add tool to get webpage content + replace DDG by Brave search + replace buggy remote code interpreter by local runtime + avoid limit of TPM

.gitignore

@@ -1 +1,3 @@
-__pycache__
+__pycache__
+.DS_Store
+tmp*

app.py

@@ -2,6 +2,8 @@ import os
 import gradio as gr
 import requests
 import pandas as pd
+from time import sleep
+from tqdm import tqdm
 
 
 # (Keep Constants as is)
@@ -89,7 +91,7 @@ async def run_and_submit_all(profile: gr.OAuthProfile | None):
             if file_name:
                 # add the URL of the data source to the question (so that the agent can deal with it)
                 file_url = f"{DEFAULT_API_URL}/files/{task_id}"
-                question_text += f"\nFile URL: {file_url}"
+                question_text += f'\nFile URL: "{file_url}"'
                 # get the extension of the file to help the agent
                 try:
                     ext = file_name.split('.')[-1]
@@ -104,6 +106,11 @@ async def run_and_submit_all(profile: gr.OAuthProfile | None):
                 submitted_answer = agent(question_text)
             answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
+
+            # wait 1 minute before next call to avoid reaching limit of token per minute (TPM)
+            print('\n\n-> Sleeping for 1 minute to avoid reaching limit of token per minute (TPM)')
+            for _ in tqdm(range(60)): # tqdm to see time we have to wait
+                sleep(1)
         except Exception as e:
              print(f"Error running agent on task {task_id}: {e}")
              results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})

langgraph_dir/agent.py

@@ -1,3 +1,4 @@
+import os
 import json
 
 from typing import Literal
@@ -6,11 +7,11 @@ from langgraph.graph import MessagesState
 from langchain_core.messages import SystemMessage, HumanMessage, ToolMessage
 from langgraph.graph import StateGraph, START, END
 from langchain.agents import load_tools
-from langchain_community.tools.riza.command import ExecPython
+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)
+    query_image, automatic_speech_recognition, get_webpage_content, python_repl_tool)
 
 
 class LangGraphAgent:
@@ -20,20 +21,21 @@ class LangGraphAgent:
                  show_prompt=True):
 
         # =========== LLM definition ===========
-        llm = ChatOpenAI(model=model_name, temperature=0) # needs OPENAI_API_KEY
+        llm = ChatOpenAI(model=model_name, temperature=0) # needs OPENAI_API_KEY in env
         print(f"LangGraphAgent initialized with model \"{model_name}\"")
 
         # =========== Augment the LLM with tools ===========
-        community_tool_names = [
-            "ddg-search", # DuckDuckGo search
-            "wikipedia",
+        community_tools = [
+            BraveSearch.from_api_key(   # Web search (more performant than DuckDuckGo)
+                api_key=os.getenv("BRAVE_SEARCH_API_KEY"), # needs BRAVE_SEARCH_API_KEY in env
+                search_kwargs={"count": 3}),
         ]
-        community_tools = load_tools(community_tool_names)
-        community_tools += [ExecPython(runtime_revision_id='01JT97GJ20BC83Y75WMAS364ZT')] # Riza code interpreter (needs RIZA_API_KEY) (not supported by load_tools, custom runtime with basic packages (pandas, numpy, etc.))
         custom_tools = [
-            multiply, add, subtract, divide, modulus, power,  # basic arithmetic
+            multiply, add, subtract, divide, modulus, power,  # Basic arithmetic
             query_image, # Ask anything about an image using a VLM
             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
         ]
 
         tools = community_tools + custom_tools

langgraph_dir/config.py

@@ -1,3 +1,6 @@
 # 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"
+# OPENAI_MODEL_NAME = "gpt-4.1"
+
+# QUERY_IMAGE_MODEL_NAME = "gpt-4.1-mini"
+QUERY_IMAGE_MODEL_NAME = "o4-mini"

langgraph_dir/custom_tools.py

@@ -1,7 +1,11 @@
 import requests
-from langchain_core.tools import tool
+from pydantic import BaseModel, Field
 from huggingface_hub import InferenceClient
 from openai import OpenAI
+from bs4 import BeautifulSoup
+from markdownify import markdownify as md
+from langchain_core.tools import tool, Tool
+from langchain_experimental.utilities import PythonREPL
 
 
 # --- Basic operations --- #
@@ -116,10 +120,11 @@ def query_image(query: str, image_url: str) -> str:
             return completion.choices[0].message
 
         elif PROVIDER == 'openai':
-            client = OpenAI()
+            from .config import QUERY_IMAGE_MODEL_NAME
 
+            client = OpenAI()
             response = client.responses.create(
-                model="gpt-4.1-mini",
+                model=QUERY_IMAGE_MODEL_NAME,
                 input=[{
                     "role": "user",
                     "content": [
@@ -180,3 +185,46 @@ def automatic_speech_recognition(file_url: str, file_extension: str) -> str:
 
     except Exception as e:
         return f"automatic_speech_recognition failed: {e}"
+
+
+@tool
+def get_webpage_content(page_url: str) -> str:
+    """Load a web page and return it to markdown if possible
+
+    Args:
+        page_url (str): the URL of web page to get
+    """
+    try:
+        r = requests.get(page_url)
+        soup = BeautifulSoup((r.text), 'html.parser')
+        if soup.body:
+            # convert to markdown
+            out = md(str(soup.body))
+        else:
+            # return the raw content
+            out = r.text
+        return out
+    except Exception as e:
+        return f"get_webpage_content failed: {e}"
+
+
+# ======= Python code interpreter =======
+# WARNING: Python REPL can execute arbitrary code on the host machine (e.g., delete files, make network requests). Use with caution.
+
+class PythonREPLInput(BaseModel):
+    code: str = Field(description="The Python code string to execute.")
+
+python_repl = PythonREPL()
+
+python_repl_tool = Tool(
+    name="python_repl",
+    description="""A Python REPL shell (Read-Eval-Print Loop).
+Use this to execute single or multi-line python commands.
+Input should be syntactically valid Python code.
+Always end your code with `print(...)` to see the output.
+Do NOT execute code that could be harmful to the host system.
+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
+)

langgraph_dir/prompt.py

@@ -6,4 +6,5 @@ YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma sepa
 If you are asked for a number, don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise.
 If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities), and write the digits in plain text unless specified otherwise.
 If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number or a string.
+If you use the python_repl tool (code interpreter), always end your code with `print(...)` to see the output.
 """

requirements.txt

@@ -1,7 +1,7 @@
 gradio
 requests
 llama-index
-llama-index-llms-huggingface-api @ git+https://github.com/guillaumefrd/llama_index.git@add-provider-HF-API#subdirectory=llama-index-integrations/llms/llama-index-llms-huggingface-api
+llama-index-llms-huggingface-api
 llama_index.tools.wikipedia
 llama_index.tools.duckduckgo
 llama_index.tools.code_interpreter
@@ -10,4 +10,6 @@ langgraph
 langchain-openai
 langchain-community
 duckduckgo-search
-rizaio
+markdownify
+beautifulsoup4
+langchain_experimental