Build Specialized Agents on a Stack You Control with NVIDIA NemoClaw
According to NVIDIA Developer. NVIDIA announces NemoClaw open blueprints combining a tuned agent harness, Nemotron models, NeMo optimization tools, and the OpenShell secure runtime.
- • NemoClaw open blueprints assemble an open agent stack combining a tuned harness, models, optimization, and runtime controls.
- • LangChain tuned its Deep Agents harness for Nemotron 3 Ultra, achieving near-frontier performance at roughly 10x lower cost per run.
- • OpenShell governs agent access to files, networks, credentials, and tools to enforce security policies during execution.
Introduction to NemoClaw
NVIDIA NemoClaw provides open blueprints that combine a tuned agent harness, Nemotron models, NeMo optimization tools, and the OpenShell secure runtime into a full-stack agent platform. The blueprints currently support OpenClaw, Hermes Agent, and LangChain Deep Agents as the orchestration layer, letting developers keep their existing harness while adding optimization, telemetry, and governance.
Performance and Tuning
LangChain tuned its Deep Agents harness for Nemotron 3 Ultra by adjusting prompts, tools, and middleware without retraining the model. This tuning brought Nemotron 3 Ultra to near-frontier performance at approximately 10x lower cost per run compared with leading closed models, and the tuned profile is available on GitHub.
Runtime Governance
OpenShell governs agent access to files, networks, credentials, and tools, keeping agents bound by policy even when memory or retrieved context suggests unintended actions. The OpenShell tutorial covers how to create policies such as preventing data exfiltration to GitHub.
- 1. developer.nvidia.comOpen source