Description

During nemoclaw onboard --fresh, step [3/8] prints the banner [3/8] Configuring inference (NIM) even when the user selects a non-NIM provider such as NVIDIA Endpoints, OpenAI, or Local Ollama and never configures a local NIM backend. The current QA test assumes the (NIM) qualifier should only appear when a NIM-local provider is actually being set up, so this generic banner causes the test to fail even though onboarding itself succeeds.

Component area: Onboard.

Environment

Host OS:        Linux (bare metal), NVIDIA GPU present
Docker:         Docker CE, user in docker group
nemoclaw:       v0.0.46
openshell:      0.0.39
State:          no prior NemoClaw onboarding (or running with --fresh)

Steps to Reproduce

Preconditions:

• NemoClaw CLI installed and Docker running.
• No existing onboarding state for this user, or run with --fresh to reset.

Repro:

1. Run onboarding from the host shell:

   nemoclaw onboard --fresh

2. Watch the output until step [3/8]:

   [3/8] Configuring inference (NIM)
   ──────────────────────────────────────────────────
   Detected local inference option: Ollama
   Inference options:
     1) NVIDIA Endpoints
     2) OpenAI
     3) Other OpenAI-compatible endpoint
     4) Anthropic
     5) Other Anthropic-compatible endpoint
     6) Google Gemini
     7) Local Ollama (localhost:11434) — running (suggested)
     8) Model Router (experimental)
   Choose :
   

3. At the inference menu, choose any provider that is NOT a NIM-local profile (e.g., 1 NVIDIA Endpoints, 2 OpenAI, or 7 Local Ollama).
4. Continue through the rest of onboarding and complete provider-specific configuration.

Expected Result

For step [3/8], the banner should be provider-agnostic unless and until a NIM-local provider is actually being configured, e.g.:

• [3/8] Configuring inference, or
• [3/8] Inference provider

The (NIM) qualifier should only appear if:

• The user explicitly chooses a NIM-local profile, AND
• The wizard is configuring NIM as the inference backend.

After the user selects a non-NIM provider (NVIDIA Endpoints, OpenAI, Local Ollama, etc.):

• No log line should imply that a NIM backend is being configured.
• Onboarding should proceed directly into the prompts relevant to the chosen provider.

Actual Result

Step [3/8] is always labeled:

[3/8] Configuring inference (NIM)

even when the user selects only non-NIM providers and never configures a local NIM server.

Onboarding then continues normally into the chosen provider's configuration, so functionality is correct, but the step label is misleading and conflicts with tests that treat (NIM) as meaning "you are configuring a NIM-local backend right now."

Impact

• QA tests that assert "the step 3 banner must not contain (NIM) when a non-NIM provider is selected" will consistently fail, even though provider selection and configuration work as intended.
• The banner text blurs the distinction between "Configuring inference" (generic) and "Configuring a local NVIDIA NIM backend" (specific), which can confuse users and makes it harder to reason about which inference path is actually in use.

Suggested Fix

• Make step [3/8] label provider-agnostic by default — change header to [3/8] Configuring inference or [3/8] Inference provider.
• Optionally add a separate, conditional note only on NIM-local flows, e.g., "Using local NVIDIA NIM profile …".
• Alternatively, if (NIM) should remain in the header, make it conditional — show [3/8] Configuring inference (NIM) ONLY when a NIM-local provider is being configured, and a generic version for all other providers.

NVB#6197212