This project is an agentic AI system that provides a structured, multi-perspective analysis of academic papers. By taking an arXiv ID as input, the agent ingests a paper, and then three distinct AI personas—the Optimist, Skeptic, and Ethicist—debate its claims. The final output is a cited, claim-by-claim debate summary and a concise TL;DR.
- PDF Ingestion: Automatically downloads and parses papers from arXiv using only the paper’s ID.
- Persona-Driven Analysis: Utilizes three specialized AI agents to analyze a paper from optimistic, skeptical, and ethical viewpoints.
- Structured Debate: Simulates a multi-round debate where agents respond to each other’s claims and counter-claims.
- Cited Arguments: All claims are grounded in the original paper’s text to minimize hallucination.
- Concise Summarization: Generates a final TL;DR that distills the entire debate into a clear, neutral summary.
Prerequisites: • Python 3.10 or higher • An OpenAI API Key
Clone the repository:
git clone https://github.com/midnightoatmeal/arxiv-agent.git
cd arxiv-agent
Create a virtual environment:
python3 -m venv venv
source venv/bin/activate
venv\Scripts\activate #windows
Install the required dependencies:
pip install -r requirements.txt
In the root directory, you will find a template file named .env.example.
Make a copy of this file and rename it to .env. You can do this with the following command in your terminal:
cp .env.example .env
Add your OpenAI API Key to .env in the following format:
OPENAI_API_KEY="sk-your-key-here"
The .env file is already included in .gitignore to keep your API key safe.
To run the agent and analyze a paper, execute the main script from your terminal:
python3 main.py <arxiv_id>
Example: To analyze the seminal “Attention Is All You Need” paper, use the ID 1706.03762.
python3 main.py 1706.03762
The script will download the paper and print a structured, claim-by-claim analysis and debate to your terminal.
arxiv-agent/
├── .env.example # Template for environment variables
├── main.py # Main script to run the project
├── pdf_ingestion.py # Handles PDF downloading and parsing
├── agentic_core.py # Contains the LLM agent logic and debate engine
├── requirements.txt # Project dependencies
└── .gitignore # Specifies files to ignore
Contributions are welcome! If you have ideas for new personas, better prompts, or improved debate logic, please open an issue or submit a pull request.
This project is licensed under the Apache-2.0 License. See the LICENSE file for details.
This project is an independent open-source tool and is not affiliated with, endorsed by, or sponsored by arXiv.org or Cornell University.
The use of the term arXiv in the project name is purely descriptive, referring to the fact that the tool can process papers available on arXiv using their IDs.