System requirements

Native client

Platform	Minimum public package
Windows	Windows 10 or 11, x64
macOS Apple Silicon	macOS 14 or later, arm64
macOS Intel	Intel x64 package
Linux Debian family	Ubuntu or Debian, x64
Linux RPM family	Fedora or RHEL, x64

The client always detects hardware before advertising capacity. A supported operating system does not guarantee that every GPU backend is certified.

Browser worker

The browser worker requires:

• a current browser with Web Workers;
• HTTPS outside local development;
• WebGPU for GPU contribution, or an accepted CPU fallback;
• a visible active page for continued execution;
• wake-lock support when the platform allows it.

iOS and mobile browsers may suspend background work aggressively. Use the native desktop client for persistent availability.

Source development

• Node.js 24 or later.
• Python 3.12 for the physical distributed runtime.
• PowerShell for the repository's Windows automation.
• A backend accepted by the selected runtime, such as Vulkan for the local llama.cpp Qwen demonstration.

Network

Private coordinators should use HTTPS/WSS outside loopback and restrict access to trusted machines. Performance depends on directed latency, bandwidth, stage size, model architecture, and concurrency; a generic internet speed result is not sufficient to predict tokens per second.