src/receptionist_agent.py

from langchain_core.prompts import ChatPromptTemplate
from langgraph.graph import StateGraph, END, MessagesState
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import ToolNode
from langchain_core.messages import HumanMessage

from dotenv import load_dotenv

from tools import transfer_call, leave_message, go_to
from prompts import system_prompt, system_prompt_original

load_dotenv()

llm = ChatOpenAI(
    model="gpt-4.1",
    temperature=0.3,
)

# 1. Define the tools
tools = [transfer_call, leave_message, go_to]
tool_node = ToolNode(tools)

# 2. Create the graph
graph = StateGraph(MessagesState)

# 3. Define a new prompt
prompt = ChatPromptTemplate.from_messages(
    [
        (
            "system",
            system_prompt_original,
        ),
        ("placeholder", "{messages}"),
    ]
)

# 4. Define the agent
agent = prompt | llm.bind_tools(tools)


# 5. Define the nodes
def agent_node(state: MessagesState):
    response = agent.invoke(state)
    return {"messages": [response]}


# 6. Define the edges
def should_continue(state: MessagesState):
    last_message = state["messages"][-1]
    if last_message.tool_calls:
        return "continue"
    return "end"


# 7. Build the graph
graph.add_node("agent", agent_node)
graph.add_node("action", tool_node)

graph.set_entry_point("agent")

graph.add_conditional_edges(
    "agent",
    should_continue,
    {
        "continue": "action",
        "end": END,
    },
)
graph.add_edge("action", "agent")

# 8. Compile the graph
app = graph.compile()


# This is a simplified conversation loop for demonstration.
def get_agent_response(conversation: list):
    response = app.invoke({"messages": conversation})
    return response["messages"][len(conversation) :]