NVIDIA has released "PAIR (Personal AI Router)," a free tool designed to coordinate available computing resources across a network.
PAIR detects PCs within a network that meet specific requirements and assigns inference tasks to the appropriate nodes. The system works by having a proxy receive a request, identify the engine and model requirements, select a node, and return the execution results to the source.
This tool is compatible with Ollama and LM Studio and does not require a new API. Additionally, it includes a feature to power off or put unnecessary nodes into sleep mode when no jobs are assigned.
In a demonstration involving a workload that executed five sub-agents, NVIDIA reported that clustering multiple devices with PAIR significantly reduced processing time compared to using a single PC.
Supported operating environments include Windows 11, DGX OS, Ubuntu 14.04, and macOS Tahoe. Supported GPUs include GeForce RTX (20 series and later), DGX Spark, and Mac M4 or later, with a minimum requirement of 8GB of memory.
Source: NVIDIAが重たいAI処理を同一ネットワーク内のPCに割り振れる無料ツール「PAIR」を公開 (GIGAZINE, 2026-09-04)