agent.py

"""LangGraph Agent – versione GPT-4.1 / Hugging Face Spaces"""
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

# LLM providers
from langchain_openai import ChatOpenAI            # NEW (GPT-4.1)
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_groq import ChatGroq
from langchain_huggingface import (
    ChatHuggingFace,
    HuggingFaceEndpoint,
    HuggingFaceEmbeddings,
)

# Tools & loaders
from langchain_community.tools.tavily_search import TavilySearchResults
from langchain_community.document_loaders import WikipediaLoader, ArxivLoader
from langchain_community.vectorstores import SupabaseVectorStore
from langchain_core.messages import SystemMessage, HumanMessage
from langchain_core.tools import tool
from langchain.tools.retriever import create_retriever_tool
from supabase.client import Client, create_client

# --------------------------------------------------------------------------- #
#  Carica variabili d'ambiente (.env eventuale + secrets di HF Spaces)        #
# --------------------------------------------------------------------------- #
load_dotenv()  # nei Spaces le secrets sono già in os.environ

# --------------------------------------------------------------------------- #
#  TOOL di esempio (aritmetica)                                               #
# --------------------------------------------------------------------------- #
@tool
def multiply(a: int, b: int) -> int: return a * b

@tool
def add(a: int, b: int) -> int: return a + b

@tool
def subtract(a: int, b: int) -> int: return a - b

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

@tool
def modulus(a: int, b: int) -> int: return a % b

# --------------------------------------------------------------------------- #
#  TOOL: Wikipedia                                                             #
# --------------------------------------------------------------------------- #
@tool
def wiki_search(query: str) -> str:
    """Search Wikipedia (max 2 docs) and return formatted result."""
    docs = WikipediaLoader(query=query, load_max_docs=2).load()
    return "\n\n---\n\n".join(
        f'<Document source="{d.metadata["source"]}" page="{d.metadata.get("page","")}"/>\n'
        f"{d.page_content}\n</Document>"
        for d in docs
    )

# --------------------------------------------------------------------------- #
#  TOOL: Tavily web search                                                     #
# --------------------------------------------------------------------------- #
@tool
def web_search(query: str) -> str:
    """Search Tavily (max 3 docs) and return formatted result."""
    docs = TavilySearchResults(max_results=3).invoke(query=query)
    return "\n\n---\n\n".join(
        f'<Document source="{d.metadata["source"]}" page="{d.metadata.get("page","")}"/>\n'
        f"{d.page_content}\n</Document>"
        for d in docs
    )

# --------------------------------------------------------------------------- #
#  TOOL: ArXiv                                                                 #
# --------------------------------------------------------------------------- #
@tool
def arxiv_search(query: str) -> str:
    """Search ArXiv (max 3 docs) and return formatted snippet."""
    docs = ArxivLoader(query=query, load_max_docs=3).load()
    return "\n\n---\n\n".join(
        f'<Document source="{d.metadata["source"]}" page="{d.metadata.get("page","")}"/>\n'
        f"{d.page_content[:1000]}\n</Document>"
        for d in docs
    )

# --------------------------------------------------------------------------- #
#  System prompt                                                               #
# --------------------------------------------------------------------------- #
with open("system_prompt.txt", "r", encoding="utf-8") as f:
    system_prompt = f.read()
sys_msg = SystemMessage(content=system_prompt)

# --------------------------------------------------------------------------- #
#  Vector store per il retriever                                               #
# --------------------------------------------------------------------------- #
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
supabase: Client = create_client(
    os.environ.get("SUPABASE_URL"),
    os.environ.get("SUPABASE_SERVICE_KEY"),
)
vector_store = SupabaseVectorStore(
    client=supabase,
    embedding=embeddings,
    table_name="documents",
    query_name="match_documents_langchain",
)
question_search_tool = create_retriever_tool(
    retriever=vector_store.as_retriever(),
    name="Question Search",
    description="A tool to retrieve similar questions from a vector store.",
)

# --------------------------------------------------------------------------- #
#  Registrazione tool list                                                    #
# --------------------------------------------------------------------------- #
tools = [
    multiply,
    add,
    subtract,
    divide,
    modulus,
    wiki_search,
    web_search,
    arxiv_search,
    question_search_tool,
]

# --------------------------------------------------------------------------- #
#  Costruzione del graph LangGraph                                             #
# --------------------------------------------------------------------------- #
def build_graph(provider: str = "openai"):
    """Restituisce un graph LangGraph pronto all'uso.

    provider: "openai" (default), "google", "groq", "huggingface"
    """
    # --- Selezione LLM ------------------------------------------------------ #
    if provider == "openai":
        openai_key = os.getenv("OPENAI_KEY")
        if not openai_key:
            raise ValueError(
                "❌ Environment variable OPENAI_KEY mancante. "
                "Aggiungi la secret dal tab 'Secrets' dello Space."
            )
        llm = ChatOpenAI(
            model_name="gpt-4.1",
            temperature=0,
            openai_api_key=openai_key,
        )

    elif provider == "google":
        llm = ChatGoogleGenerativeAI(model="gemini-2.0-flash", temperature=0)

    elif provider == "groq":
        llm = ChatGroq(model="qwen-qwq-32b", temperature=0)

    elif provider == "huggingface":
        llm = ChatHuggingFace(
            llm=HuggingFaceEndpoint(
                url="https://api-inference.huggingface.co/models/Meta-DeepLearning/llama-2-7b-chat-hf",
                temperature=0,
            )
        )
    else:
        raise ValueError(
            "Invalid provider. Choose 'openai', 'google', 'groq' or 'huggingface'."
        )

    # Abilita tool calling
    llm_with_tools = llm.bind_tools(tools)

    # ------------------------- NODES --------------------------------------- #
    def assistant(state: MessagesState):
        """Invoca il modello."""
        return {"messages": [llm_with_tools.invoke(state["messages"])]}

    def retriever(state: MessagesState):
        """Aggiunge alla history un Q/A simile come esempio."""
        similar = vector_store.similarity_search(state["messages"][0].content)
        if similar:
            example_msg = HumanMessage(
                content=(
                    "Here I provide a similar question and answer for reference:\n\n"
                    f"{similar[0].page_content}"
                )
            )
            return {"messages": [sys_msg] + state["messages"] + [example_msg]}
        return {"messages": [sys_msg] + state["messages"]}

    # --------------------------- GRAPH ------------------------------------- #
    builder = StateGraph(MessagesState)
    builder.add_node("retriever", retriever)
    builder.add_node("assistant", assistant)
    builder.add_node("tools", ToolNode(tools))

    builder.add_edge(START, "retriever")
    builder.add_edge("retriever", "assistant")
    builder.add_conditional_edges("assistant", tools_condition)
    builder.add_edge("tools", "assistant")

    return builder.compile()


# --------------------------------------------------------------------------- #
#  Quick test (python agent.py)                                               #
# --------------------------------------------------------------------------- #
if __name__ == "__main__":
    graph = build_graph(provider="openai")
    question = "When was a picture of St. Thomas Aquinas first added to the Wikipedia page on the Principle of double effect?"
    msgs = [HumanMessage(content=question)]
    result = graph.invoke({"messages": msgs})
    for m in result["messages"]:
        m.pretty_print()