부록 · 치트시트 · 용어집 · 추가 학습

from agents import Agent, Runner, function_tool, SQLiteSession

@function_tool
def my_tool(x: str) -> str:
    """도구 설명 — 명확하게."""
    return ...

agent = Agent(name="A", instructions="역할", tools=[my_tool])
result = Runner.run_sync(agent, "입력")          # 동기
# await Runner.run(agent, "입력", session=...)   # 비동기+메모리

# 멀티에이전트: 도구로
manager = Agent(name="M", tools=[agent.as_tool("name", "desc")])
# 핸드오프
triage = Agent(name="T", handoffs=[billing_agent, refund_agent])
# 가드레일
agent = Agent(name="A", input_guardrails=[my_guardrail])

from langgraph.graph import StateGraph, START, END
from langgraph.checkpoint.memory import InMemorySaver

builder = StateGraph(State)
builder.add_node("n", node_fn)
builder.add_edge(START, "n")
builder.add_conditional_edges("n", router_fn, {"a": "node_a", "b": END})
graph = builder.compile(checkpointer=InMemorySaver())
graph.invoke({...}, {"configurable": {"thread_id": "t1"}})

from crewai import Agent, Task, Crew, Process

agent = Agent(role="역할", goal="목표", backstory="배경", tools=[...])
task = Task(description="...", expected_output="...", agent=agent)
crew = Crew(agents=[...], tasks=[...], process=Process.sequential)
crew.kickoff()
# Hierarchical은 manager_llm 또는 manager_agent 필수

from fastmcp import FastMCP
mcp = FastMCP("Server")

@mcp.tool
def add(a: int, b: int) -> int:
    """두 수를 더한다."""
    return a + b

mcp.run(transport="http", port=8000)
# 연결: agents.mcp.MCPServerStreamableHttp(params={"url": ...})

트리거 노드(Chat/Webhook/Schedule)
  └→ AI Agent 노드
        ├─ Chat Model (OpenAI/Claude/Ollama)
        ├─ Memory (Simple/Window/Postgres)
        ├─ Tool (검색/HTTP/앱/벡터스토어)
        └─ Output Parser (구조화 출력)
  └→ 출력 노드(Slack/Sheets/Gmail) · Switch로 분기