arxiv-ai-agent 0.1.2

An LLM powered research assistant that finds, ranks, and explains recent AI papers on arxiv.org.

Arxiv.org Agent

An LLM powered research assistant that finds, ranks, and explains recent AI papers on arxiv.org.

▶️ Watch a demo: https://youtu.be/PqJiYTvOP1M

🚀 Try it Online (no installation needed)

You can use the hosted Streamlit version here:

👉 https://arxiv-agent-atclu7bhsjhbqjk8lndeus.streamlit.app/

You'll provide your own OpenAI API key in the sidebar.
Your key is used only in memory for the session and never saved or logged.

---

📦 Install Locally (via PyPI)

pip install arxiv-ai-agent

Then launch the app:

arxiv-ai-agent

This will start the Streamlit interface and open it in your browser (or show you a Local URL if the browser does not auto-launch).

💻 Install from GitHub (development mode)

git clone https://github.com/nurtekinsavasai/arxiv-agent.git
cd arxiv-agent

python3 -m venv .venv
source .venv/bin/activate

pip install -e .

Then run:

arxiv-ai-agent

🔑 OpenAI API Key

The app uses the OpenAI API for:

• Relevance classification
• Citation prediction
• Plain-English paper summaries
• Embeddings (text-embedding-3-large)

Your key is:

• Provided manually in the sidebar
• Used only for the active session
• Never saved
• Never exported
• Never written to disk

You are billed directly under your own OpenAI account.

📚 What the Agent Does

You provide:

• A short research brief in natural language
• Optional exclusions (things you don't want)
• A date range (3 days, 7 days, or 30 days)

The agent:

• Fetches recent cs.AI + cs.LG papers from arXiv
• Uses embeddings to rank semantic similarity
• Uses an LLM to classify papers as primary, secondary, or off-topic
• Selects ~20 promising papers
• Predicts 1-year citation impact using an LLM
• Ranks papers by predicted influence

Provides:

• Paper metadata
• Abstract
• PDF link
• Plain-English summary
• Explanation of the citation prediction
• Lets you download all artifacts as a ZIP

⚠️ Limitations & Ethical Notes

• Citation predictions are heuristic and approximate.
• They may reflect academic biases, trends, and popularity.
• This tool is for exploration and inspiration — not formal evaluation of researchers, institutions, or grant proposals.
• Always read the actual papers before relying on any automated ranking.

🧭 Future Work

• Expand beyond cs.AI + cs.LG
• Support more LLM providers and models
• Add multi-step agentic reasoning
• Add full background workflows for autonomous research assistants
• Enhance citation prediction framework