app.py

from smolagents import CodeAgent, DuckDuckGoSearchTool, InferenceClientModel, load_tool, tool
import datetime
import requests
import pytz
import yaml
import gradio as gr
from tools.final_answer import FinalAnswerTool

# Below is an example of a tool that does nothing. Amaze us with your creativity !
@tool
def my_custom_tool(arg1: str, arg2: int) -> str:  # it's important to specify the return type
    # Keep this format for the description / args / args description but feel free to modify the tool
    """Fetch and nicely format top headlines from popular news sources for today.

    Args:
        arg1: Comma-separated list of sources to include (options: "bbc", "nyt", "guardian", "hn", "all").
              Use "all" or empty string to fetch from all supported sources.
        arg2: Maximum number of headlines per source (must be > 0).

    This tool scrapes public RSS feeds (no API key needed) and returns a
    markdown-formatted string grouped by source, with each headline on its own line
    and linked to the original article when available.
    """
    import xml.etree.ElementTree as ET

    if arg2 <= 0:
        return "Please provide a positive integer number of headlines per source for arg2."

    # Normalize requested sources
    requested = [s.strip().lower() for s in arg1.split(",")] if arg1 else []
    if not requested or "all" in requested:
        requested = ["bbc", "nyt", "guardian", "hn"]

    # Supported sources and their RSS URLs
    sources = {
        "bbc": {
            "name": "BBC News",
            "url": "https://feeds.bbci.co.uk/news/rss.xml",
        },
        "nyt": {
            "name": "The New York Times",
            "url": "https://rss.nytimes.com/services/xml/rss/nyt/HomePage.xml",
        },
        "guardian": {
            "name": "The Guardian",
            "url": "https://www.theguardian.com/world/rss",
        },
        "hn": {
            "name": "Hacker News (Top)",
            "url": "https://hnrss.org/frontpage",
        },
    }

    picked_keys = [k for k in requested if k in sources]
    if not picked_keys:
        return (
            "No valid news sources selected. Valid options: bbc, nyt, guardian, hn, all."
        )

    output_sections = []

    for key in picked_keys:
        meta = sources[key]
        name = meta["name"]
        url = meta["url"]

        try:
            resp = requests.get(url, timeout=8)
            resp.raise_for_status()
            root = ET.fromstring(resp.content)

            # RSS structure: channel/item/title/link
            items = []
            for item in root.findall(".//item"):
                title_el = item.find("title")
                link_el = item.find("link")
                if title_el is not None and title_el.text:
                    title = title_el.text.strip()
                    link = link_el.text.strip() if link_el is not None and link_el.text else ""
                    if title:
                        items.append((title, link))
                if len(items) >= arg2:
                    break

            if not items:
                output_sections.append(f"### {name}\n\n_(no headlines found)_")
            else:
                lines = []
                for idx, (title, link) in enumerate(items, start=1):
                    if link:
                        lines.append(f"{idx}. [{title}]({link})")
                    else:
                        lines.append(f"{idx}. {title}")
                joined = "\n".join(lines)
                output_sections.append(f"### {name} (top {len(items)})\n\n{joined}")
        except Exception as e:
            output_sections.append(f"### {name}\n\n_Error fetching headlines: {e}_")

    header = "## Top headlines from popular news sources\n"
    return header + "\n\n" + "\n\n".join(output_sections)

@tool
def get_current_time_in_timezone(timezone: str) -> str:
    """A tool that fetches the current local time in a specified timezone.
    Args:
        timezone: A string representing a valid timezone (e.g., 'America/New_York').
    """
    try:
        # Create timezone object
        tz = pytz.timezone(timezone)
        # Get current time in that timezone
        local_time = datetime.datetime.now(tz).strftime("%Y-%m-%d %H:%M:%S")
        return f"The current local time in {timezone} is: {local_time}"
    except Exception as e:
        return f"Error fetching time for timezone '{timezone}': {str(e)}"


final_answer = FinalAnswerTool()

