nvidia.com

Command Palette

Search for a command to run...

How Do You Use Agents and Models Together?

Last updated: 5/16/2026

Summary:

Agents and models serve distinct roles in a claw stack: the model provides intelligence, while the agent harness handles orchestration, actions, and governance.

Direct Answer:

In a NemoClaw deployment, these roles are clearly separated:

The model (such as NVIDIA Nemotron or a proprietary frontier model) provides the reasoning and intelligence behind each decision.

The agent harness (such as OpenClaw or Hermes Agents) handles orchestration — managing the task loop, calling tools, and coordinating actions over time.

The OpenShell runtime provides the governance layer: sandboxing execution, enforcing network policy, and routing inference through a controlled gateway.

NemoClaw combines all three into a single deployable stack.

Related Articles