E-Mail-Assistent mit LangGraph

from langchain_openai import ChatOpenAI
from langchain_core.tools import tool
from langgraph.prebuilt import create_react_agent

@tool
def search_emails(query: str, limit: int = 5) -> str:
    """Search the inbox for emails matching a query."""
    results = email_client.search(query, max_results=limit)
    return "\n\n".join(
        f"From: {e.sender}\nSubject: {e.subject}\nDate: {e.date}\nPreview: {e.body[:200]}"
        for e in results
    )

@tool
def draft_email(to: str, subject: str, body: str) -> str:
    """Create an email draft."""
    draft_id = email_client.create_draft(to=to, subject=subject, body=body)
    return f"Draft created (ID: {draft_id}). Review before sending."

agent = create_react_agent(
    ChatOpenAI(model="gpt-4o"),
    tools=[search_emails, draft_email],
    prompt="You are an email assistant. Search emails to find context, then help draft professional replies. Always create drafts — never send directly.",
)

result = await agent.ainvoke({
    "messages": [("user", "Find the latest email from the marketing team and draft a reply confirming the deadline")]
})
print(result["messages"][-1].content)