Nvidia introduced its Personal AI Router (PAIR) at IFA 2026, a software solution that enables users to create a personal AI cluster from multiple devices on a local network. This free, open-source tool routes independent AI inference requests to compatible systems with available capacity, optimizing the execution of agentic AI tasks. The company states that over half of U.S. households possess two or more PCs, with much of their computing power remaining idle daily. PAIR aims to utilize this untapped resource, allowing for more parallel task execution and freeing up primary systems for other activities like gaming or creative work.
PAIR operates by discovering compatible PCs on a local network and then intelligently dispatching sub-tasks from a central AI agent to systems with spare GPU cycles. This approach can prevent performance bottlenecks that occur when multiple sub-agents compete for resources on a single GPU. The software is designed to work with existing local AI inference engines such as Ollama and LM Studio. When a user runs a local AI agent and assigns it a goal, the agent may generate several sub-tasks. PAIR then distributes these sub-tasks among the connected devices, which can lead to faster overall completion of the larger agentic task.
The PAIR beta is available for Windows, macOS, and Linux platforms. It supports Nvidia GeForce RTX 20-series GPUs and newer, Nvidia RTX Pro workstation GPUs based on Turing architecture and newer, Nvidia DGX Spark systems, and Apple M4 or newer silicon. Nvidia reports that PAIR can improve throughput, citing an example of up to 1.9 times faster llama.cpp throughput on an RTX 5090 when utilizing kernel optimizations. The tool is designed to be elastic, adapting as devices join or leave the local network and assigning work without requiring a fixed cluster configuration.
PAIR functions as a local traffic director for separable AI workloads. It evaluates factors such as whether a paired node is online, if a supported engine is enabled, if the exact model is present, the existing workload on the node and engine, and current GPU utilization. This allows PAIR to bypass machines actively in use, such as those running games, and direct inference calls to available hardware. The software uses mDNS for device discovery on the local network and MTLS for secure communication between devices. Users maintain control over which devices are connected to the router.
The introduction of PAIR aligns with Nvidia's broader strategy to expand personal PCs into AI agent execution hubs and to promote local AI processing. The company also announced the upcoming launch of RTX Spark Windows PCs in October, which will feature Blackwell-based RTX GPUs and Grace CPUs. These new systems are intended to further support the operation of local AI agents. PAIR aims to simplify the process of running AI models locally by reducing the complex setup previously required for model and inference program configuration.