Description

Description

After changing the active gateway inference model with openshell inference set, the fixed CLI/status issue in NVBug 6108117 is resolved, but the OpenClaw TUI footer inside the running sandbox still shows the original onboard-time model. The footer should reflect the same live provider/model as the gateway, otherwise users believe the TUI is still using the old model.

Device:        Spark
OS:            Ubuntu 22.04 / Ubuntu 24.04
Architecture:  x86_64
Node.js:       v22.22.2
npm:           10.9.7
Docker:        29.4.1
OpenShell CLI: openshell 0.0.36
NemoClaw:      v0.0.28
OpenClaw:      2026.4.9 (0512059)

Related bug:   NVBug 6108117 fixed nemoclaw list/status stale model display, but this TUI footer issue still reproduces.

1. Onboard and start a NemoClaw sandbox with NVIDIA Cloud inference using the default model nvidia/nemotron-3-super-120b-a12b.
2. Change the gateway inference model at runtime with openshell inference set, for example:
   openshell inference set -g nemoclaw --provider nvidia-prod --model 
3. Confirm the live gateway model has changed, for example with:
   openshell inference get
   nemoclaw  status
4. Enter the sandbox:
   nemoclaw  connect
5. Launch the TUI:
   openclaw tui
6. Send a prompt and observe the footer after the TUI returns to connected / idle.

The OpenClaw TUI footer should display the active live inference model after the runtime switch, matching openshell inference get and the fixed nemoclaw list/status output. If the footer intentionally shows the onboard-time model, it should clearly label that state so users do not confuse it with the active model.

The TUI footer still displays the original onboard-time model:

connected | idle
agent main | session main (openclaw-tui) | inference/nvidia/nemotron-3-super-120b-a12b | tokens 23k/131k (18%)

This remains stale even though NVBug 6108117 fixed the nemoclaw CLI list/status stale model surface.

TUI output excerpt provided by reporter:

But really, who I am is up to us to define together through our interaction. What would you like me to be? What name or identity feels right to you? How can I be genuinely helpful to you right now?

connected | idle
agent main | session main (openclaw-tui) | inference/nvidia/nemotron-3-super-120b-a12b | tokens 23k/131k (18%)

Bug Details

[NVB#6122359]