agent.py

from selenium import webdriver
from selenium.webdriver.common.by import By
from selenium.webdriver.common.keys import Keys
from langchain_core.tools import tool
from selenium.webdriver.support.ui import WebDriverWait
from selenium.webdriver.support import expected_conditions as EC
import time
from youtube_transcript_api import YouTubeTranscriptApi
import re
import yaml
from langchain_core.messages import SystemMessage
from langgraph.graph import StateGraph, START
from langgraph.prebuilt import ToolNode, tools_condition
from typing import TypedDict, Annotated
from langgraph.graph.message import add_messages
from langchain_openai import ChatOpenAI
from langchain_ollama import ChatOllama
from langchain_groq import ChatGroq
from dotenv import load_dotenv
import os
from tools import tools
from langchain_core.callbacks import BaseCallbackHandler
from langchain_core.tools import tool
from duckduckgo_search import DDGS
from selenium.webdriver.chrome.service import Service
from webdriver_manager.chrome import ChromeDriverManager


driver = None

def get_driver():
    global driver
    if driver is None:
        options = webdriver.ChromeOptions()
        options.add_argument("--no-sandbox")
        options.add_argument("--disable-dev-shm-usage")
        options.add_argument("--headless")  # 🔥 IMPORTANT

        driver = webdriver.Chrome(
            service=Service(ChromeDriverManager().install()),
            options=options
        )
    return driver

@tool
def search_in_page(keyword: str, context_chars: int = 400) -> str:
    """
    Search for a keyword in the current page and return all occurrences
    with surrounding context. Use this to find specific information.
    """
    d = get_driver()
    matches = d.execute_script("""
        const keyword = arguments[0];
        const ctx = arguments[1];
        const body = document.body.innerText;
        const lower = body.toLowerCase();
        const kw = keyword.toLowerCase();
        let matches = [], idx = 0;
        while ((idx = lower.indexOf(kw, idx)) !== -1) {
            matches.push({
                context: body.substring(Math.max(0, idx - ctx), 
                                        Math.min(body.length, idx + kw.length + ctx))
            });
            idx += kw.length;
            if (matches.length >= 10) break;  // cap at 10
        }
        return matches;
    """, keyword, context_chars)
    
    if not matches:
        return f"Keyword '{keyword}' not found on page."
    
    results = []
    for i, m in enumerate(matches, 1):
        results.append(f"[Match {i}]\n{m['context']}\n{'─'*40}")
    return f"Found {len(matches)} match(es) for '{keyword}':\n\n" + "\n".join(results)

    
@tool
def go_back() -> str:
    """Goes back to previous page."""
    driver = get_driver()
    driver.back()
    return "Went back to previous page"

@tool
def close_popups() -> str:
    """
    Closes any visible modal or pop-up on the page. Use this to dismiss pop-up windows!
    This does not work on cookie consent banners.
    """
    driver = get_driver()
    webdriver.ActionChains(driver).send_keys(Keys.ESCAPE).perform()
    return "Attempted to close popups"

@tool
def go_to(url: str) -> str:
    """Open a webpage using URL."""
    driver = get_driver()
    driver.get(url)
    WebDriverWait(driver, 10).until(
        lambda d: d.execute_script("return document.readyState") == "complete"
    )
    return f"Opened {url}"

@tool
def click_element(selector: str, by: str = "css") -> str:
    """
    Click an element. by = 'css', 'xpath', 'text', or 'id'.
    Example: click_element("a[href*='trending']")
    """
    d = get_driver()
    by_map = {
        "css": By.CSS_SELECTOR, "xpath": By.XPATH,
        "id": By.ID, "text": By.LINK_TEXT
    }
    locator = by_map.get(by, By.CSS_SELECTOR)
    el = WebDriverWait(d, 8).until(EC.element_to_be_clickable((locator, selector)))
    el.click()
    time.sleep(1)
    return f"Clicked. Now at: {d.current_url}"

@tool
def get_page_text(max_chars: int = 8000) -> str:
    """Get the full visible text of the current page (truncated)."""
    d = get_driver()
    text = d.execute_script("return document.body.innerText;")
    return text[:max_chars] if text else "No text found."

