Bug Description

Running nemoclaw onboard fails at step 4/7 ("Setting up inference provider") with the error:

failed to verify inference endpoint for provider 'nvidia-prod' and model 'nvidia/nemotron-3-super-120b-a12b'
at 'https://integrate.api.nvidia.com/v1': failed to connect to
https://integrate.api.nvidia.com/v1/chat/completions: error sending request for url

The verification request made by openshell inference set fails to connect, even though:

• The NVIDIA API is reachable from the host machine (confirmed via curl)
• The API is reachable from inside the gateway container (confirmed via openssl s_client)
• The API key is valid (confirmed by receiving a successful chat completion response)

This blocks onboarding completely — nemoclaw onboard --resume retries the same step and fails again.

Root Cause

In source/bin/lib/onboard.js, the REMOTE_PROVIDER_CONFIG.build (NVIDIA) config block is missing skipVerify: true, while other providers like OpenAI, Gemini, and custom endpoints all include it:

// NVIDIA config — missing skipVerify
build: {
    label: "NVIDIA Endpoints",
    providerName: "nvidia-prod",
    providerType: "nvidia",
    credentialEnv: "NVIDIA_API_KEY",
    endpointUrl: BUILD_ENDPOINT_URL,
    helpUrl: "https://build.nvidia.com/settings/api-keys",
    modelMode: "catalog",
    defaultModel: DEFAULT_CLOUD_MODEL,
    // skipVerify: true  <-- missing
},

// OpenAI config — has skipVerify
openai: {
    ...
    skipVerify: true,  // ✓
},

This means openshell inference set is called without --no-verify for NVIDIA, and the gateway's internal verification request fails to connect to integrate.api.nvidia.com even though the endpoint is functional.

Fix

Add skipVerify: true to the build entry in REMOTE_PROVIDER_CONFIG:

  build: {
      label: "NVIDIA Endpoints",
      providerName: "nvidia-prod",
      providerType: "nvidia",
      credentialEnv: "NVIDIA_API_KEY",
      endpointUrl: BUILD_ENDPOINT_URL,
      helpUrl: "https://build.nvidia.com/settings/api-keys",
      modelMode: "catalog",
      defaultModel: DEFAULT_CLOUD_MODEL,
+     skipVerify: true,
  },

Environment

• macOS
• nemoclaw installed via source at ~/.nemoclaw/source
• openshell v0.0.16
• Gateway running locally at 127.0.0.1:8080