aggiunto tool get_youtube_transcript

agent.py

@@ -15,7 +15,7 @@ from langchain_google_genai import ChatGoogleGenerativeAI
 # Local imports
 from tools import get_search_tool, get_tavily_search_tool, get_wikipedia_tool, wikipedia_search, wikipedia_search_3,\
                   execute_python_code_from_file, download_taskid_file, analyze_excel_file, get_analyze_mp3_tool,\
-                  get_analyze_image_tool, arxiv_search
+                  get_analyze_image_tool, arxiv_search, get_youtube_transcript
 
 # Nota: per i test in locale si usa il .env
 #       su HuggingFace invece si usano le variabili definite in Settings/"Variables and secrets"
@@ -106,7 +106,15 @@ search_tool = get_tavily_search_tool()
 analyze_mp3_tool = get_analyze_mp3_tool(chat)
 analyze_png_tool = get_analyze_image_tool(chat)
 
-tools = [search_tool, wikipedia_search_3, execute_python_code_from_file, download_taskid_file, analyze_excel_file, analyze_mp3_tool, analyze_png_tool, arxiv_search]
+tools = [search_tool,
+         wikipedia_search_3,
+         execute_python_code_from_file,
+         download_taskid_file,
+         analyze_excel_file,
+         analyze_mp3_tool,
+         analyze_png_tool,
+         arxiv_search,
+         get_youtube_transcript]
 
 # Bind tools to the model
 chat_with_tools = chat.bind_tools(tools)

tools.py

@@ -11,6 +11,7 @@ from langchain_community.document_loaders import WikipediaLoader
 import wikipedia
 from langchain_tavily import TavilySearch
 from langchain_community.document_loaders import ArxivLoader
+from langchain.document_loaders import YouTubeLoader
 from langchain_core.tools import tool
 from langchain.tools import Tool
 from langchain_core.messages import HumanMessage
@@ -313,4 +314,17 @@ def arxiv_search(query: str) -> str:
             for doc in search_docs
         ]
     )
-    return {"arxiv_results": formatted_search_docs}
+    return {"arxiv_results": formatted_search_docs}
+
+
+@tool
+def get_youtube_transcript(url: str) -> str:
+    """
+    Fetches the transcript from a YouTube video URL and returns it as plain text.
+    """
+    loader = YouTubeLoader.from_youtube_url(url, add_video_info=True)
+    docs = loader.load()
+
+    # Combine all transcript chunks into a single string
+    transcript = "\n".join(doc.page_content for doc in docs)
+    return transcript