Summary

Paper-Agent is an intelligent academic research and survey tool designed for researchers and students, addressing the common pain points of time-consuming literature reviews and shallow analysis. Leveraging a multi-agent collaborative architecture powered by AutoGen and LangGraph, it automates the end-to-end process of literature retrieval, reading, analysis, synthesis, and report generation, delivering in-depth, insightful domain review reports.

How It Works

The system employs a modular, multi-agent architecture orchestrated by LangGraph. Core agents include a search_agent for retrieving papers from arXiv based on natural language queries (with manual review), a reading_agent that parallelly extracts structured information (problem, methods, results, limitations) into JSON and a ChromaDB vector store, and an analyse_agent performing clustering (KMeans), deep, and global analysis to identify research trends. A writing_agent then generates a report outline, splits tasks, and performs parallel, retrieval-augmented writing, culminating in a report_agent that consolidates chapters into a Markdown report with real-time progress updates via SSE.

Quick Start & Requirements

• Primary install: Use poetry install to set up the Python environment and dependencies. Run the backend with poetry run python main.py and the web UI with cd web && npm install && npm run dev.
• Prerequisites: Python 3.12+, API keys for LLM providers (e.g., OPENAI_API_KEY in .env), and configuration in models.yaml.
• Links: QQ Group: 340020097.

Highlighted Details

• Multi-agent collaboration architecture (AutoGen).
• Intelligent literature retrieval from arXiv with manual review.
• Structured information extraction into JSON.
• Deep domain analysis via clustering, deep, and global analysis stages.
• Automated domain review report generation in Markdown.
• Real-time streaming output of task progress via SSE.
• Parallel processing optimization for reading, analysis, and writing.
• Vector database support (ChromaDB) for retrieval-augmented writing.
• User interaction and review checkpoints throughout the workflow.

Maintenance & Community

The project is undergoing comprehensive refactoring and iterative maintenance. Community interaction and feedback are encouraged via GitHub Issues. A QQ group (340020097) is available for collaboration and discussion. Contribution guidelines are detailed in CONTRIBUTING.md.

Licensing & Compatibility

The project is licensed under the MIT License, which permits commercial use and integration into closed-source projects.

Limitations & Caveats

Several core functionalities, including paper downloading, deduplication, and filtering, are explicitly marked as "currently not supported, to be improved," indicating these features are incomplete or missing in the current implementation.