app_langgraph.py

"""LangGraph Agent"""
import os
from dotenv import load_dotenv
from langgraph.graph import START, StateGraph, MessagesState
from langgraph.prebuilt import tools_condition
from langgraph.prebuilt import ToolNode
from langchain_huggingface import HuggingFaceEndpoint, ChatHuggingFace, HuggingFaceEmbeddings
from langchain_core.messages import SystemMessage, HumanMessage
from langchain_core.globals import set_debug
from langchain_groq import ChatGroq
from search_tools import web_search, arvix_search, wiki_search
from math_tools import multiply, add, subtract, divide
# from supabase.client import Client, create_client
# from langchain.tools.retriever import create_retriever_tool
# from langchain_community.vectorstores import SupabaseVectorStore
import json
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_openai import ChatOpenAI
from langchain_core.messages import HumanMessage

# set_debug(True)
load_dotenv()

tools = [
    multiply,
    add,
    subtract,
    divide,
    web_search,
    wiki_search,
    arvix_search
]

def build_graph():
    dashscope_key = os.getenv("CODING_PLAN_API_KEY")  # 从 Coding Plan 页面获取
    
    llm = ChatOpenAI(
        api_key=dashscope_key,
        base_url="https://coding.dashscope.aliyuncs.com/v1",  # ⚠️ 关键：DashScope 兼容地址
        model="qwen3.5-plus",  # 支持 qwen-plus, qwen-max, qwen-turbo 等
        temperature=0.1,
        verbose=True
    )
    
    chat_with_tools = llm.bind_tools(tools)

    def assistant(state: MessagesState):
        sys_msg = "You are a helpful assistant with access to tools. Understand user requests accurately. Use your tools when needed to answer effectively. Strictly follow all user instructions and constraints." \
        "Pay attention: your output needs to contain only the final answer without any reasoning since it will be strictly evaluated against a dataset which contains only the specific response." \
        "Your final output needs to be just the string or integer containing the answer, not an array or technical stuff."
        return {
            "messages": [chat_with_tools.invoke([sys_msg] + state["messages"])],
        }

    ## The graph
    builder = StateGraph(MessagesState)

    builder.add_node("assistant", assistant)
    builder.add_node("tools", ToolNode(tools))

    builder.add_edge(START, "assistant")
    builder.add_conditional_edges(
        "assistant",
        # If the latest message requires a tool, route to tools
        # Otherwise, provide a direct response
        tools_condition,
    )
    builder.add_edge("tools", "assistant")
    return builder.compile()

# test
if __name__ == "__main__":

    graph = build_graph()
    with open('sample.jsonl', 'r') as jsonl_file:
        json_list = list(jsonl_file)
    
    start = 10 #revisit 5, 8, 
    end = start + 1
    for json_str in json_list[start:end]:
        json_data = json.loads(json_str)
        print(f"Question::::::::: {json_data['Question']}")
        print(f"Final answer::::: {json_data['Final answer']}")
    
        question = json_data['Question']
        messages = [HumanMessage(content=question)]
        messages = graph.invoke({"messages": messages})
        for m in messages["messages"]:
            m.pretty_print()