An RL-central Framework for Language and Vision Assistant, which decouples algorithm logic from distributed execution, enables modular customization.