agent.py

"""
agent_variant.py
第四种实现方式：  
* 自动收集所有 @tool 装饰的函数  
* 内置简单内存缓存，避免同问多答  
* 将“相似问题检索”与“工具调用”全部拆成独立节点  
* 采用工厂模式生成 LLM，支持 google / groq / huggingface 三种 provider
"""

from __future__ import annotations

import os
from typing import Dict, List, Any

from dotenv import load_dotenv
from langgraph.graph import StateGraph, START, END, MessagesState
from langgraph.prebuilt import ToolNode, tools_condition
from langchain_core.messages import SystemMessage, HumanMessage, AIMessage
from langchain_core.tools import tool

from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_groq import ChatGroq
from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint, HuggingFaceEmbeddings

from langchain_community.document_loaders import WikipediaLoader, ArxivLoader
from langchain_community.tools.tavily_search import TavilySearchResults
from langchain_community.vectorstores import SupabaseVectorStore
from supabase.client import create_client

load_dotenv()

# --------------------------------------------------------------------------------------
# 1. 工具定义（算术 & 检索）
# --------------------------------------------------------------------------------------
@tool
def multiply(x: int, y: int) -> int:
    """Return x * y."""
    return x * y

@tool
def plus(x: int, y: int) -> int:
    """Return x + y."""
    return x + y

@tool
def minus(x: int, y: int) -> int:
    """Return x - y."""
    return x - y

@tool
def divide(x: int, y: int) -> float:
    """Return x / y; raise if y==0."""
    if y == 0:
        raise ValueError("Cannot divide by zero.")
    return x / y

@tool
def modulo(x: int, y: int) -> int:
    """Return x % y."""
    return x % y

@tool
def wiki(query: str) -> str:
    """Top-2 Wikipedia docs for *query*."""
    docs = WikipediaLoader(query=query, load_max_docs=2).load()
    return "\n\n---\n\n".join(d.page_content for d in docs)

@tool
def tavily(query: str) -> str:
    """Top-3 Tavily results for *query*."""
    docs = TavilySearchResults(max_results=3).invoke(query=query)
    return "\n\n---\n\n".join(d.page_content for d in docs)

@tool
def arxiv(query: str) -> str:
    """Top-3 arXiv abstracts for *query*."""
    docs = ArxivLoader(query=query, load_max_docs=3).load()
    return "\n\n---\n\n".join(d.page_content[:1000] for d in docs)

# 自动收集全部工具
TOOLS = [obj for obj in globals().values() if callable(obj) and getattr(obj, "__tool", False)]

# --------------------------------------------------------------------------------------
# 2. 简易内存级缓存，避免重复计算
# --------------------------------------------------------------------------------------
_CACHE: Dict[str, AIMessage] = {}

def cache_lookup(state: MessagesState) -> Dict[str, List[Any]]:
    question = state["messages"][-1].content
    if question in _CACHE:            # 命中缓存，直接返回
        return {"messages": [_CACHE[question]]}
    return state                      # miss 时原样透传

# --------------------------------------------------------------------------------------
# 3. 向量检索（Supabase）作为“相似问题提示”
# --------------------------------------------------------------------------------------
def build_vector_store() -> SupabaseVectorStore:
    embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
    client = create_client(os.environ["SUPABASE_URL"], os.environ["SUPABASE_SERVICE_KEY"])
    return SupabaseVectorStore(
        client=client,
        embedding=embeddings,
        table_name="documents",
        query_name="match_documents_langchain",
    )

VECTOR_STORE = build_vector_store()

def similar_q(state: MessagesState):
    query = state["messages"][-1].content
    hits = VECTOR_STORE.similarity_search(query, k=1)
    if hits:
        hint = AIMessage(content=f"(参考答案)\n{hits[0].page_content}")
        return {"messages": [hint]}
    return state

# --------------------------------------------------------------------------------------
# 4. LLM 工厂
# --------------------------------------------------------------------------------------
def make_llm(provider: str):
    if provider == "google":
        return ChatGoogleGenerativeAI(model="gemini-2.0-pro", temperature=0)
    if provider == "groq":
        return ChatGroq(model="llama3-70b-8192", temperature=0)
    if provider == "huggingface":
        return ChatHuggingFace(
            llm=HuggingFaceEndpoint(
                url="/static-proxy?url=https%3A%2F%2Fapi-inference.huggingface.co%2Fmodels%2Fmicrosoft%2FPhi-3-mini-4k-instruct%26quot%3B%3C%2Fspan%3E%2C
                temperature=0
            )
        )
    raise ValueError(f"Unknown provider: {provider}")

# --------------------------------------------------------------------------------------
# 5. 加载系统提示
# --------------------------------------------------------------------------------------
with open("system_prompt.txt", encoding="utf-8") as fp:
    SYSTEM_PROMPT = SystemMessage(content=fp.read())

# --------------------------------------------------------------------------------------
# 6. 构建 LangGraph
# --------------------------------------------------------------------------------------
def build_graph(provider: str = "groq"):
    llm = make_llm(provider).bind_tools(TOOLS)

    def assistant(state: MessagesState):
        return {"messages": [llm.invoke(state["messages"])]}

    builder = StateGraph(MessagesState)

    # 节点
    builder.add_node("cache", cache_lookup)
    builder.add_node("retriever", similar_q)
    builder.add_node("assistant", assistant)
    builder.add_node("toolcaller", ToolNode(TOOLS))

    # 边
    builder.add_edge(START, "cache")
    builder.add_edge("cache", "retriever")
    builder.add_edge("retriever", "assistant")

    # 工具调用与结束判断
    builder.add_conditional_edges(
        "assistant",
        tools_condition,
        {
            "toolcaller": "toolcaller",
            END: END
        }
    )
    builder.add_edge("toolcaller", "assistant")

    return builder.compile()

# --------------------------------------------------------------------------------------
# 7. CLI 用法示例
# --------------------------------------------------------------------------------------
if __name__ == "__main__":
    import sys
    question = " ".join(sys.argv[1:]) or "Who won the Nobel physics prize in 1921?"
    graph = build_graph("groq")
    init_msgs = [SYSTEM_PROMPT, HumanMessage(content=question)]
    result = graph.invoke({"messages": init_msgs})
    for msg in result["messages"]:
        msg.pretty_print()
    # 写入缓存
    _CACHE[question] = result["messages"][-1]