An elegant, locally deployable tool to browse, filter, and discover the latest ArXiv papers with a modernized card-style interface.
Staying updated with the latest research should feel as effortless as scrolling through a news-feed app. CodeArXiv transforms the raw ArXiv feed into a visually engaging, personalized experience.
Experience research like never before. Say goodbye to dense text lists and explore papers through an elegant Card UI that brings abstracts, authors, and key figures to life.
We leverage state-of-the-art LLMs to summarize, tag, and translate every paper, turning raw data into actionable insights.
- ⚡ AI-Powered Tl;dr: Powered by GPT-5.2 (via OpenAI Codex CLI), we distill complex papers into a precise one-sentence summary for rapid browsing.
- 🧠 Multi-dimensional Tagging: Every paper is deeply analyzed and tagged across multiple dimensions—including
task,method,property, andspecial paper type—to help you find exactly what you need. - 🌐 Chinese-Friendly & Bilingual: Break language barriers with high-quality translations from GPT-5.2 (via OpenAI Codex CLI). Seamlessly switch between English and Chinese for titles, abstracts, and conclusions.
- 🔎 Semantic Similarity: Paper/favorite embeddings are generated with Qwen3-Embedding to power similarity scores and recommendations.
Don't just read—curate and control your feed.
- 📂 Multi-Folder Favorites: Organize your research effectively. Create multiple favorite folders to categorize papers by project, topic, or interest.
- 🎯 Context-Aware Recommendations: The system calculates semantic similarities between new papers and your specific collections, ensuring you never miss relevant research.
- 🎛️ Tag Whitelists & Blacklists: Take full control of your feed. Set up custom Tag Whitelists to prioritize topics or Blacklists to filter out noise.
- 🔒 Full Local Control: Deploy the web interface entirely on your local machine. Your reading history, favorites, and preferences stay private.
- 👥 Flexible User Modes: Supports multi-user registration and login for shared deployments. For personal use, it defaults to a frictionless "No-User Mode." (Passwords are hashed and verified.)
- 🔄 Effortless Data Sync: We process and update ArXiv metadata and AI insights daily on the CodeArXiv-data HuggingFace repository. Your local instance can automatically sync with our dataset using a single command/script.
Get up and running in minutes.
```
git clone https://github.com/caopulan/CodeArXiv.git
cd CodeArXiv
```
```
# Install uv first if needed: https://docs.astral.sh/uv/
# Install Python 3.12 if you don't have it yet
uv python install 3.12
uv sync -p 3.12
# Optional: configure env vars
cp .env.example .env
# Initialize database
uv run flask --app app init-db
```
To use the hf CLI, install the optional extra:
uv sync -p 3.12 --extra hf
a) Download all data without images
uv run hf download caopu/CodeArXiv-data \
--include "*.json" \
--repo-type dataset \
--local-dir ./CodeArXiv-data
b) Download specific date data without images
uv run hf download caopu/CodeArXiv-data \
--include "2025-12-04.json" \
--repo-type dataset \
--local-dir ./CodeArXiv-data
c) Download specific date data with images
Note: Downloading image data requires 1~2GB per day.
Change --include to:
--include "2025-12-04.json" "images/2025-12-04/*"
d) Download all data
uv run hf download caopu/CodeArXiv-data \
--repo-type dataset \
--local-dir ./CodeArXiv-data
```
uv run flask --app app run --debug
```
Visit http://localhost:5000.
- Basic Card UI & Responsive Design
- Local Deployment Support
- Data Mirror: Upload dataset to ModelScope for broader accessibility
- Open Source Pipeline: Release the code for paper extraction, summarization, and tag processing
- Mobile Access Guide: Tutorial for deploying on a personal server for remote access via mobile devices
- Public Web Service: Build an official hosted website. (We are looking for experts to help us architect a robust public demo. If you are interested, please check the issues or reach out!)
Contributions are what make the open-source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated.
- Fork the Project
- Create your Feature Branch (
git checkout -b feature/AmazingFeature) - Commit your Changes (
git commit -m 'Add some AmazingFeature') - Push to the Branch (
git push origin feature/AmazingFeature) - Open a Pull Request
This project was developed with the help of AI assistants: the core repository is built upon OpenAI's Codex, and the GitHub page documentation was generated via Gemini.