[DGX Spark][Onboard] nemoclaw onboard --non-interactive does not auto-select managed local vLLM on DGX Spark

[lyral-commits]

Description

On DGX Spark, running nemoclaw onboard --non-interactive --agent langchain-deepagents-code does not auto-select the managed local vLLM provider as expected. Instead, it falls back to NVIDIA Endpoints (cloud API), even when no cloud API key is configured. Interactive onboarding correctly shows the managed local vLLM option for DGX Spark.

Platform scope: Reproduced on DGX Spark only; other platforms not tested.
Regression: Unknown — earlier versions not tested.

Environment

Device:        DGX Spark (GB10, aarch64)
OS:            Ubuntu 24.04 (aarch64)
Architecture:  aarch64
Node.js:       v22.22.2
npm:           10.9.7
Docker:        28.3.3
OpenShell CLI: 0.0.85
NemoClaw:      v0.0.89
OpenClaw:      N/A (LangChain Deep Agents Code agent)

Steps to Reproduce

1. On DGX Spark, install NemoClaw v0.0.89.
2. Ensure no cloud API key is pre-configured:

nemoclaw credentials reset NVIDIA_API_KEY

3. Run:

nemoclaw onboard --non-interactive --agent langchain-deepagents-code --name dcode-spark-toolcall --gpu --yes --yes-i-accept-third-party-software --fresh --recreate-sandbox

4. Observe the provider selected in the onboard output.

Expected Result

The non-interactive onboard auto-selects managed local vLLM (install-vllm) as the default provider on DGX Spark, using the Qwen3.6-family NVFP4 checkpoint. Output contains no provider fallback to cloud API.

Actual Result

The provider falls back to NVIDIA Endpoints (cloud API) instead of using managed local vLLM:

[non-interactive] Provider: build
Using NVIDIA Endpoints with model: nvidia/nemotron-3-ultra-550b-a55b
Provider:      nvidia-prod
Model:         nvidia/nemotron-3-ultra-550b-a55b

Logs

[3/8] Configuring inference provider
[non-interactive] Provider: build
Using NVIDIA Endpoints with model: nvidia/nemotron-3-ultra-550b-a55b
Provider:      nvidia-prod
Model:         nvidia/nemotron-3-ultra-550b-a55b
[4/8] Setting up inference provider
Updated provider nvidia-prod
Route: inference.local
Provider: nvidia-prod
Model: nvidia/nemotron-3-ultra-550b-a55b