langgraph swarm agent

 

https://github.com/bekingcn/langgraph_swarm

LangGraph Swarm

The LangGraph Swarm is a alternative implementation to the OpenAI's Swarm framework with langchain and langgraph.

Coming soon... with full configuration implementation in LcStack

Overview

Why LangGraph Swarm?

LangGraph Swarm reimplements the Swarm framework with langchain and langgraph. With LangGraph Swarm, you can create agents and integrate them with Langchain components. If you already have a langchain agent, you can benefit from Swarm's features without having to re-invent the wheel.

 

import datetime
from collections import defaultdict
from typing import Callable

from langchain_core.runnables import RunnableConfig
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent
from langgraph_swarm import create_handoff_tool, create_swarm

model = ChatOpenAI(model="gpt-4o")

# Mock data for tools
RESERVATIONS = defaultdict(lambda: {"flight_info": {}, "hotel_info": {}})

TOMORROW = (datetime.date.today() + datetime.timedelta(days=1)).isoformat()
FLIGHTS = [
    {
        "departure_airport": "BOS",
        "arrival_airport": "JFK",
        "airline": "Jet Blue",
        "date": TOMORROW,
        "id": "1",
    }
]
HOTELS = [
    {
        "location": "New York",
        "name": "McKittrick Hotel",
        "neighborhood": "Chelsea",
        "id": "1",
    }
]


# Flight tools
def search_flights(
    departure_airport: str,
    arrival_airport: str,
    date: str,
) -> list[dict]:
    """Search flights.

    Args:
        departure_airport: 3-letter airport code for the departure airport. If unsure, use the biggest airport in the area
        arrival_airport: 3-letter airport code for the arrival airport. If unsure, use the biggest airport in the area
        date: YYYY-MM-DD date
    """
    # return all flights for simplicity
    return FLIGHTS


def book_flight(
    flight_id: str,
    config: RunnableConfig,
) -> str:
    """Book a flight."""
    user_id = config["configurable"].get("user_id")
    flight = [flight for flight in FLIGHTS if flight["id"] == flight_id][0]
    RESERVATIONS[user_id]["flight_info"] = flight
    return "Successfully booked flight"


# Hotel tools
def search_hotels(location: str) -> list[dict]:
    """Search hotels.

    Args:
        location: offical, legal city name (proper noun)
    """
    # return all hotels for simplicity
    return HOTELS


def book_hotel(
    hotel_id: str,
    config: RunnableConfig,
) -> str:
    """Book a hotel"""
    user_id = config["configurable"].get("user_id")
    hotel = [hotel for hotel in HOTELS if hotel["id"] == hotel_id][0]
    RESERVATIONS[user_id]["hotel_info"] = hotel
    return "Successfully booked hotel"


# Define handoff tools
transfer_to_hotel_assistant = create_handoff_tool(
    agent_name="hotel_assistant",
    description="Transfer user to the hotel-booking assistant that can search for and book hotels.",
)
transfer_to_flight_assistant = create_handoff_tool(
    agent_name="flight_assistant",
    description="Transfer user to the flight-booking assistant that can search for and book flights.",
)


# Define agent prompt
def make_prompt(base_system_prompt: str) -> Callable[[dict, RunnableConfig], list]:
    def prompt(state: dict, config: RunnableConfig) -> list:
        user_id = config["configurable"].get("user_id")
        current_reservation = RESERVATIONS[user_id]
        system_prompt = (
            base_system_prompt
            + f"\n\nUser's active reservation: {current_reservation}"
            + f"Today is: {datetime.datetime.now()}"
        )
        return [{"role": "system", "content": system_prompt}] + state["messages"]

    return prompt


# Define agents
flight_assistant = create_react_agent(
    model,
    [search_flights, book_flight, transfer_to_hotel_assistant],
    prompt=make_prompt("You are a flight booking assistant"),
    name="flight_assistant",
)

hotel_assistant = create_react_agent(
    model,
    [search_hotels, book_hotel, transfer_to_flight_assistant],
    prompt=make_prompt("You are a hotel booking assistant"),
    name="hotel_assistant",
)

# Compile and run!
builder = create_swarm(
    [flight_assistant, hotel_assistant], default_active_agent="flight_assistant"
)
app = builder.compile()

 

from langchain.chat_models import init_chat_model
from langgraph.prebuilt import create_react_agent
from langgraph_swarm import create_handoff_tool, create_swarm

# from swarm_researcher.configuration import Configuration
from swarm_researcher.prompts import planner_prompt, researcher_prompt
from swarm_researcher.utils import fetch_doc

# LLM
model = init_chat_model(model="gpt-4o", model_provider="openai")

# Handoff tools
transfer_to_planner_agent = create_handoff_tool(
    agent_name="planner_agent",
    description="Transfer the user to the planner_agent for clarifying questions related to the user's request.",
)
transfer_to_researcher_agent = create_handoff_tool(
    agent_name="researcher_agent",
    description="Transfer the user to the researcher_agent to perform research and implement the solution to the user's request.",
)

# LLMS.txt
llms_txt = "LangGraph:https://langchain-ai.github.io/langgraph/llms.txt"
num_urls = 3
planner_prompt_formatted = planner_prompt.format(llms_txt=llms_txt, num_urls=num_urls)

# Planner agent
planner_agent = create_react_agent(
    model,
    prompt=planner_prompt_formatted,
    tools=[fetch_doc, transfer_to_researcher_agent],
    name="planner_agent",
)

# Researcher agent
researcher_agent = create_react_agent(
    model,
    prompt=researcher_prompt,
    tools=[fetch_doc, transfer_to_planner_agent],
    name="researcher_agent",
)

# Swarm
agent_swarm = create_swarm(
    [planner_agent, researcher_agent], default_active_agent="planner_agent"
)
app = agent_swarm.compile()

 

https://zhuanlan.zhihu.com/p/18555531485

https://zhuanlan.zhihu.com/p/25617183612