Systèmes multi-agents
Définition
Les systèmes multi-agents impliquent plusieurs agents IA qui interagissent pour résoudre des tâches: 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: par ex. one agent for RAG récupération, 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.
Comment ça fonctionne
L'utilisateur envoie une tâche à un orchestrateur (qui peut être un LLM ou un workflow fixe). L'orchestrateur attribue le travail à Agent1, Agent2, etc.tc., 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 (par 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.
Cas d'utilisation
Les systèmes multi-agents aident lorsqu'un seul agent ne suffit pas : vous avez besoin de rôles distincts, de débats ou de pipelines modulaires.
- Orchestrating planner, executor, and critic agents for complex tasks
- Debate or review flows where multiple agents refine an answer
- Specialized pipelines (par ex. one agent for récupération, one for generation)
Documentation externe
- 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