@tool
def get_elements_text(css_selector: str, limit: int = 10) -> str:
    """
    Extract text from multiple matching elements.
    Great for lists, tables, repo cards, profile stats, etc.
    Example: get_elements_text("article.Box-row", limit=5)
    """
    d = get_driver()
    elements = d.find_elements(By.CSS_SELECTOR, css_selector)[:limit]
    if not elements:
        return f"No elements found for selector: {css_selector}"
    results = [f"[{i+1}] {el.text.strip()}" for i, el in enumerate(elements) if el.text.strip()]
    return "\n\n".join(results)

@tool
def get_youtube_transcript(video_url: str = None) -> str:
    """Get transcript of current or given YouTube video."""

    driver = get_driver()
    url = video_url or driver.current_url
    
    match = re.search(r"(?:v=|youtu\.be/|embed/)([a-zA-Z0-9_-]{11})", url)
    if not match:
        return "FAILED: Could not extract video ID from URL"
    
    video_id = match.group(1)

    try:
        ytt_api = YouTubeTranscriptApi()  # instantiate first
        fetched = ytt_api.fetch(video_id)
        full_text = " ".join([snippet.text for snippet in fetched])
        return full_text[:5000]

    except Exception as e:
        return f"FAILED: {str(e)}"

        
@tool
def web_search(query: str) -> str:
    """
    Search the web for real-time information.

    Use this when:
    - You need up-to-date information
    - The answer is not on the current webpage
    - You need external knowledge

    Input: a clear search query
    Output: top search results with title, snippet, and URL
    """

    results = []

    try:
        with DDGS() as ddgs:
            for r in ddgs.text(query, max_results=5):
                results.append(
                    f"Title: {r['title']}\n"
                    f"Snippet: {r['body']}\n"
                    f"URL: {r['href']}\n"
                )

        if not results:
            return "No results found."

        return "\n\n".join(results)

    except Exception as e:
        return f"Search error: {str(e)}"

toolsused = [
    go_to,
    click_element,
    search_in_page,
    get_elements_text,
    get_page_text,
    go_back,
    close_popups,
    get_youtube_transcript,
    web_search   
]

load_dotenv()

# Load prompt
with open("config.yaml", encoding="utf-8") as f:
    SYSTEM_PROMPT = yaml.safe_load(f)["prompt"]

# LLM (local)
# llm = ChatOpenAI(
#     base_url="https://router.huggingface.co/v1",
#     api_key=os.getenv("HF_TOKEN"),
#     model="openai/gpt-oss-120b:groq",
# )

llm = ChatGroq(model="qwen/qwen3-32b", temperature=0)
print(type(tools))
llm_with_tools = llm.bind_tools(toolsused)

# State
class AgentState(TypedDict):
    messages: Annotated[list, add_messages]

# Assistant node
def assistant(state: AgentState):
    sys_msg = SystemMessage(content=SYSTEM_PROMPT)

    response = llm_with_tools.invoke(
        [sys_msg] + state["messages"]
    )

    return {"messages": [response]}

# Graph
builder = StateGraph(AgentState)

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

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

agent = builder.compile()

class ToolLogger(BaseCallbackHandler):
    def on_tool_start(self, serialized, input_str, **kwargs):
        print(f"\n🔧 TOOL CALLED: {serialized['name']}")
        print(f"   INPUT: {input_str}")
    
    def on_tool_end(self, output, **kwargs):
        print(f"   OUTPUT: {str(output)[:200]}")  # first 200 chars

def run_agent(question: str) -> str:
    response = agent.invoke(
        {
            "messages": [
                {"role": "user", "content": question}
            ]
        },
        config={
            "recursion_limit": 75,
            "callbacks": [ToolLogger()]
        }
    )
    print(response["messages"][-1].content)
    return response["messages"][-1].content