arxivflow 0.2.2

Automate arXiv paper tracking with LLM-powered metadata extraction and Google Sheets sync.

arXivFlow 🚀

arXivFlow is a powerful Python-based automation tool designed to streamline the research paper discovery and tracking process. It autonomously fetches metadata from arXiv, performs local AI-driven analysis using Ollama (e.g., Llama 3.2), and synchronizes the results with Google Sheets and local databases.

---

✨ Features

• Asynchronous API: Fully rewritten with asyncio for high-performance paper retrieval and PDF processing.
• Automated Retrieval: Fetch the latest papers from specific arXiv categories (e.g., cs.AI, cs.LG, hep-ph) within any date range.
• Local AI Analysis: Uses Ollama models (e.g., Llama 3.2) to extract keywords and contact information (emails/affiliations) directly from PDF text. No cloud API costs or data privacy concerns.
• Intelligent PDF Handling: Automatically downloads PDFs and extracts text for deep analysis. Supports custom storage paths and atomic PDF writes.
• Robust arXiv Requests: Built-in compliance with arXiv's API guidelines (3-second request intervals), paged metadata retrieval, 429 cooldown handling, retry backoff, and duplicate-result cleanup.
• Multi-Format Export: Save your research data to CSV, JSON, Excel, or SQLite for flexible offline analysis.
• Google Sheets Sync: Seamlessly push compiled research data to a shared Google Sheet for team collaboration.
• Type-Safe & Modular: Clean, documented Python code with full type hinting and a class-based architecture.

---

🛠️ Prerequisites

1. Python 3.13+: Ensure you have a modern Python environment.
2. Ollama: Install Ollama and download the required model (e.g., Llama 3.2):

   ollama pull llama3.2
   

3. Google Cloud Credentials:
   • Enable the Google Sheets and Google Drive APIs.
   • Create a Service Account and download the JSON key as credentials.json.
   • Ensure the service account has 'Editor' permissions on the sheet.

---

🚀 Installation

From PyPI (Recommended)

pip install arxivflow

From Source (For Development)

1. Clone the repository:

   git clone https://github.com/zjzhao/arXivFlow.git
   cd arXivFlow
   

2. Set up virtual environment:

   python -m venv .
   source bin/activate  # On Windows: Scripts\activate
   

3. Install dependencies:

   pip install -e .
   

---

📖 Usage

Quick Start (Async)

import asyncio
import datetime
from arxivflow import arXivFlow

async def main():
    # 1. Initialize the flow
    flow = arXivFlow(
        categories=["cs.AI", "cs.CV"], 
        ollama_model="llama3.2",
        max_results=20,
        start_date=datetime.datetime.now() - datetime.timedelta(days=7),
        request_timeout=60.0
    )

    # 2. Fetch data & Extract info (Keywords/Contacts)
    df = await flow.get_arxiv_data(download_pdfs=True)

    # 3. Save to your preferred formats
    flow.save_to_csv("my_research.csv")
    flow.save_to_sqlite("research.db")

    # 4. Sync with Google Sheets
    flow.save_to_google_sheet(
        sheet_id="YOUR_SHEET_ID", 
        credentials_file="credentials.json"
    )
    
    # 5. Close the client
    await flow.close()

if __name__ == "__main__":
    asyncio.run(main())

---

🧱 Request Stability

arXiv can occasionally return slow responses, rate limits, or temporary service errors. arXivFlow now makes the request path more stable by:

• Fetching arXiv metadata in smaller pages instead of relying on one large request.
• Fetching metadata for all requested categories before starting PDF downloads, which avoids PDF download bursts interfering with the next category query.
• Serializing arXiv requests and preserving the recommended 3-second interval.
• Retrying transient failures (429, 500, 502, 503, 504, timeouts, and network errors) with exponential backoff and jitter.
• Applying a longer cooldown after 429 rate-limit responses before making the next arXiv request.
• Respecting Retry-After headers when arXiv provides them.
• Using a default 60-second request timeout, configurable with request_timeout.
• Writing PDFs to temporary .part files first, then atomically replacing the final file only after validating PDF-like content.
• Deduplicating merged output by arXiv ID.

For especially large date ranges, prefer smaller max_results values or narrower date windows. arXivFlow will page requests internally, but smaller slices are still easier for arXiv and more reliable in practice.

---

🏗️ Architecture

The project follows a modular structure for easy extension:

• src/arxivflow/arxivflow.py: The main orchestrator class (arXivFlow).
• src/arxivflow/ollama_functions.py: Local LLM interface using the Ollama API.
• src/arxivflow/arxiv_functions.py: Asynchronous arXiv API interaction layer, including paging, rate limiting, retries, and PDF downloads.
• src/arxivflow/categories.py: arXiv category definitions.

---

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

🤝 Contributing

Contributions are what make the open-source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated.

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
3. Commit your Changes (git commit -m 'Add some AmazingFeature')
4. Push to the Branch (git push origin feature/AmazingFeature)
5. Open a Pull Request