app.py

from smolagents import CodeAgent, HfApiModel, load_tool, tool
from dotenv import load_dotenv
import datetime
import requests
import base64
from requests import post, get
import os
import pytz
import yaml
import json
from urllib.parse import urlencode
import typing
from typing import List, Dict, Optional
from tools.final_answer import FinalAnswerTool
from tools.web_search import DuckDuckGoSearchTool
from Gradio_UI import GradioUI

Any = typing.Any

@tool
def search_youtube_guitar_tutorials(song_name: str) -> List[Dict[str, Any]]:
    """Searches YouTube for guitar tutorials for a specific song.
    
    Args:
        song_name: The name of the song to search for guitar tutorials
        
    Returns:
        A list of dictionaries containing information about the tutorials found,
        including title, channel name, view count, and URL.
    """
    api_key = os.getenv("YOUTUBE_API_KEY")
    if not api_key:
        raise ValueError("YouTube API key not found. Please set the YOUTUBE_API_KEY environment variable.")
    
    base_url = "https://www.googleapis.com/youtube/v3/search"
    search_params = {
        'part': 'snippet',
        'q': f"{song_name} guitar tutorial how to play",
        'type': 'video',
        'videoDefinition': 'high',
        'order': 'relevance',
        'maxResults': 10,
        'key': api_key
    }
    search_url = f"{base_url}?{urlencode(search_params)}"
    
    try:
        search_response = requests.get(search_url)
        search_response.raise_for_status()
        search_data = search_response.json()
        
        # Get video IDs
        video_ids = [item['id']['videoId'] for item in search_data.get('items', [])]
        if not video_ids:
            return []
        
        # Get video details (including view counts)
        video_url = "https://www.googleapis.com/youtube/v3/videos"
        video_params = {
            'part': 'snippet,statistics',
            'id': ','.join(video_ids),
            'key': api_key
        }
        video_details_url = f"{video_url}?{urlencode(video_params)}"
        video_response = requests.get(video_details_url)
        video_response.raise_for_status()
        video_data = video_response.json()

        # Filter for actual guitar tutorials
        tutorial_keywords = ['tutorial', 'lesson', 'how to play', 'guitar cover', 'chords', 'tabs']

        filtered_items = []
        for item in video_data.get('items', []):
            title_lower = item['snippet']['title'].lower()
            if any(keyword in title_lower for keyword in tutorial_keywords):
                filtered_items.append(item)

        # Sort by views if we have enough tutorials
        if len(filtered_items) >= 3:
            filtered_items.sort(key=lambda x: int(x['statistics'].get('viewCount', 0)), reverse=True)

        if not filtered_items:
            return "No guitar tutorials found."
            
        formatted_results = ["Here are the top guitar tutorials:"]
        for i, item in enumerate(filtered_items[:3], 1):
            views = int(item['statistics'].get('viewCount', 0))
            formatted_views = f"{views:,}" if views > 0 else "N/A"
            
            formatted_results.append(
                f"\n{i}. {item['snippet']['title']}\n"
                f"   Channel: {item['snippet']['channelTitle']}\n"
                f"   Views: {formatted_views}\n"
                f"   Link: https://www.youtube.com/watch?v={item['id']}\n"
            )
        
        return "\n".join(formatted_results)
    
    except requests.exceptions.RequestException as e:
        print(f"Error searching YouTube: {e}")
        return []
    
# Initialize tools
final_answer = FinalAnswerTool()

# Initialize model with alternative endpoint
model = HfApiModel(
    max_tokens=2096,
    temperature=0.5,
    model_id='Qwen/Qwen2.5-Coder-32B-Instruct',
    custom_role_conversions=None,
)

# Load prompts
with open("prompts.yaml", 'r') as stream:
    prompt_templates = yaml.safe_load(stream)

# Initialize agent
agent = CodeAgent(
    model=model,
    tools=[
        search_youtube_guitar_tutorials,
        final_answer  
    ],
    max_steps=6,
    verbosity_level=1,
    grammar=None,
    planning_interval=None,
    name=None,
    description=None,
    prompt_templates=prompt_templates
)

# Launch UI
GradioUI(agent).launch()