DEV Community

News
News

Posted on

Kora: an AI-first Linux desktop OS that spreads model work across CPU, GPU and NPU

Infographic cover

πŸ“… November 11, 2026 β€’ Tom's Hardware

Kirt McMaster, co-founder of Cyanogen, has unveiled Kora, a Linux desktop operating system with AI models and agents built in, made by Kora Systems Inc., formerly Playtron, under the tagline "The Operating System for Intelligence". The system runs local models by default and turns to cloud models when needed. PCs with Kora are expected in 2027, and the system is downloadable for most modern PCs. The announcement was covered by Tom's Hardware in an interview with McMaster, with further reporting from heise online and It's FOSS.

The core of the system is Cortex, a hybrid inference engine that splits AI workloads across the CPU, GPU and NPU, and across other machines on the same local network. Cortex is not open source and is optimized for local models. Inside it sits Lemma, an inference engine built on llama.cpp, whisper.cpp, stable-diffusion.cpp and ONNX, with model support predominantly in the GGUF format. On the hardware side, McMaster said the Snapdragon X2 Elite Hexagon NPU runs Google's Gemma 4 26B model at 25 tokens per second for decode.
Qualcomm support exists today, AMD XDNA 2 support is planned and likely limited to speech and embedding models, and Intel NPUs are planned with talks underway.

The desktop itself is built around AI. Kora ships neural workspaces and a neural file system, meaning sandboxed workspaces and files indexed by semantic meaning, and offers three modes: Work, Create and Play. The GameFrame gaming stack descends from Playtron GameOS. A separate enterprise variant, Humain OS, developed with the Saudi AI company Humain, is set for commercial release in 2027 on OEM PCs with the Snapdragon X2 Elite, and Kora itself follows later in 2027.

The architectural point matters for how AI is deployed on desktops. On Windows and macOS, AI mostly lives in a browser tab or inside an application sandbox, with data sent to the cloud. Kora builds the operating system from kernel to shell around local models, so AI behaves like a system resource on the level of memory or storage. Sandboxing is enforced per workspace, which addresses the security question local-model setups usually leave open, and pooling machines on the same network into one inference engine turns idle PCs into usable compute without new hardware.

If the shipping product matches the announced design, the line between an OS with an AI assistant and an OS built for AI becomes concrete: model placement across CPU, GPU, NPU and the local network is handled by the system itself rather than by each application, starting with Humain OS on OEM hardware in 2027 and Kora later that year.

Sources: tomshardware.com/software/linux/cyanogen-co-founder-unveils-kora-a-linux-os-that-splits-ai-models-across-the-cpu-gpu-and-npu-snapdragon-x2-elite-and-hexagon-npu-get-full-support-first - heise.de/en/news/Kora-OS-Ex-Cyanogen-CEO-unveils-AI-focused-Linux-desktop-OS-11519373.html - itsfoss.com/news/kora-ai-desktop-os-unveiled

Key figures

IMPACT checklist


Continue the conversation: https://t.me/AIAutomationArchitecture Β· Bluesky

AI Automation

Top comments (0)