langraph_agent.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_google_genai import ChatGoogleGenerativeAI
from langchain_groq import ChatGroq
from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint, HuggingFaceEmbeddings
from langchain_community.tools.tavily_search import TavilySearchResults
from langchain_community.document_loaders import WikipediaLoader
from langchain_community.document_loaders import 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
import requests  # NEW: for HTTP requests to scoring API
from dataclasses import dataclass
import time  # For timestamp in code agent wrapper
from code_agent import run_agent  # Compiled code-interpreter graph helper
from langchain_core.messages import AIMessage

from langfuse.langchain import CallbackHandler

# Initialize Langfuse CallbackHandler for LangGraph/Langchain (tracing)
try:
    langfuse_handler = CallbackHandler()
except Exception as e:
    print(f"Warning: Could not initialize Langfuse handler: {e}")
    langfuse_handler = None

# Load environment variables - try multiple files
load_dotenv()  # Try .env first
load_dotenv("env.local")  # Try env.local as backup

print(f"SUPABASE_URL loaded: {bool(os.environ.get('SUPABASE_URL'))}")
print(f"GROQ_API_KEY loaded: {bool(os.environ.get('GROQ_API_KEY'))}")

# Base URL of the scoring API (duplicated here to avoid circular import with basic_agent)
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"

@tool
def wiki_search(input: str) -> str:
    """Search Wikipedia for a query and return maximum 2 results.
    
    Args:
        input: The search query."""
    try:
        search_docs = WikipediaLoader(query=input, load_max_docs=2).load()
        if not search_docs:
            return {"wiki_results": "No Wikipedia results found for the query."}
        formatted_search_docs = "\n\n---\n\n".join(
            [
                f'<Document source="{doc.metadata.get("source", "Unknown")}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content}\n</Document>'
                for doc in search_docs
            ])
        return {"wiki_results": formatted_search_docs}
    except Exception as e:
        print(f"Error in wiki_search: {e}")
        return {"wiki_results": f"Error searching Wikipedia: {e}"}

@tool
def web_search(input: str) -> str:
    """Search Tavily for a query and return maximum 3 results.
    
    Args:
        input: The search query."""
    try:
        # TavilySearchResults.invoke expects the query string as its first ("input") argument.
        # Passing it positionally avoids the earlier `missing 1 required positional argument: 'input'` error.
        search_docs = TavilySearchResults(max_results=3).invoke(input)
        if not search_docs:
            return {"web_results": "No web search results found for the query."}
        formatted_search_docs = "\n\n---\n\n".join(
            [
                f'<Document source="{doc.get("url", "Unknown")}" />\n{doc.get("content", "No content")}\n</Document>'
                for doc in search_docs
            ])
        return {"web_results": formatted_search_docs}
    except Exception as e:
        print(f"Error in web_search: {e}")
        return {"web_results": f"Error searching web: {e}"}

@tool
def arvix_search(input: str) -> str:
    """Search Arxiv for a query and return maximum 3 result.
    
    Args:
        input: The search query."""
    try:
        search_docs = ArxivLoader(query=input, load_max_docs=3).load()
        if not search_docs:
            return {"arvix_results": "No Arxiv results found for the query."}
        formatted_search_docs = "\n\n---\n\n".join(
            [
                f'<Document source="{doc.metadata.get("source", "Unknown")}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content[:1000]}\n</Document>'
                for doc in search_docs
            ])
        return {"arvix_results": formatted_search_docs}
    except Exception as e:
        print(f"Error in arvix_search: {e}")
        return {"arvix_results": f"Error searching Arxiv: {e}"}

@tool
def run_python(input: str) -> str:
    """Execute Python code in a restricted sandbox (code-interpreter).

