Summary

The blueprint.yaml defines profiles for NVIDIA Cloud (default), NCP (ncp), local NIM (nim-local), and vLLM (vllm) — but no Azure-hosted inference option.

NVIDIA NIM is available on Azure via Microsoft Foundry. Adding an azure-nim profile with Azure-specific authentication would unlock enterprise deployments on Azure infrastructure.

Current behavior

There are 4 inference profiles in blueprint.yaml:

• default → integrate.api.nvidia.com/v1 with NVIDIA_API_KEY
• ncp → dynamic endpoint with NVIDIA_API_KEY
• nim-local → nim-service.local:8000/v1 with NIM_API_KEY
• vllm → localhost:8000/v1 with OPENAI_API_KEY

The ncp profile supports custom endpoints but does not handle Entra ID / Managed Identity authentication, which is the standard auth pattern for Azure-hosted NIM.

Proposed change

Add an azure-nim profile to blueprint.yaml and the TypeScript/JS onboarding flow:

1. blueprint.yaml — new azure-nim profile with:

   • provider_type: "openai" (NIM on Azure exposes an OpenAI-compatible API)
   • credential_env: "AZURE_NIM_API_KEY"
   • dynamic_endpoint: true for user-configurable Microsoft Foundry endpoint URL

2. nemoclaw/src/commands/onboard.ts — add "azure-nim" to the EndpointType union and interactive prompt

3. Network policy — add an azure_nim policy entry allowing egress to *.inference.ml.azure.com and *.services.ai.azure.com

4. Docs — document the new profile in inference-profiles.md and README.md

Why this matters

• Many enterprises run NIM on Azure via Microsoft Foundry
• Entra ID / Managed Identity is the standard auth pattern — not just API keys
• SAP + Azure shops running Nemotron on Azure infrastructure need this for production
• NVIDIA's partnership with Microsoft/Azure makes this a natural addition

References

• NVIDIA NIM on Azure
• Current profiles: docs/reference/inference-profiles.md