NVIDIA has begun beta testing Personal AI Router (PAIR), a new tool that distributes AI inference requests across multiple local machines to boost performance without altering existing setups, promising faster processing and enhanced privacy.
NVIDIA has begun beta testing Personal AI Router, or PAIR, a local network tool that is designed to spread AI inference requests across several compatible machines instead of pushing them through a single GPU. The company says the software is aimed at local multi-agent workloads, where one model call can trigger many more and quickly overwhelm a lone system. According to NVIDIA, PAIR sits between the agent and the inference service, then routes each request to an eligible node on the network.
The appeal is straightforward: more of the available hardware can be kept busy without changing the underlying agent stack. NVIDIA says PAIR works with popular local inference tools such as Ollama and LM Studio, and that an agent still sees a single connection while the router decides where each request should run. The documentation says the router chooses a node based on model compatibility, engine availability and current load.
NVIDIA has also stressed what PAIR does not do. It does not combine separate GPUs into one larger accelerator, nor does it pool video memory into a single shared resource. Instead, it distributes independent inference jobs across multiple systems. The company says the software supports Windows 11, Linux and macOS, including x64 and arm64 machines, and can coordinate mixed operating systems so long as the model or engine is supported on the target node.
In a demonstration cited by InfoQ and NVIDIA, PAIR was shown alongside Hermes Desktop and Ollama, with the workload split across an RTX Spark, a DGX Spark and an RTX 5090. NVIDIA said the setup roughly halved completion time compared with running the same task on a single RTX Spark laptop, although it added that results vary depending on workload, model, settings, hardware, network conditions and node availability. The company says the router keeps prompts, files and agent context on the home network, positioning it as a privacy-oriented option for users who want to scale local AI without sending data elsewhere.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
