Skip to content

[Multiple Platforms][Onboard] NemoClaw does not pull Ollama model during onboard when no models are present #6038

Closed

Description

@wangericnv

Description

When running nemoclaw onboard with Ollama installed but no models pulled, NemoClaw should detect the absence of models, prompt the user to select a model, and pull it during onboard. On DGX Spark, the model pull step is silently skipped — model count remains 0 after onboard completes. On ubuntu22/ubuntu24/DGX Station, the interactive wizard also fails to complete (wizard navigation timeout), suggesting the onboard flow for the Ollama-no-models case is broken across multiple platforms.

Environment

Device:        DGX Spark (GB10) / Ubuntu 22.04 / Ubuntu 24.04 / DGX Station
OS:            Linux
Architecture:  x86_64 / arm64
NemoClaw:      v0.0.69
OpenClaw:      2026.5.27
Ollama:        installed, 0 models pulled

Steps to Reproduce

  1. Install NemoClaw v0.0.69 on DGX Spark
  2. Ensure Ollama is installed with no models pulled (ollama list returns empty)
  3. Run: nemoclaw onboard
  4. Select Ollama as inference provider
  5. When prompted, select a model to pull
  6. Complete onboard
  7. Run: ollama list

Expected Result

NemoClaw pulls the selected Ollama model during onboard. After onboard completes, ollama list shows at least 1 model.

Actual Result

On DGX Spark (non-interactive test): Onboard completes but model count is 0.

Test failure: "NemoClaw should have pulled the selected model during onboard: expected 0 to be greater than 0"

On ubuntu22/ubuntu24/DGX Station (interactive test): Wizard navigation times out — onboard never reaches completion step.

Test failure: "Not all wizard steps completed."

Logs

CI pipeline 56214563 (v0.0.69 non-sanity3):
- dgxspark job 350535726: T5882219 FAIL 14s — "expected 0 to be greater than 0" (model count)
- ubuntu22 job 350535730: T5882219 FAIL 13m — "Not all wizard steps completed"
- ubuntu24 job 350535729: T5882219 FAIL 15m — "Not all wizard steps completed"
- dgx-station job 350535727: T5882219 FAIL 4s — "Not all wizard steps completed"
- u24-gpu job 350535728: T5882219 PASS
- wsl-x86 job 350535724: T5882219 PASS
- macos job 350535725: T5882219 PASS

NVB#6395743

Activity

  1. changed the title [-][DGX Spark][Onboard] NemoClaw does not pull Ollama model during onboard when no models are present[/-] [+][Multiple Platforms][Onboard] NemoClaw does not pull Ollama model during onboard when no models are present[/+] on Jun 30, 2026
  2. added
    NV QABugs found by the NVIDIA QA Team
    needs: triageAwaiting maintainer classification
    on Jun 30, 2026
  3. added
    area: local-modelsLocal model providers, downloads, launch, or connectivity
    area: onboardingOnboarding FSM, provider setup, sandbox launch, or first-run flow
    provider: ollamaOllama local model provider behavior
    on Jun 30, 2026
  4. added theissue type on Jun 30, 2026
  5. self-assigned this
    on Jul 8, 2026
  6. added a commit that references this issue on Jul 8, 2026
    5ddf9a1
  7. added a commit that references this issue on Jul 12, 2026
    9e1f887
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

NV QABugs found by the NVIDIA QA Teamarea: local-modelsLocal model providers, downloads, launch, or connectivityarea: onboardingOnboarding FSM, provider setup, sandbox launch, or first-run flowprovider: ollamaOllama local model provider behavior

Type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions