Sistemas multi-agente
Definição
Sistemas multi-agente envolvem múltiplos agentes de IA que interagem para resolver tarefas: collaboration (divide work, share state), debate (argue and refine answers), or specialized roles (planner, executor, critic).
They extend single agents when one model or one loop is insufficient: por ex. one agent for RAG recuperação, another for generation, another for critique. Subagents are a hierarchical form where a root agent delegates to children; here we focus on flat or peer-to-peer multi-agent patterns.
Como funciona
O usuário envia uma tarefa a um orquestrador (que pode ser um LLM ou um fluxo de trabalho fixo). O orquestrador atribui trabalho ao Agente1, Agente2, etc., each with its own role, tools, and optionally model. Agents may share a common state, pass messages, or be invoked in sequence/parallel. Their outputs are aggregated (por ex. combined, voted, or summarized) and returned to the user. Design choices include role assignment, communication protocol, and conflict resolution. MAS are useful when you want modularity (each agent has a clear responsibility), specialization (different models or tools per role), reusability (same agent in different workflows), and structured control flow.
Casos de uso
Sistemas multi-agente ajudam quando um único agente não é suficiente: você precisa de papéis distintos, debate ou pipelines modulares.
- Orchestrating planner, executor, and critic agents for complex tasks
- Debate or review flows where multiple agents refine an answer
- Specialized pipelines (por ex. one agent for recuperação, one for generation)
Documentação externa
- From Prototypes to Agents with ADK – Google Codelabs — ADK supports composing multiple agents into a multi-agent system
- LangChain – Multi-agent — Multi-agent orchestration patterns