get_youtube_transcript + use o4-mini

langgraph_dir/agent.py

@@ -11,7 +11,8 @@ from langchain_community.tools import BraveSearch
 
 from .prompt import system_prompt
 from .custom_tools import (multiply, add, subtract, divide, modulus, power,
-    query_image, automatic_speech_recognition, get_webpage_content, python_repl_tool)
+    query_image, automatic_speech_recognition, get_webpage_content, python_repl_tool,
+    get_youtube_transcript)
 
 
 class LangGraphAgent:
@@ -21,7 +22,11 @@ class LangGraphAgent:
                  show_prompt=True):
 
         # =========== LLM definition ===========
-        llm = ChatOpenAI(model=model_name, temperature=0) # needs OPENAI_API_KEY in env
+        if model_name.startswith('o'):
+            # reasoning model (no temperature setting)
+            llm = ChatOpenAI(model=model_name) # needs OPENAI_API_KEY in env
+        else:
+            llm = ChatOpenAI(model=model_name, temperature=0)
         print(f"LangGraphAgent initialized with model \"{model_name}\"")
 
         # =========== Augment the LLM with tools ===========
@@ -36,6 +41,7 @@ class LangGraphAgent:
             automatic_speech_recognition, # Transcribe an audio file to text
             get_webpage_content, # Load a web page and return it to markdown
             python_repl_tool, # Python code interpreter
+            get_youtube_transcript, # Get the transcript of a YouTube video
         ]
 
         tools = community_tools + custom_tools

langgraph_dir/config.py

@@ -1,6 +1,4 @@
-# OPENAI_MODEL_NAME = "gpt-4.1-nano" # Overall Score: 10.0% (2/20 correct)
-OPENAI_MODEL_NAME = "gpt-4.1-mini"
+# OPENAI_MODEL_NAME = "gpt-4.1-nano"
+# OPENAI_MODEL_NAME = "gpt-4.1-mini"
 # OPENAI_MODEL_NAME = "gpt-4.1"
-
-# QUERY_IMAGE_MODEL_NAME = "gpt-4.1-mini"
-QUERY_IMAGE_MODEL_NAME = "o4-mini"
+OPENAI_MODEL_NAME = "o4-mini"

langgraph_dir/custom_tools.py

@@ -8,6 +8,8 @@ from langchain_core.tools import tool, Tool
 from langchain_experimental.utilities import PythonREPL
 from pypdf import PdfReader
 from io import BytesIO
+from youtube_transcript_api import YouTubeTranscriptApi
+from pytube import extract
 
 
 # --- Basic operations --- #
@@ -241,4 +243,25 @@ You are allowed to download files from URLs.
 Do NOT send commands that block indefinitely (e.g., `input()`).""",
     func=python_repl.run,
     args_schema=PythonREPLInput
-)
+)
+
+@tool
+def get_youtube_transcript(page_url: str) -> str:
+    """Get the transcript of a YouTube video
+
+    Args:
+        page_url (str): YouTube URL of the video
+    """
+    try:
+        # get video ID from URL
+        video_id = extract.video_id(page_url)
+
+        # get transcript
+        ytt_api = YouTubeTranscriptApi()
+        transcript = ytt_api.fetch(video_id)
+
+        # keep only text
+        txt = '\n'.join([s.text for s in transcript.snippets])
+        return txt
+    except Exception as e:
+        return f"get_youtube_transcript failed: {e}"

requirements.txt

@@ -12,4 +12,7 @@ langchain-community
 duckduckgo-search
 markdownify
 beautifulsoup4
-langchain_experimental
+langchain_experimental
+pypdf
+youtube-transcript-api
+pytube