You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Per-invocation sub-agent model overrides (port of upstream mpfaffenberger#377): new invoke_agent_with_model + list_available_models tools let routing agents (e.g. the Orchestrator) run a sub-agent on a specific model for a single call — route cheap/mechanical steps to a fast model, reserve a strong model for hard fixes. Built on BaseAgent.temporary_model_name_override (precedence: runtime → pinned → global); JSONAgent honors it too. The chosen model is surfaced in AgentInvokeOutput.model_name and the invocation UI.
Graceful no-model handling: get_global_model_name() now returns None instead of a phantom gpt-5; the statusline shows [None]; a one-shot warning points at /add_model.
Onboarding nag: after the wizard, if no model is configured (and OAuth wasn't chosen), prompts to run /add_model.
models.json ships empty: first run requires explicit model setup via /add_model or OAuth. Existing users' extra_models.json / OAuth models still merge in automatically.