多智能体系统架构

横向 16:9 高保真系统图，描绘多智能体 LLM 架构，风格类似细致的 AutoGen / LangGraph / Anthropic Managed Agents 图 1。细微阴影，暖铜色高光，编号流程标记 ①②③④。 区域 1 — "User interface"：圆角用户框，内含占位任务 "research question: summarize recent red-teaming attacks and reproduce the top three"。 区域 2 — "Orchestrator layer"：中心六边形节点 "Planner LLM"，顶部边缘暖铜色。三颗卫星芯片："Task decomposition"、"Agent routing"、"Re-plan on failure"。小嵌入芯片 "prompt cache hit ~98%"。 区域 3 — "Specialised workers"：2×2 六边形，分别为 "Researcher" / "Coder" / "Critic" / "Writer"，每个带有符号和状态条带（"idle"、"running step 3/5"、"done"、"running step 2/4"）。中央标注 "async message bus"。 区域 4 — "Tools & memory"： (a) "Tool registry" 面板列出 "web_search ×41"、"python_exec ×27"、"read_file ×18"、"write_file ×12"、"browser_use ×7"；(b) "Memory" 面板含 "Short-term scratchpad" 和圆柱体 "Long-term vector store — 1.8M episodes"。 底部嵌入 "Example trace"：8步水平时间线，从 "User asks" 到 "Planner decomposes"、"Researcher: web_search(...)"、"Coder: python_exec(...)"、"Critic: verify"、"Re-plan"（循环箭头）、"Writer: compose final answer"。 标题："Agentic LLM system: planner orchestrates specialised workers over a shared tool and memory layer"。副标题："adapted from AutoGen (Wu et al., 2023), LangGraph, and Anthropic Managed Agents patterns"。