Skip to content

ArxivAgent Documentation

ArxivAgent is a class that helps fetch, process, and summarize scientific papers from arXiv. It uses LLMs to generate summaries of papers relevant to a given query and context.

Basic Usage

from ursa.agents import ArxivAgent

# Initialize the agent
agent = ArxivAgent()

# Run a query
result = agent.invoke(
    arxiv_search_query="Experimental Constraints on neutron star radius", 
    context="What are the constraints on the neutron star radius and what uncertainties are there on the constraints?"
)

# Print the summary
print(result)

Parameters

When initializing ArxivAgent, you can customize its behavior with these parameters:

Parameter Type Default Description
llm str "openai/o3-mini" The LLM model to use for summarization
summarize bool True Whether to summarize the papers or just fetch them
process_images bool True Whether to extract and describe images from papers
max_results int 3 Maximum number of papers to fetch from arXiv
database_path str 'arxiv_papers' Directory to store downloaded PDFs
summaries_path str 'arxiv_generated_summaries' Directory to store paper summaries
vectorstore_path str 'arxiv_vectorstores' Directory to store vector embeddings
download_papers bool True Whether to download papers or use existing ones

Advanced Usage

Customizing the Agent

agent = ArxivAgent(
    llm="openai/o3",  # Use a more powerful model
    max_results=5,       # Fetch more papers
    process_images=False,  # Skip image processing to save time
    download_papers=False  # Use only papers already in database_path
)

Running Multiple Queries

# First query
result1 = agent.invoke(
    arxiv_search_query="quantum computing error correction", 
    context="Summarize recent advances in quantum error correction techniques"
)

# Second query (will reuse downloaded papers if applicable)
result2 = agent.invoke(
    arxiv_search_query="quantum computing algorithms", 
    context="What are the most promising quantum algorithms for near-term devices?"
)

How It Works

  1. Fetching Papers: The agent searches arXiv for papers matching your query and downloads them as PDFs.

  2. Processing: If summarize=True, each paper is:

  3. Converted to text
  4. Split into chunks
  5. Embedded into a vector database

  6. If process_images=True, images are extracted and described using GPT-4 Vision

  7. Summarization: The agent:

  8. Retrieves the most relevant chunks based on your context
  9. Generates a summary for each paper

  10. Creates a final summary addressing your specific context

  11. Output: Returns a comprehensive summary that synthesizes information from all relevant papers.

Notes

  • Summaries and vector stores are cached, making subsequent queries faster.
  • The agent uses a ThreadPoolExecutor to process papers in parallel.
  • You can find the combined summaries in 'summaries_combined.txt' and the final summary in 'final_summary.txt'.