# NVIDIA Details NemoClaw Support Across Three AI Agent Frameworks

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/45026
Published: 2026-10-30T18:50:31.000Z
Updated: 2026-10-30T18:50:31.000Z
Section: Technology

> The blueprints pair agent harnesses with Nemotron models, NeMo tools and OpenShell controls. Agents built with NemoClaw can run in cloud environments, data centers and on NVIDIA PCs.

NVIDIA’s NemoClaw blueprints bring together three AI agent frameworks with the company’s models, developer tools and access controls, giving developers configurations for building specialized agents around different harnesses.

The currently supported harnesses are OpenClaw, Hermes Agent and LangChain Deep Agents. Each blueprint combines a tuned harness with Nemotron open models, NeMo tools and NVIDIA OpenShell. NeMo Relay provides visibility into agent execution, while OpenShell governs access to files, networks, credentials and tools.

A harness provides an agent with tools, skills and memory, and orchestrates its task execution. OpenShell sets boundaries on what the agent can access. NemoClaw is the blueprint system that assembles these components; it is not itself a competing harness or runtime.

Agents built with NemoClaw can run in cloud environments and data centers, as well as on DGX Spark and RTX PCs and workstations. The combinations let developers use different harnesses with NVIDIA’s models and controls across those deployment environments.

On the Deep Agents suite, a tuned setup using Nemotron 3 Ultra scored 0.86, compared with 0.87 for Opus 4.8. The full-suite runs cost about $4.48 and $43.48, respectively. The scores and costs apply to the tested configurations on the Deep Agents suite.

NVIDIA announced NemoClaw on March 16 as a stack for OpenClaw. LangChain announced its Deep Agents blueprint on July 8.