Deep-Research-Agent mit Sub-Agenten

from deepagents import create_deep_agent
from langchain.chat_models import init_chat_model
from langchain_core.tools import tool

@tool
def web_search(query: str, max_results: int = 5) -> dict:
    """Search the web for current information."""
    from tavily import TavilyClient
    client = TavilyClient()
    return client.search(query, max_results=max_results)

research_sub_agent = {
    "name": "researcher",
    "description": "Delegate research to a specialist sub-agent.",
    "system_prompt": "You are a research specialist. Search thoroughly and synthesize findings.",
    "tools": [web_search],
}

agent = create_deep_agent(
    model=init_chat_model("anthropic:claude-sonnet-4-5-20250929"),
    tools=[web_search],
    system_prompt="You are a research orchestrator. Break complex topics into sub-tasks and delegate to researchers.",
    subagents=[research_sub_agent],
)

result = agent.invoke({"messages": [("user", "Research the latest advances in AI agent architectures")]})