Nvidia has released a beta tool that distributes local AI requests across compatible computers on the same network. Personal AI Router, or PAIR, sits between applications and existing runtimes such as Ollama or LM Studio, forwarding each request to an available machine and returning the result to the calling app.
The approach is designed for parallel agent workloads that can overwhelm one laptop or graphics card. In Nvidia’s demonstration, three devices completed a task involving five subagents in just under nine minutes, compared with 18 minutes on one laptop. That example shows the potential benefit but is not a guarantee for models or jobs that cannot run concurrently.
PAIR automatically discovers supported systems and uses mutual TLS encryption for traffic between machines. Hardware coverage includes GeForce RTX cards from the 20 series onward, RTX Pro workstations, DGX Spark and Apple computers beginning with the M4 generation. The beta runs on Windows, macOS and Linux.
Users do not need to rewrite their agents, but every participating device still needs enough memory and compatible software for the assigned model. Network speed, mixed hardware and the number of independent requests will affect performance. PAIR’s immediate value is therefore less like pooling memory into one giant computer and more like adding a local traffic controller that keeps several existing machines busy at once.