mesh-llm-moa-plugin

Built-in virtual-model plugin that exposes MeshLLM's Mixture-of-Agents engine through the generic plugin protocol. The plugin owns MoA orchestration while the host retains model discovery, placement, nested inference, accounting, and OpenAI-compatible response framing.

The built-in MoA policy strips thinking text from direct and aggregated public answers. The current path does not preserve a separate copy of that text for later use; a future arbitration policy that needs it requires an explicit change to the worker response handling.