Organisation-controlled assistants and agent capabilities are combined with models that can be swapped.
Proposed mechanism: Keep key context and agent layers controllable while switching model providers.
Conditions: Interchangeable components.
Limit: Portability is asserted, not independently tested.