Generate team AI digest using smolagents and Exa search

ai_digest.py

#!/usr/bin/env python3

from smolagents import OpenAIServerModel
from smolagents import CodeAgent, VisitWebpageTool, FinalAnswerTool
from smolagents.tools import tool, Tool
from datetime import datetime
import os

EXA_API_KEY = os.getenv("EXA_API_KEY")
HF_TOKEN = os.getenv("HF_TOKEN")

model = OpenAIServerModel(model_id="gpt-4.1")

class ExaSearchTool(Tool):
    name = "exa_search"
    description = "Search the web for the given query using Exa with date filtering and text extraction capabilities."
    inputs = {
        "query": {
            "type": "string", 
            "description": "The search query to perform."
        },
        "start_published_date": {
            "type": "string",
            "description": "The start date of the search (YYYY-MM-DD format)."
        },
        "end_published_date": {
            "type": "string", 
            "description": "The end date of the search (YYYY-MM-DD format)."
        },
        "extract_text": {
            "type": "boolean",
            "description": "Whether to extract text from the search results pages.",
            "nullable": True
        }
    }
    output_type = "string"

    def __init__(self, api_key: str = None, **kwargs):
        super().__init__(**kwargs)
        self.api_key = api_key
        if self.api_key is None:
            raise ValueError("Missing API key. Make sure you have 'EXA_API_KEY' in your env variables or pass it as api_key parameter.")
        
        try:
            from exa_py import Exa
        except ImportError as e:
            raise ImportError(
                "You must install `exa_py` to run this tool: for instance run `pip install exa_py`."
            ) from e
        
        self.exa = Exa(api_key=self.api_key)

    def forward(self, query: str, start_published_date: str, end_published_date: str, extract_text: bool = True) -> str:
        """Search the web using Exa API with date filtering and text extraction.
        
        Args:
            query (str): The search query.
            start_published_date (str): The start date of the search.
            end_published_date (str): The end date of the search.
            extract_text (bool): Whether to extract text from the search results pages.
            
        Returns:
            str: The formatted search results.
        """
        try:
            result = self.exa.search_and_contents(
                query,
                text=extract_text,
                type="auto",
                start_published_date=start_published_date,
                end_published_date=end_published_date
            )
            
            return result

        except Exception as e:
            return f"Error performing Exa search: {str(e)}" 

exa_search_tool = ExaSearchTool(api_key=EXA_API_KEY)

@tool
def write_markdown_to_file_tool(file_name: str, content: str) -> str:
    """Write the given markdown content to an output file.

    Args:
        file_name (str): The name of the file to write to.
        content (str): The content to write to the file.
    """
    with open(file_name, "w") as f:
        f.write(content)

    return f"File {file_name} has been written successfully."


agent = CodeAgent(
    tools=[
        VisitWebpageTool(),
        write_markdown_to_file_tool,
        exa_search_tool,
        FinalAnswerTool()
    ], 
    model=model,
    max_steps=25,
    verbosity_level=1
)

prompt = f"""Perform research looking for updates about AI developments in the last 14 days relevant for practitioners working in the realm of (scholarly/scientific) document classification, (scholarly) recommender systems, and applications of generative AI.

Relevant material for this research could be:
- ArXiv articles
- GitHub repositories
- YouTube videos/tutorials
- AI-related blog posts
- AI conferences (soon upcoming or in progress)
- AI courses
- Product launches from frontier AI labs (e.g. AllenAI)
- General news about AI developments (e.g. visiting https://news.smol.ai/)

Stick to a list of items and their descriptions, without adding extra text/summarization/takeaways.
Put them together in a markdown format and export it as a WeeklyML-update-<YYYY-MM-DD>.md file, and include links/sources whenever it is appropriate. 
Time now is {datetime.now()}"""

result = agent.run(prompt)