Tom’s Hardware reported that Kirt McMaster, co-founder of Cyanogen, unveiled Kora, a new Linux desktop operating system designed to distribute artificial intelligence workloads across multiple hardware components.reported The system aims to utilize CPU, GPU, and NPU resources simultaneously, as well as leverage other machines connected to the same local network for processing tasks.The new Linux operating system uses all parts of your computer, plus nearby devices, to run AI programs locally.The system implements a proprietary scheduler that distributes tensor operations across heterogeneous accelerators and peer nodes.
Evidence: reported
Cyanogen co-founder unveils Kora, a Linux OS that splits AI models across the CPU, GPU, and NPUaifleet-tomshardware
The figures
Tom’s Hardware cited Qualcomm’s specification of 80 TOPS for the Hexagon NPU found in the Snapdragon X2 Elite chip.reported McMaster told the outlet that this specific NPU configuration runs Google’s Gemma 4 26B model at a decode speed of 25 tokens per second.reported The company expects the OS to arrive on personal computers in the year 2027.
Evidence: reported
Cyanogen co-founder unveils Kora, a Linux OS that splits AI models across the CPU, GPU, and NPUaifleet-tomshardware
Evidence: reported
Cyanogen co-founder unveils Kora, a Linux OS that splits AI models across the CPU, GPU, and NPUaifleet-tomshardware
Behind the figures
The performance metrics provided by McMaster relate specifically to the Snapdragon X2 Elite hardware and its Hexagon NPU.reported Tom’s Hardware noted that the press kit did not provide speed figures, model formats, or quantization details for Intel or AMD platforms. The outlet also referenced its own earlier Windows testing, which showed Intel’s NPU work often running through IPEX on the GPU because it was faster than using the NPU directly.
Evidence: reported
Cyanogen co-founder unveils Kora, a Linux OS that splits AI models across the CPU, GPU, and NPUaifleet-tomshardware
What it means for your stack
The OS includes a proprietary technology called Cortex, described as a hybrid inference engine optimized for local models. This engine is not open source. Kora runs local models by default and switches to cloud models only when necessary. The system features "neural workspaces" and a "neural file system," which refer to sandboxed environments and files indexed by semantic meaning rather than traditional directory structures.
Building on this
Developers can expect three distinct operating modes: Work, Create, and Play. The Play mode utilizes GameFrame, which descends from Playtron’s previous GameOS. Humain OS, developed in partnership with Saudi AI company Humain, serves as the enterprise variant of this operating system. Currently, Intel and AMD NPU support is in development, though McMaster indicated AMD’s NPU may be limited to speech and embedding models rather than large language models.
Why it matters
- The proprietary nature of the Cortex engine means developers cannot inspect or modify the scheduling logic for cross-hardware inference.
- Lack of documented performance metrics for Intel and AMD NPUs creates uncertainty for developers targeting those specific silicon vendors.
- The reliance on semantic indexing for the file system suggests a shift in how applications might retrieve and manage local data assets.
- The plan to utilize local network machines for distributed inference introduces new complexity in latency and security management for local clusters.
Topics: linux, npu, inference, local-ai, kora-os
Updated
Then / now / next
Playtron focused on gaming operating systems with GameOS.
The company, now Kora Systems Inc., is building a general-purpose Linux OS with integrated AI agents.
Developers should watch for announcements regarding Intel and AMD NPU runtime support and performance benchmarks.
Story timeline
- Then Playtron focused on gaming operating systems with GameOS.
- 11 Nov 2026 Cyanogen co-founder unveils Kora, a Linux OS that splits AI models across the CPU, GPU, and NPU
- Now The company, now Kora Systems Inc., is building a general-purpose Linux OS with integrated AI agents.
- Next Developers should watch for announcements regarding Intel and AMD NPU runtime support and performance benchmarks.
Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Terms in this story
Cortex
A proprietary hybrid inference engine that schedules AI workloads across CPU, GPU, NPU, and networked devices.
Neural workspaces
Sandboxed environments within the OS designed for specific AI-driven tasks.
Neural file system
A file system that indexes files by semantic meaning rather than traditional directory paths.
TOPS
Trillions of Operations Per Second, a metric used to measure computing performance, particularly for NPUs.
Questions readers ask
Will Kora OS support Intel and AMD NPUs?
Support is in development, but AMD NPUs may be limited to speech and embedding models rather than large language models.
Is the Cortex inference engine open source?
No, Tom’s Hardware reported that the Cortex hybrid inference engine is not open source.
When will Kora OS be available for PCs?
The operating system is expected to arrive on personal computers in 2027.
Connected stories
Reading receipt
- Kora OS targets a 2027 release for PCs, including Snapdragon X2 Elite devices.
- The proprietary Cortex engine schedules tasks across CPU, GPU, NPU, and local network machines.
- Qualcomm Hexagon NPU runs Gemma 4 26B at 25 tokens per second for decode on X2 Elite.
Still open: How will the Cortex engine handle latency when distributing workloads to other machines on the local network?
Still open: What specific model formats and quantization methods will be supported for Intel and AMD NPUs?