A team led by Cyanogen co-founder Kirt McMaster has introduced Kora, a new Linux desktop operating system designed to handle artificial intelligence tasks locally. The operating system, developed by Kora Systems Inc., uses a proprietary technology called Cortex to distribute AI model processing across a computer’s central processing unit, graphics processing unit, and neural processing unit.
Kora Systems, formerly known as Playtron, describes the product as “The Operating System for Intelligence.” The company states that the OS is built from the kernel to the shell to prioritize local model execution, switching to cloud-based models only when necessary. The desktop environment features “neural workspaces” and a “neural file system,” which organize files based on semantic meaning rather than traditional folder structures.
Snapdragon X2 Elite and NPU Support
Initial hardware support for Kora focuses on the Qualcomm Snapdragon X2 Elite platform. The system utilizes the Hexagon NPU found in this chip, which Qualcomm rates at 80 trillion operations per second. According to McMaster, the Cortex engine can schedule AI workloads not only across the local CPU, GPU, and NPU but also across other machines on the same local network.
McMaster provided specific performance metrics for the current setup. He stated that the Snapdragon X2 Elite NPU runs Google’s Gemma 4 26B model at 25 tokens per second for decoding. The system achieves this by using the llama.cpp Hexagon backend and the upstream FastRPC driver in the mainline kernel, rather than Qualcomm’s proprietary QAIRT or QNN runtimes.
The company recommends 64GB of system memory to access larger models, though this is not a strict requirement. The Gemma 4 26B model requires approximately 14.4GB to 17GB of memory depending on quantization settings. McMaster noted that the X2 Elite reaches 550 tokens per second for prefill throughput while running entirely on the Hexagon NPU with a 32K context window.
Intel and AMD Compatibility Plans
Support for other hardware vendors is in development, though with varying degrees of certainty. McMaster confirmed that Intel NPU support is planned and that conversations are underway. However, he indicated that AMD’s Ryzen AI chips will likely face limitations. According to the developer, AMD NPUs will probably be restricted to speech recognition and embedding models, rather than large language models.
This distinction arises because each NPU vendor prefers its own software stack. Qualcomm uses QAIRT/QNN, Intel relies on OpenVINO, and AMD utilizes Ryzen AI software. Kora predominantly supports the GGUF model format, which is associated with the llama.cpp project. While llama.cpp has an OpenVINO backend for Intel, it lacks a dedicated AMD NPU backend, which explains Kora’s decision to route AMD tasks through ONNX.
Kora will launch in two variants. The enterprise-focused Humain OS, developed in partnership with Saudi AI company Humain, is scheduled for commercial availability in 2027 on OEM PCs using Snapdragon X2 Elite. The consumer desktop version, Kora, will follow later in the same year. McMaster said the OS will be downloadable from the company website, with a list of recommended configurations provided to users.
Source: Tom’s Hardware
