open-weight models is creating a stronger foundation for AI innovation. PyTorch, vLLM, and llama.cpp provide reusable engineering components; open-weight models such as those from DeepSeek lower barriers to model adaptation; and open-source communities continue to expand opportunities for upstream collaboration. openKylin believes the key challenge is no longer model capability alone, but the system infrastructure needed to connect AI with everyday computing. Model capabilities remain disconnected from desktop tasks, AI stacks are fragmented, and hardware–software integration remains highly complex. openKylin views this as a shared challenge for operating systems transitioning into the AI era, rather than one specific to any single community or company.

To bridge this gap, openKylin proposes a shift in the operating system management paradigm from resource management to intent management. Traditional operating systems focus on managing CPUs, memory, and files. Operating systems in the AI era, by contrast, need to understand user goals and break them down into system actions that can be coordinated and observed, forming a closed loop of understanding, planning, execution, and observation. The architecture comprises three layers. The runtime layer includes KART, the unified AI runtime, and the kytensor local inference engine, which provide shared inference services. The agent execution layer supports KylinBot, Kylin-CUA, and third-party agents. The capability layer includes API tools and desktop GUI skills that make existing Linux capabilities accessible, with permission boundaries and execution monitoring governing their use.

In openKylin 3.0, these capabilities are reflected in three core desktop features. AI Assistant shifts interaction from finding features to expressing intent. Global Intelligent Search moves search from location-based lookup to semantic discovery. Intelligent Space organizes information around tasks rather than folders.

openKylin is also expanding its AI stack to RISC-V and open hardware ecosystems. Its focus areas include hardware enablement, RISC-V Vector extension optimization, AI runtimes, and heterogeneous computing. For inference acceleration, RVV kernel optimizations deliver a 2–5× speedup in FP32 inference with ONNX Runtime. For large-model support, the AI stack supports CPU inference with vLLM on SpacemiT K3. Delivery options include .deb packages, Python wheels, Conda, and Docker. openKylin says this work is being advanced through open contributions, with plans to contribute both RVV kernel optimizations and inference framework enablement work upstream.

During the summit, AI PCs and other devices running openKylin 3.0 were shown at the openKylin booth, where attendees tried the systems and spoke with the technical team. Haihe Laboratory presented an interactive VR experience. SpacemiT exhibited a K3 laptop with openKylin 3.0 preinstalled. Singapore-based Nanyang Singtech showcased its SingNova series of AI edge computing products.