Description

Users currently must use openshell inference set to switch the active inference provider or model at runtime. NemoClaw should expose a native nemoclaw inference wrapper that surfaces this capability without requiring direct use of the lower-level openshell CLI.

Proposed UX

nemoclaw inference set --provider nvidia-prod --model <model-name>
nemoclaw inference get

Current Workaround

openshell inference set -g nemoclaw --provider <provider> --model <model-name>
openshell inference get

Acceptance Criteria

• nemoclaw inference set changes the active provider/model for the running sandbox
• nemoclaw inference get returns the current live provider and model
• Output matches what nemoclaw <sandbox> status reports
• Related: [Linux][Agent] openclaw tui footer shows stale model after runtime inference switch #2608 (TUI footer stale after runtime switch — should reflect the value set via this command)