tools.py

from smolagents import DuckDuckGoSearchTool, Tool, VisitWebpageTool
from langchain_community.agent_toolkits.load_tools import load_tools
import os
from typing import List, Optional
from youtube_transcript_api import YouTubeTranscriptApi, TranscriptsDisabled, NoTranscriptFound
from smolagents import tool

@tool
def youtube_transcript_search(
    video_id: str,
    query: str,
    top_k: Optional[int] = 5
) -> List[dict]:
    """
    Search a YouTube transcript for occurrences of a query string.

    Args:
        video_id (str): The YouTube video ID (after 'v=' in the URL).
        query (str): The search term to look for in the transcript (case-insensitive).
        top_k (Optional[int]): Maximum number of matches to return (default: 5).

    Returns:
        List[dict]: A list of up to top_k matches, each a dict with:
            - 'time': float start time in seconds
            - 'text_snippet': snippet of transcript containing the query
    """
    try:
        # Fetch transcript (choose auto-generated or manually created)
        transcript_list = YouTubeTranscriptApi.list_transcripts(video_id)
        # prefer generated if manual unavailable
        transcript = transcript_list.find_manually_created_transcript(['en']) or transcript_list.find_generated_transcript(['en'])
        segments = transcript.fetch()

    except (TranscriptsDisabled, NoTranscriptFound) as e:
        return [{"time": 0.0, "text_snippet": f"No transcript found: {str(e)}"}]

    # Search query in segments
    query_lower = query.lower()
    hits = []
    for seg in segments:
        if query_lower in seg["text"].lower():
            hits.append({
                "time": seg["start"],
                "text_snippet": seg["text"]
            })
            if len(hits) >= top_k:
                break

    if not hits:
        return [{"time": 0.0, "text_snippet": "No matches found."}]

    return hits

# Initialize the search tools

duck_search_tool = DuckDuckGoSearchTool()

visit_web_page_tool = VisitWebpageTool()

google_search_tool = Tool.from_langchain(load_tools(["serpapi"])[0])