[Brev][Ubuntu 22.04][Github Issue #] nemoclaw status does not reflect runtime inference changes from openshell inference set

[JoyceChenNV]

Description

Running: openshell inference set --provider nvidia-nim --model nvidia/nemotron-3-super-120b-a12b

successfully updates the gateway inference route. openshell inference get shows the new provider/model, but nemoclaw status still shows the sandbox with the previous model, which makes it appear that the sandbox inference provider was not updated.

Reproduction Steps

1. Open https://build.nvidia.com/nemoclaw and click "Try Early Preview" button to open brev launchable like https://brev.nvidia.com/launchable/deploy/now?launchableID=env-3Azt0aYgVNFEuz7opyx3gscmowS
2. Click on "Deploy Launchable" button
3. Click "Code-Server" when instance is ready
4. During sandbox onboard, set the inference provider to qwen/qwen3.5-397b-a17b
5. Then follow the guide "Switch inference providers" as run openshell inference set --provider nvidia-nim --model nvidia/nemotron-3-super-120b-a12b

Environment

• Device: Brev instance Ubuntu 22.04.5 LTS
• Node.js: v22.22.1
• Docker: Docker Engine 29.3.0
• OpenShell CLI: 0.0.10
• NemoClaw: v0.1.0 (6e1208c)
• OpenClaw: 2026.3.11 (29dc654)

Debug Output

Logs

ubuntu@brev-1gfvh46jl:~$ openshell inference set --provider nvidia-nim --model nvidia/nemotron-3-super-120b-a12b
Gateway inference configured:

  Route: inference.local
  Provider: nvidia-nim
  Model: nvidia/nemotron-3-super-120b-a12b
  Version: 2
  Validated Endpoints:
    - https://integrate.api.nvidia.com/v1/chat/completions (openai_chat_completions)

ubuntu@brev-1gfvh46jl:~$ openshell inference get
Gateway inference:

  Provider: nvidia-nim
  Model: nvidia/nemotron-3-super-120b-a12b
  Version: 2
ubuntu@brev-1gfvh46jl:~$ nemoclaw brev-test status

  Sandbox: brev-test
    Model:    qwen/qwen3.5-397b-a17b
    Provider: nvidia-nim
    GPU:      no
    Policies: pypi, npm
Sandbox:

  Id: c6df6f45-8c00-40e4-8be8-2c3b39fe58ce
  Name: brev-test
  Namespace: openshell
  Phase: Ready

Checklist

• I confirmed this bug is reproducible
• I searched existing issues and this is not a duplicate

[NVB#6014112]