Add search_arxiv tool

app.py

@@ -1,14 +1,26 @@
-from smolagents import CodeAgent,DuckDuckGoSearchTool, HfApiModel,load_tool,tool
+from smolagents import CodeAgent,DuckDuckGoSearchTool, HfApiModel,load_tool,tool, VisitWebpageTool
 import datetime
 import requests
 import pytz
 import yaml
 from tools.final_answer import FinalAnswerTool
 from datetime import date
+from collections import namedtuple
+from typing import Tuple
 
 
 from Gradio_UI import GradioUI
 
+
+web_search = DuckDuckGoSearchTool()
+visit_webpage = VisitWebpageTool()
+
+final_answer = FinalAnswerTool()
+
+SearchResult = namedtuple("SearchResult", ["title", "url", "description"])
+
+
+
 # Below is an example of a tool that does nothing. Amaze us with your creativity !
 @tool
 def my_custom_tool(arg1:str, arg2:int)-> str: #it's import to specify the return type
@@ -29,6 +41,55 @@ def todays_date() -> str:
     """
     today = date.today()
     return today.strftime("%b %d, %Y")
+
+def parse_md_link(md_link: str) -> Tuple[str, str]:
+    """ parse markdown link to the title and link"""
+    pattern = r'\[([^\]]+)\]\(([^)]+)\)'
+    matches = re.findall(pattern, md_link)
+    return matches
+
+def search_specific_website(topic: str, url: str, num_results: int):
+    query = f"{url} {topic}"
+
+    query_result =  web_search(query).split("\n\n")[1:n]
+    output = []
+    for result in query_result:
+        link, description = result.split("\n")
+        title, url = parse_md_link(link)
+        output.append(SearchResult(title=title, url=url, description=description))
+    return output
+    
+
+def get_arxiv_paper_abstract(url):
+    result = visit_webpage(url)
+    for line in result.split("\n"):
+        if line.startswith("> Abstract:"):
+            return line.strip("> Abstract:")
+    return ""
+
+
+@tool
+def search_arxiv(topic: str, year: str = None, num_results: int = 10) -> str:
+    """
+    Search arxiv.org for a given topic. It will return the title and abstract of the top results
+
+    Args:
+        topic: The topic to search for.
+        year: Specic year to search for. Default is None
+        num_results: Indicating how many top search results should be included in the output
+
+    Returns:
+        A list of query result of title and abstract contcatenated by "\n\n".
+    """
+    if year is None:
+        year = ""
+    query_results =  search_specific_website(topic + " " + year, "site:arxiv.org", num_results)
+
+    output = []
+    for result in query_results:
+        output.append("Title: {} Abstract: {}".format(result.title, get_arxiv_paper_abstract(result.url)))
+    return "\n\n".join(output)
+
     
 
 @tool
@@ -47,9 +108,6 @@ def get_current_time_in_timezone(timezone: str) -> str:
         return f"Error fetching time for timezone '{timezone}': {str(e)}"
 
 
-final_answer = FinalAnswerTool()
-web_search = DuckDuckGoSearchTool()
-
 # If the agent does not answer, the model is overloaded, please use another model or the following Hugging Face Endpoint that also contains qwen2.5 coder:
 # model_id='https://pflgm2locj2t89co.us-east-1.aws.endpoints.huggingface.cloud' 
 
@@ -69,7 +127,7 @@ with open("prompts.yaml", 'r') as stream:
     
 agent = CodeAgent(
     model=model,
-    tools=[final_answer, web_search, todays_date], ## add your tools here (don't remove final answer)
+    tools=[final_answer, web_search, todays_date, search_arxiv], ## add your tools here (don't remove final answer)
     max_steps=6,
     verbosity_level=1,
     grammar=None,