Skip to main content

System Overview

YuLan-OneSim is a next-generation social simulation platform powered by Large Language Models (LLMs) that enables researchers to create, execute, and analyze complex social simulations through natural language interactions.

Core Architecture Components​

YuLan-OneSim consists of four integrated subsystems:

1. Scenario Construction System​

  • Purpose: Transform natural language descriptions into executable simulation environments
  • Components: Code generation, scenario validation, environment initialization
  • Key Feature: Code-free scenario creation through conversational interface

2. Simulation Execution System​

  • Purpose: Provide reactive agent framework and distributed execution capabilities
  • Components: Agent runtime, event bus, environment management
  • Key Feature: Real-time agent interactions with event-driven architecture

3. Feedback Evolution System​

  • Purpose: Automatically optimize LLM performance based on external feedback
  • Components: Performance monitoring, model tuning, adaptive learning
  • Key Feature: Self-improving simulation quality over time

4. AI Social Researcher System​

  • Purpose: End-to-end automated research from problem formulation to report generation
  • Components: Research planning, experiment design, analysis automation
  • Key Feature: Autonomous scientific research workflow

Extensibility & Customization​

YuLan-OneSim is designed for extensibility:

  • Plugin Architecture: Custom components can be easily integrated
  • Configurable Models: Support for different LLM providers and models
  • Custom Environments: 50+ built-in scenarios with template for custom ones

Deployment Modes​

Single Node Mode​

  • Use Case: Small-scale simulations (< 1,0000 agents)
  • Setup: Single machine deployment
  • Benefits: Simple configuration, easy debugging

Distributed Mode​

  • Use Case: Large-scale simulations (1,0000+ agents)
  • Setup: Master-Worker cluster
  • Benefits: High throughput, fault tolerance, horizontal scaling

This architecture enables researchers to focus on their research questions while YuLan-OneSim handles the complexity of multi-agent simulation, distributed computing, and LLM integration.