About Inference Routing
NemoClaw gives agents a managed inference path while OpenShell handles the selected upstream provider on the host. This design keeps provider credentials outside the sandbox.
Request Path
New NVIDIA Endpoints, OpenAI, Anthropic, Google Gemini, OpenRouter, and Hermes Provider selections use sandbox-attached OpenShell providers. The agent calls the selected provider’s API hostname with a credential placeholder. OpenShell checks the destination, executable, and allowed request paths, then supplies the host-managed credential.
Each sandbox records its selected provider attachment. Changing one sandbox’s native selection does not change another sandbox’s route. OpenRouter attribution headers remain in the managed agent configuration. For an endpoint returned by Hermes login, NemoClaw records its accepted public IP addresses and restricts the provider and sandbox policy to those addresses. A different destination cannot gain access through a later DNS change.
Existing sandboxes without a native attachment retain their recorded shared route. Legacy local routes have the recovery restrictions described under Credential Boundary. New Deep Agents Code, Hermes, OpenClaw, and Pi selections for Ollama, standalone vLLM, and llama.cpp use a sandbox-attached provider for the qualified host-local endpoint. Deep Agents Code also uses an attached provider for hosted compatible endpoints. Deep Agents Code, Hermes, and Pi use attached providers for qualified host-local compatible endpoints. OpenClaw compatible endpoints, other custom endpoints without an attachment, Local NVIDIA NIM, Model Router, and unchanged published Pi and Hermes shared local selections retain the shared route. Hermes live qualification for the new local lifecycle remains pending.
Use nemoclaw inference get to inspect the selected sandbox’s recorded route.
status, doctor, and startup checks verify its recorded native provider identity before using a native attachment.
Missing or conflicting ownership stops the operation instead of falling back to the shared route.
Credential Boundary
Provider credentials stay on the host and flow through the OpenShell provider system. The sandbox does not receive the raw upstream API key.
Local Ollama and local vLLM routes do not require the host OPENAI_API_KEY.
NemoClaw uses provider-specific local tokens for those routes.
For retained shared local routes, rebuilds migrate away from stale OpenAI credential requirements.
Legacy Deep Agents Code local sandboxes without an attachment require a backup and fresh recreation. Existing OpenClaw Ollama, standalone vLLM, and llama.cpp sandboxes without an attachment retain supported model switches on their shared route, but cannot resume or rebuild. Back one up and re-onboard with --fresh --name <sandbox-name> --recreate-sandbox. Owned Local NVIDIA NIM sandboxes retain the shared route and can resume or rebuild.
Legacy Hermes Ollama, standalone vLLM, llama.cpp, and qualifying host-local compatible routes without a native attachment cannot resume or rebuild.
Back up required sandbox state before running nemohermes onboard --fresh --name <sandbox-name> --recreate-sandbox to replace the sandbox with a native attachment.
Owned Local NVIDIA NIM and nonqualifying custom routes retain their shared route.
For an attached Hermes route, startup refreshes the credential reference before starting the gateway.
Rebuild refreshes only the restored credential references, preserving the selected model and provider settings.
When a rebuild reuses an automatically bridged compatible-endpoint route without a host API key, NemoClaw reapplies the config-only bridge rewrite without reading or passing the credential stored in OpenShell.
Related Topics
- Choose an Inference Provider compares the upstream routes that OpenShell can use.
- View the Active Inference Route shows the provider and model currently selected.
- Verify the Sandbox Inference Route tests the configured path.