Multi-Agent Architecture (MAA) is an artificial intelligence system architecture in which multiple agents collaborate to complete complex tasks. This architecture achieves more efficient and reliable task execution by breaking down tasks into multiple sub-tasks, each handled by a different specialized agent. Originating from research on multi-agent systems, this concept gained widespread application around 2023 with the development of agent technologies based on large language models.

In a multi-agent architecture, different agents typically assume different roles. These agents interact through message passing or sharing context to collaboratively complete the overall task. Compared to a single-agent architecture, this model reduces the cognitive burden on a single model through division of labor and cooperation, thereby improving the system's robustness and scalability.