Summary

WorldSeed is a multi-agent world engine designed for emergent outcomes, allowing users to "seed a world" rather than build a workflow. It targets engineers, researchers, and power users interested in creating complex simulations, games, or fictional worlds where AI agents interact based on defined rules and consequences. The primary benefit is enabling the observation and exploration of unpredictable, emergent behaviors arising from agent interactions within a flexible, scene-agnostic framework.

How It Works

The engine operates on a discrete "tick" loop, advancing the world state incrementally. In each tick, agents perceive a filtered, asymmetric slice of the world, propose actions, and the engine resolves them. Predictable actions are handled by a deterministic rule engine (DSL), while uncertain actions are adjudicated by an LLM-based "Dungeon Master." Worlds are entirely defined in YAML, specifying entities, rules, physics, and per-character perception, allowing the engine zero hardcoded domain knowledge. This approach facilitates complex interactions, allows for pluggable LLMs, and logs all state changes for replayability.

Quick Start & Requirements

• Prerequisites: Python 3.11+, Node.js 18+, uv.
• Installation & Run:

  git clone https://github.com/AIScientists-Dev/WorldSeed && cd WorldSeed
  uv sync --extra dm
  cd frontend && npm install && npm run build && cd ..
  cp .env.example .env
  # Add your API key (any LiteLLM provider) to .env
  uv run worldseed play configs/ai_layoffs.yaml
  

• Access: Dashboard available at http://localhost:8000.
• Links: Demo: https://worldseed.morphmind.ai/demo, Docs: docs/ARCHITECTURE.md.

Highlighted Details

• Scene-Agnostic Engine: The core engine is independent of specific world content, running any defined YAML configuration.
• Asymmetric Information: Perception rules ensure each agent possesses a unique, filtered view of the world state.
• AI-Assisted World Creation: YAML scene configurations can be generated from natural language prompts.
• Interactive Modes: Users can observe (Watch), influence (Intervene), or embody an agent (Play).
• Replayability: All past world runs are preserved and can be replayed.

Maintenance & Community

• Community: Discord (https://discord.gg/x9mtbMEx), WeChat, X (@morphmind__ai).
• Homepage: https://morphmind.ai.

Licensing & Compatibility

• License: MIT License.
• Compatibility: The MIT license generally permits commercial use and integration with closed-source projects without significant restrictions.

Limitations & Caveats

The provided README does not explicitly detail limitations, alpha status, or known bugs. The project appears to be actively developed with clear setup instructions and community channels.