Nvidia wants enterprises to run agents safely. NemoClaw is how.

Short summary

Nvidia released NemoClaw, an open-source toolkit for running agents safely in enterprise production. The three-component system—OpenShell (sandbox and granular policy controls), Nemotron models (routable locally or with frontier models under privacy controls), and NeMo Agent Toolkit (workflow and function orchestration)—isolates agent execution without constraining reasoning. The design shifts focus from limiting agent capabilities to controlling where agents operate, addressing the enterprise governance gap blocking widespread agent deployment.

• •NemoClaw combines OpenShell (sandboxing + policy controls), Nemotron models (local or routed with privacy), and NeMo Agent Toolkit for production-ready agent governance
• •Enterprise agents get isolated, purpose-built sandboxes per agent and sub-agent; policy engine evaluates permissions at binary, path, and method level
• •Nvidia's business logic: enterprise GPU adoption depends on safe agent deployment; NemoClaw removes the governance blocker for widespread agent production use

Generated with AI, which can make mistakes.