Kora Systems has unveiled Kora, a new Linux-based desktop operating system designed around local AI processing and a dedicated gaming experience. Announced on November 10, 2026, the operating system incorporates GameFrame, a gaming interface derived from Playtron’s GameOS, alongside an AI engine capable of distributing workloads across a PC’s CPU, GPU and neural processing unit (NPU). Kora is scheduled to arrive on consumer PCs later in 2027, with Qualcomm Snapdragon X2 Elite hardware receiving the first confirmed NPU support.
The project is led by Kirt McMaster, co-founder of Cyanogen and Playtron. Previously known as Playtron, the company developed GameOS as a Linux-based operating system focused on gaming devices. Its transition to Kora Systems expands that technology into a general-purpose desktop platform combining gaming, productivity and AI capabilities.
Kora introduces three operating modes: Work, Create and Play. The Play mode uses GameFrame to provide a console-style interface for launching and playing PC games, retaining part of the company’s original gaming focus.
McMaster also outlined plans for AI-powered game development tools that would allow users to create, publish and play games directly within the operating system. Android application support is on the roadmap, although the company has not announced when it will become available.
The company has not yet provided a detailed game compatibility list, gaming performance benchmarks or confirmation of which Windows games will run through its planned compatibility layer. GameFrame’s actual compatibility and performance will therefore need to be assessed when Kora becomes available.
Beyond gaming, Kora’s central technology is Cortex, a proprietary AI engine designed to coordinate local models and distribute inference workloads between available processors. Its underlying Lemma inference system can divide model layers across computing devices, using technologies including llama.cpp, whisper.cpp, stable-diffusion.cpp and ONNX.
Cortex can also use computing resources from other machines connected to the same local network. The approach is intended to make better use of available hardware rather than depending exclusively on a single processor or cloud-based AI services.
Qualcomm has worked with Kora Systems for nearly a year to optimize the operating system for Snapdragon-powered computers. The companies have demonstrated support for Snapdragon X2 Elite processors, combining the Oryon CPU, Adreno GPU and Hexagon NPU for AI workloads.
The Hexagon NPU, rated by Qualcomm at up to 80 TOPS, is currently the only NPU platform Kora has explicitly confirmed as supported. McMaster told Tom’s Hardware that Kora uses the llama.cpp Hexagon backend and the upstream FastRPC driver to access the processor.
McMaster also supplied performance figures for Google’s Gemma 4 26B A4B language model running entirely on the Snapdragon X2 Elite Hexagon NPU. According to the developer, the system achieves approximately 25 tokens per second during decoding and 550 tokens per second during prompt prefill with a 32,000-token context window.
These are developer-reported results rather than independently verified benchmarks. The company has not disclosed the precise Snapdragon X2 Elite configuration or installed RAM used for the measurements, making direct comparisons with other PCs difficult.
Memory capacity will be another consideration for users planning to run large AI models locally. McMaster recommends 64GB of RAM to provide access to more capable models, although this is not a minimum system requirement.
The Gemma 4 26B A4B model itself requires substantial memory. Google’s estimate places its quantized weights at approximately 14.4GB using Q4_0 quantization, before accounting for the context window and other runtime requirements. The Q4_K_M variant used by Kora occupies approximately 16GB to 17GB.
Kora is also being developed for conventional Intel and AMD PCs, but their NPU support remains less mature than Qualcomm’s implementation.
Intel NPU integration is planned, with discussions still underway. For AMD Ryzen AI processors, McMaster indicated that the operating system will likely use the NPU for speech recognition and embedding models through ONNX rather than running large language models directly on the NPU.
These differences highlight the challenge of supporting AI acceleration across multiple hardware vendors, each using different drivers, software frameworks and model execution methods.
Interest in alternative NPU workloads has also extended beyond language models. An experimental Windows 11 desktop rendering driver recently demonstrated an Intel NPU handling graphical output, illustrating another potential use for processors traditionally reserved for AI acceleration.
Kora’s other desktop features include neural workspaces, which separate projects and manage application and AI-agent access, and a neural file system that supports searching files by their meaning rather than relying entirely on filenames or folder locations.
Although Kora uses Linux and other open-source technologies, its Cortex engine and desktop interface remain proprietary. The company plans to offer the operating system as a free download, with paid subscriptions providing access to additional AI models and agent-based services. Subscription prices have not been announced.
Kora Systems is also developing HUMAIN OS, an enterprise-oriented operating system created in partnership with Saudi Arabian AI company HUMAIN. That platform is expected to become commercially available on Snapdragon X2 Elite PCs in 2027, ahead of Kora’s planned consumer release later that year.
For PC gaming enthusiasts, GameFrame is the most directly relevant feature, potentially providing a Linux gaming interface alongside conventional desktop applications and hardware-accelerated local AI. However, Kora remains an announced platform rather than a publicly available operating system, and its gaming compatibility, system requirements and performance across Intel, AMD and Qualcomm hardware have yet to be established through independent testing.
