Nvidia Launches PAIR Beta to Distribute AI Agents Locally

Key Takeaways

1Nvidia launches PAIR, an open-source AI agent router in beta

2The software distributes AI sub-agents across PCs on the local network based on several criteria

3PAIR does not share GPU or memory and requires a Mac M4 or newer for macOS

4Compatible with Ollama and LM Studio, it operates locally and relies on mDNS and MTLS

Why it matters : PAIR enables faster execution of home AI agents by leveraging the resources of multiple computers while remaining local and secure.

Nvidia releases PAIR as open-source, a "Personal AI Router" that delegates AI sub-agents to other machines on the home network. The software relies on real-time allocation criteria but does not share GPU or memory resources and imposes defined prerequisites, particularly on the Mac side.

No Shared GPUs and One Sub-Agent per Machine

PAIR does not pool hardware resources: neither GPU power nor memory is shared between computers. It therefore does not allow the execution of larger models than those that a single system can support on its own. The software also does not distribute the same sub-agent across multiple machines, with each sub-agent assigned to a single system. Optimal usage may involve keeping many computers on the network powered on, and the behavior when using a machine simultaneously for other tasks, such as video or gaming, is not specified.

Orchestration: The Agent Decides, PAIR Routes and Collects

Once the systems are connected, a main agent, such as Hermes Desktop or OpenClaw, orchestrates the decomposition into sub-agents and delegates their execution to PAIR, which is used as a proxy. PAIR then distributes the sub-agents across the other computers, collects the results, and sends them back to the main agent. A dashboard provides real-time status of these operations.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Task Assignment Based on Resources and Compatibility

PAIR assigns sub-agents taking into account several factors: each machine's acceptance of assignments, the presence of the required inference engine and model, current workload, and available GPU bandwidth. Nvidia indicates that it is working to expand and refine these allocation criteria.

Platforms, Prerequisites, and Open-Source Availability

PAIR works with Windows, Mac, and Linux. For macOS, an Apple M4 generation processor or later is required, while Nvidia does not specify any specific requirements for Windows or Linux. The software is compatible with the inference engines Ollama and LM Studio. Designed as a "Personal AI Router," PAIR accelerates the execution of AI agent tasks by distributing sub-agents across the local network when tasks can be parallelized, while maintaining local processing. It is open-source, available in beta on Github, and relies on mDNS for discovery and MTLS for security.