- Kirt McMaster, co-founder of Cyanogen, has announced Kora, a new Linux-based desktop operating system.
- Kora is designed for AI developers, content creators, and power users, integrating AI models and autonomous agents at a system level.
- Scheduled for a 2027 launch, the OS will optimize AI workloads across CPU, GPU, and NPU resources, with initial support for Qualcomm Snapdragon X2 Elite.
Kora: The Operating System for the AI Era
Kora Systems, a startup led by Cyanogen co-founder Kirt McMaster, has unveiled Kora, a new Linux-based desktop operating system. Positioned as a free-to-use alternative to Windows, Kora is specifically designed for users heavily reliant on AI models and agents. The system is slated for release in 2027.
System-Level AI Integration and Resource Optimization
Kora is engineered to unify various AI models and autonomous agents at a system level, allowing workloads to be split across CPU, GPU, NPU resources, and other machines on the same local network. At its core, Kora features Cortex, a proprietary hybrid inference engine optimized for local models. The OS runs local models by default and leverages cloud models when necessary.
Built “from the kernel to the shell,” Kora incorporates “neural workspaces” and a “neural file system,” which refer to sandboxed environments and files indexed by semantic meaning. The operating system offers three modes: “Work,” “Create,” and “Play,” with the latter including GameFrame, a descendant of Playtron’s GameOS.
Hardware Support and AI Model Capabilities
Initially, Kora will support the Hexagon NPU found in Qualcomm Snapdragon X2 Elite processors, which boasts up to 80 TOPS of performance. McMaster stated that Cortex can schedule AI workloads across the CPU, GPU, and NPU on the X2 Elite. While Intel and AMD NPU support is in development, AMD’s Ryzen AI NPUs are expected to be limited to speech and embedding models, rather than large language models (LLMs).
Kora utilizes the llama.cpp Hexagon backend and the FastRPC driver to access the Qualcomm NPU. The primary model format supported is GGUF, llama.cpp’s model file format. Lemma, the inference engine within Cortex, is designed to split model layers across processors as needed.
| NPU | Vendor’s own framework | Status in Kora | Runtime Kora uses | What it runs |
|---|---|---|---|---|
| Qualcomm Hexagon (Snapdragon X2 Elite) | QAIRT/QNN | Supported today | llama.cpp’s Hexagon backend | Language models; KORA cites Gemma 4 26B at 25 tokens per second |
| AMD XDNA 2 (Ryzen AI) | Ryzen AI software | Planned for launch | ONNX | Likely embedding and speech-recognition models; language models unlikely |
| Intel NPUs | OpenVINO | Planned, “conversations underway” | Not stated | Not stated |
McMaster indicated that for models like Gemma 4 26B, which requires approximately 14.4GB of memory for 4-bit quantized weights, 64GB of RAM is recommended for access to better models. The company projects that about 70% of model calls will run locally by 2027.
Target Audience and Business Model
Kora is not aimed at the mass market but rather at experienced AI users, developers, and content creators. An enterprise variant, Humain OS, developed with Saudi AI company Humain, will be commercially available in 2027 on OEM PCs utilizing the Snapdragon X2 Elite. Kora itself will be downloadable from the company’s website “later in 2027,” with recommended configurations provided, though it should be installable on most modern PCs. Specific pricing and the overall business model will be disclosed closer to the availability date.
The emergence of specialized operating systems like Kora underscores the growing demand for optimized AI solutions. This trend mirrors broader advancements in AI applications, such as the expansion of AI capabilities in productivity tools, as recently covered in our article Anthropic Expands Claude with New Office Productivity Tools, Challenging Google and Microsoft.
The introduction of Kora represents a significant step towards operating systems designed from the ground up for AI. For developers and content creators, it could unlock new possibilities for running complex AI models locally with highly efficient hardware utilization. For the industry, this signifies increased competition in the desktop OS segment, traditionally dominated by Windows and macOS, and a potential acceleration of innovation in edge computing and on-device AI capabilities.