    Pass **any** coding or file-manipulation task here and the agent will
    compute the answer by running Python. The entire standard library is NOT
    available; heavy networking is disabled. Suitable for: math, data-frames,
    small file parsing, algorithmic questions.
    """
    return run_agent(input)

# load the system prompt from the file
with open("system_prompt.txt", "r", encoding="utf-8") as f:
    system_prompt = f.read()

# System message
sys_msg = SystemMessage(content=system_prompt)

# build a retriever
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2") #  dim=768

# Try to create Supabase client with error handling
try:
    supabase_url = os.environ.get("SUPABASE_URL")
    supabase_key = os.environ.get("SUPABASE_SERVICE_KEY")
    
    if not supabase_url or not supabase_key:
        print("Warning: Supabase credentials not found, vector store will be disabled")
        vector_store = None
        create_retriever_tool = None
    else:
        supabase: Client = create_client(supabase_url, supabase_key)
        vector_store = SupabaseVectorStore(
            client=supabase,
            embedding= embeddings,
            table_name="documents",
            query_name="match_documents_langchain",
        )
        create_retriever_tool = create_retriever_tool(
            retriever=vector_store.as_retriever(),
            name="Question Search",
            description="A tool to retrieve similar questions from a vector store.",
        )
except Exception as e:
    print(f"Warning: Could not initialize Supabase vector store: {e}")
    vector_store = None
    create_retriever_tool = None

tools = [
    wiki_search,
    web_search,
    arvix_search,
    run_python,
]
if create_retriever_tool:
    tools.append(create_retriever_tool)

# ---------------------------------------------------------------------------
# Code-interpreter integration helpers
# ---------------------------------------------------------------------------

from code_agent import run_agent  # Executes the compiled code-interpreter graph


def _needs_code(state: dict) -> bool:  # type: ignore[override]
    """Heuristic: does *state* look like a coding request?"""
    messages = state.get("messages", [])
    if not messages:
        return False
    last_content = messages[-1].content.lower()
    triggers = [
        "```python",
        "write python",
        "run this code",
        "file manipulation",
        "csv",
        "pandas",
        "json",
        "plot",
        "fibonacci",
    ]
    return any(t in last_content for t in triggers)


def _code_exec_wrapper(state: dict):  # type: ignore[override]
    """Delegate the user query to the sandboxed Python interpreter."""
    # Get the last human message's content (fallback to empty string)
    human_msgs = [m.content for m in state.get("messages", []) if m.type == "human"]
    query = "\n\n".join(human_msgs)

    # Execute code-interpreter with full context (question + attachments)
    result = run_agent(query)
    # Persist the raw stdout so we can convert it to an AI message downstream
    return {"code_result": result}


def _code_to_message(state: dict):  # type: ignore[override]
    """Turn the interpreter's stdout into an AIMessage so the LLM can see it."""
    from langchain_core.messages import AIMessage  # local import to avoid cycles

    if not state.get("code_result"):
        return {}
    return {"messages": [AIMessage(content=state["code_result"])]}

# Build graph function
def build_graph(provider: str = "groq"):
    """Build the graph"""
    # Load environment variables from .env file
    if provider == "google":
        # Google Gemini
        llm = ChatGoogleGenerativeAI(model="gemini-2.0-flash", temperature=0)
    elif provider == "groq":
        # Groq https://console.groq.com/docs/models
        llm = ChatGroq(model="qwen-qwq-32b", temperature= 0.6) # optional : qwen-qwq-32b gemma2-9b-it
    elif provider == "huggingface":
        # TODO: Add huggingface endpoint
        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 'google', 'groq' or 'huggingface'.")
    # Bind tools to LLM
    llm_with_tools = llm.bind_tools(tools)

    # Node
    def assistant(state: MessagesState):
        """Assistant node"""
        try:
            print(f"Assistant node: Processing {len(state['messages'])} messages")
            result = llm_with_tools.invoke(state["messages"])
            print(f"Assistant node: LLM returned result type: {type(result)}")
            return {"messages": [result]}
        except Exception as e:
            print(f"Error in assistant node: {e}")
            error_msg = AIMessage(content=f"I encountered an error: {e}")
            return {"messages": [error_msg]}
    
    def retriever(state: MessagesState):
        """Retriever node"""
        try:
            print(f"Retriever node: Processing {len(state['messages'])} messages")
            if not state["messages"]:
                print("Retriever node: No messages in state")
                return {"messages": [sys_msg]}

