NVIDIA has released "PAIR (Personal AI Router)," a virtual inference router that can detect and connect compatible PCs on a network to execute inference tasks in coordination.
When running AI in a local environment using tools such as Ollama or LM Studio, PAIR detects other PCs within the network, assigns jobs, and returns the responses. It works by receiving requests via a proxy and selecting the appropriate node based on the model's requirements to perform the processing. Because PAIR handles the routing, the AI agent side only needs to send requests without needing to be aware of specific nodes.
The PAIR proxy is compatible with Ollama and LM Studio and does not require a new API. Additionally, it allows for control such as powering off unnecessary nodes when there are no active jobs.
According to a demo released by NVIDIA, in a multi-agent workload using Qwen 3.6-35B-A3B, clustering three devices significantly reduced processing time compared to running the workload on a single PC.
Supported operating systems include Windows 11, DGX OS, Ubuntu 14.04, and macOS Tahoe. GPUs must meet specific requirements, such as GeForce RTX (20 series or later) or Mac M4 and later.
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.net (Google News: LM Studio, 2026-09-04)