This is a Python framework designed for building real-time speech and multimodal dialogue agents. It can orchestrate speech recognition, large models, and speech synthesis into composable processing pipelines. It enables ultra-low latency interaction through WebSocket or WebRTC. Each pipeline functions as an Agent, supporting multi-agent collaboration modes like task handover, parallel distribution, and cross-process distributed deployment, and is applicable to scenarios such as voice assistants, online customer service, and companion applications.