Paper Circle: An Open-Source Multi-Agent Framework for Research Discovery and Analysis

Paper Circle: AI Agents That Discover, Analyze, and Organize Scientific Literature

Researchers have released Paper Circle, an open-source multi-agent system that uses LLMs to automate the tedious process of discovering, evaluating, and synthesizing academic literature.

The Problem

The exponential growth of scientific publications makes it nearly impossible for researchers to keep up:

• Over 5 million papers published annually across all fields
• Literature reviews take weeks to months of manual work
• Researchers miss relevant work due to information overload
• Cross-disciplinary connections go unnoticed

Paper Circle's Architecture

The system uses two complementary pipelines:

1. Discovery Pipeline:

• Offline + online retrieval from multiple academic sources
• Multi-criteria scoring for relevance and quality
• Diversity-aware ranking to avoid redundant results
• Structured output in JSON, CSV, BibTeX, Markdown, and HTML

2. Analysis Pipeline:

• Transforms individual papers into structured knowledge graphs
• Typed nodes: concepts, methods, experiments, figures
• Enables graph-aware question answering
• Coverage verification across research domains

Multi-Agent Orchestration

Paper Circle implements a coder-LLM-based multi-agent framework where specialized agents handle different tasks:

• Search agents — Query multiple databases simultaneously
• Scoring agents — Evaluate paper relevance and quality
• Graph agents — Build and query knowledge graphs
• Writing agents — Generate literature reviews and summaries

Benchmark Results

Why It Matters

Paper Circle represents a new paradigm in research tools:

• Fully reproducible — Every step produces synchronized outputs
• Open source — Available for community extension and adaptation
• Domain agnostic — Works across any academic field
• Scalable — Multi-agent architecture handles growing literature volumes

The system is publicly available as a web application.