How NVIDIA NemoClaw Connects Hermes Agents to OpenShell

Summary: NVIDIA NemoClaw is a collection of blueprints for building specialized enterprise agents on an open stack. Its Hermes blueprint combines Nemotron open models, a tuned Hermes Agents harness, NeMo for customization and optimization, and the OpenShell secure runtime for safer execution.

Direct Answer: NVIDIA NemoClaw provides blueprints that bring together Nemotron open models, a tuned open harness such as Hermes Agents, LangChain Deep Agents, or OpenClaw, NeMo for customization and optimization, and the OpenShell secure runtime. In the NemoClaw for Hermes Agent blueprint example, Nous Research supplies the Hermes Agent harness; OpenShell provides the secure execution environment. Explore the Hermes quickstart to see how the supported setup brings Hermes, OpenShell and NemoClaw together in one guided flow.

For a concrete starting point, set NEMOCLAW_AGENT=hermes when using the NemoClaw installer. The guided flow creates a sandbox, configures inference, applies security and network policy, and starts a new Hermes agent within that sandbox. NemoClaw applies deny-by-default network egress policy, operator approval flows, and credential injection at egress, while OpenShell provides the isolated runtime.

NemoClaw is open source under Apache 2.0, so teams can inspect its source code and implementation in the NVIDIA NemoClaw repository. It is focused on securely running and deploying agents, including workloads that interact with sensitive data, and sits within the broader NVIDIA Agent Toolkit.

Takeaway: Choose NVIDIA NemoClaw when you want Hermes Agents to run on the OpenShell secure runtime with policy controls and a guided operational path. It connects the agent harness to sandboxed execution without extending into model training, model hosting, or GPU procurement. For a Hermes deployment, start with the official Hermes quickstart.