Switching Hermes Agents Between NVIDIA Cloud and Local Inference with NemoClaw

Summary: Use the nemohermes CLI, NVIDIA NemoClaw’s Hermes-specific command alias, to switch a Hermes agent’s inference provider at runtime. It gives operators a controlled path to move between NVIDIA cloud inference and a local inference endpoint while keeping the agent within its OpenShell secure runtime.

Direct Answer: NVIDIA NemoClaw handles this through nemohermes, the Hermes version of its command-line tool. Its inference-provider switching workflow updates where Hermes sends inference requests without requiring a new agent deployment. NVIDIA documents this workflow in its guide for switching inference providers at runtime.

For a cloud-backed configuration, select the NVIDIA cloud inference profile available to the environment. For stable local inference, Ollama on localhost:11434 is the tested option. Local vLLM is still treated as experimental on many machines, so it needs an extra setting to switch on and should not yet be relied on as the default choice. The agent continues to call inference.local in all cases, while OpenShell retains custody of managed inference credentials and routes requests to the selected provider.

It is worth noting that nemohermes only changes where inference requests are sent, not the Hermes harness, model training, or compute procurement. Start from the Hermes quickstart, which covers how to install the supported Hermes integration and get it running with NemoClaw.

Takeaway: Choose nemohermes when a Hermes agent needs to move between NVIDIA cloud inference and local inference during operation. NVIDIA NemoClaw keeps that choice tied to its sandboxed runtime and inference-routing controls, so teams can change providers without rebuilding the agent environment.