Summary

Inside the sandbox, the NemoClaw/OpenClaw experience always reports build.nvidia.com / nemotron-3-super-120b-a12b regardless of the actual configured inference provider.

Problem

Users cannot trust the reported model identity when validating their setup. This is especially confusing when using Ollama or any provider other than the hardcoded default.

Expected behavior

The reported model identity should reflect the active inference provider and model configuration.

Acceptance criteria

• The UI or plugin banner shows the actual configured provider and model.
• Asking the agent which model it is using returns the active configuration, not a baked-in default.
• Configuration changes are reflected without requiring a rebuilt image unless that limitation is explicitly documented.
• A test covers at least one non-default provider.