# If the agent does not answer, the model is overloaded, please use another model or the following Hugging Face Endpoint that also contains qwen2.5 coder:
# model_id='https://pflgm2locj2t89co.us-east-1.aws.endpoints.huggingface.cloud' 

model = InferenceClientModel(
    max_tokens=2096,
    temperature=0.5,
    model_id='Qwen/Qwen2.5-Coder-32B-Instruct',  # it is possible that this model may be overloaded
    custom_role_conversions=None,
)


# Import tool from Hub
image_generation_tool = load_tool("agents-course/text-to-image", trust_remote_code=True)

with open("prompts.yaml", 'r') as stream:
    prompt_templates = yaml.safe_load(stream)
    
agent = CodeAgent(
    model=model,
    tools=[final_answer, my_custom_tool, get_current_time_in_timezone, image_generation_tool],  # don't remove final_answer
    max_steps=6,
    verbosity_level=1,
    name=None,
    description=None,
)


def chat_fn(message, history):
    """Chat handler for Gradio.

    If the user is asking for headlines, we bypass the agent and call the
    headlines tool directly so the nicely formatted markdown is returned
    as-is. For other queries, we fall back to the CodeAgent.
    """
    text = (message or "").lower().strip()

    # Special handling for capability / "what can you do" questions
    if any(phrase in text for phrase in ["what can you do", "what you can do", "help me", "how can you help"]):
        return (
            "I’m your **Daily Headlines Assistant**.\n\n"
            "- I fetch today’s top headlines from **BBC**, **The New York Times**, "
            "**The Guardian**, and **Hacker News** using live RSS feeds.\n"
            "- I format them into a clean, grouped markdown view with **clickable links**.\n"
            "- I can also tell you the **current time in any timezone** and generate images from text prompts.\n\n"
            "Try asking things like:\n"
            "- \"Show me today’s top 3 headlines from BBC, NYT and Hacker News.\"\n"
            "- \"Give me the top 5 headlines from BBC only.\"\n"
            "- \"What time is it now in America/New_York?\"\n"
        )
    wants_news = any(
        kw in text
        for kw in ["headline", "headlines", "bbc", "nyt", "new york times", "guardian", "hacker news", "hn"]
    )

    if wants_news:
        # Simple heuristic: default to all sources and 3 headlines if user
        # does not specify numbers; otherwise, try to extract a small integer.
        import re

        match = re.search(r"\b(\d{1,2})\b", text)
        count = int(match.group(1)) if match else 3
        # Map some common names to our source keys
        sources = []
        if "bbc" in text:
            sources.append("bbc")
        if "nyt" in text or "new york times" in text:
            sources.append("nyt")
        if "guardian" in text:
            sources.append("guardian")
        if "hacker news" in text or "hn" in text:
            sources.append("hn")
        # If nothing specific mentioned, use all
        arg1 = ",".join(sources) if sources else "all"

        try:
            return my_custom_tool(arg1=arg1, arg2=count)
        except Exception as e:
            return f"Error while fetching headlines: {e}"

    # Fallback: use the full agent for non-news tasks
    try:
        result = agent.run(task=message)
    except Exception as e:
        result = f"Error while running the agent: {e}"
    if result is None:
        result = (
            "The agent did not produce a final answer (the upstream model may have "
            "returned an error like 502). Please try again or with a simpler request."
        )
    return str(result)


demo = gr.ChatInterface(
    fn=chat_fn,
    title="Daily Headlines Assistant",
    description=(
        "Ask for today's top headlines from BBC, The New York Times, The Guardian, "
        "and Hacker News. Results include clickable links and are grouped by source."
    ),
    examples=[
        "Show me today’s top 3 headlines from BBC, NYT and Hacker News.",
        "Give me the top 5 headlines from BBC only.",
        "Fetch today’s main stories from The Guardian and Hacker News.",
    ],
)

demo.launch()