            # Extract the user query content early for downstream steps
            query_content = state["messages"][0].content

            # ------------------- NEW: fetch attachment if available -------------------
            attachment_msg = None
            try:
                resp = requests.get(f"{DEFAULT_API_URL}/questions", timeout=30)
                resp.raise_for_status()
                questions = resp.json()
                matched_task_id = None
                for q in questions:
                    if str(q.get("question")).strip() == str(query_content).strip():
                        matched_task_id = str(q.get("task_id"))
                        break
                if matched_task_id:
                    print(f"Retriever node: Found task_id {matched_task_id} for current question, attempting to download attachment…")
                    file_resp = requests.get(f"{DEFAULT_API_URL}/files/{matched_task_id}", timeout=60)
                    if file_resp.status_code == 200 and file_resp.content:
                        try:
                            file_text = file_resp.content.decode("utf-8", errors="replace")
                        except Exception:
                            file_text = "(binary or non-UTF8 file omitted)"
                        MAX_CHARS = 8000
                        if len(file_text) > MAX_CHARS:
                            print(f"Retriever node: Attachment length {len(file_text)} > {MAX_CHARS}, truncating…")
                            file_text = file_text[:MAX_CHARS] + "\n… (truncated)"
                        attachment_msg = HumanMessage(content=f"Attached file content for task {matched_task_id}:\n```python\n{file_text}\n```")
                        print("Retriever node: Prepared attachment message")
                    else:
                        print(f"Retriever node: No attachment found for task {matched_task_id} (status {file_resp.status_code})")
            except Exception as api_e:
                print(f"Retriever node: Error while fetching attachment – {api_e}")
            # -------------------------------------------------------------------------

            # If vector store unavailable, simply return sys_msg + user message (+ attachment if any)
            if not vector_store:
                msgs = [sys_msg] + state["messages"]
                if attachment_msg:
                    msgs.append(attachment_msg)
                print("Retriever node: Vector store not available, skipping retrieval")
                return {"messages": msgs}

            # Perform similarity search when vector store is available
            print(f"Retriever node: Searching for similar questions with query: {query_content[:100]}…")
            similar_question = vector_store.similarity_search(query_content)
            print(f"Retriever node: Found {len(similar_question)} similar questions")
            msgs = [sys_msg] + state["messages"]
            if similar_question:
                example_msg = HumanMessage(content=f"Here I provide a similar question and answer for reference: \n\n{similar_question[0].page_content}")
                msgs.append(example_msg)
                print("Retriever node: Added example message from similar question")
            else:
                print("Retriever node: No similar questions found, proceeding without example")

            # Attach the file content if we have it
            if attachment_msg:
                msgs.append(attachment_msg)
                print("Retriever node: Added attachment content to messages")

            return {"messages": msgs}
        except Exception as e:
            print(f"Error in retriever node: {e}")
            return {"messages": [sys_msg] + state["messages"]}

    builder = StateGraph(MessagesState)
    builder.add_node("retriever", retriever)
    builder.add_node("assistant", assistant)
    builder.add_node("tools", ToolNode(tools))
    builder.add_node("code_exec", _code_exec_wrapper)
    builder.add_node("code_to_message", _code_to_message)

    builder.add_edge(START, "retriever")
    # Conditional branch: decide whether to run code interpreter
    builder.add_conditional_edges(
        "retriever",
        _needs_code,
        {True: "code_exec", False: "assistant"},
    )

    # Flow after code execution: inject result then resume chat
    builder.add_edge("code_exec", "code_to_message")
    builder.add_edge("code_to_message", "assistant")

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

    # Compile graph
    return builder.compile()

# test
if __name__ == "__main__":
    question = "When was a picture of St. Thomas Aquinas first added to the Wikipedia page on the Principle of double effect?"
    # Build the graph
    graph = build_graph(provider="groq")
    # Run the graph
    messages = [HumanMessage(content=question)]
    messages = graph.invoke({"messages": messages}, config={"callbacks": [langfuse_handler]})
    for m in messages["messages"]:
        m.pretty_print()