SeedWorld is an emergent world simulation engine that leverages Large Language Models (LLMs) to generate complex, unscripted civilizations and narratives. It targets users interested in AI-driven storytelling, world-building, and complex systems, offering a unique "terrarium" for observing organic societal evolution without pre-written scripts.

How It Works

The engine operates via three steps: users Describe a world setting, the engine Generates characters with distinct personalities and relationships, and users Watch these agents independently make decisions. Core to its design are 12 interlocking "mechanism systems" (e.g., Reputation, Resource Competition) that track state across simulation ticks. These systems provide rich context to the LLM, which generates agent actions. This approach prioritizes emergence by allowing LLMs to react organically to pressures and relationships, rather than following predefined scripts. Novelty lies in the LLM's self-evaluation of action impact and routing this feedback through mechanism systems, closing a continuous loop.

Quick Start & Requirements

• Installation: Clone the repository (git clone https://github.com/zmzhace/world-slice.git), navigate into the directory (cd world-slice), and install dependencies (npm install).
• Prerequisites: Node.js environment, and access to an OpenAI or Anthropic-compatible LLM API.
• Configuration: Set up .env.local with WORLD_SLICE_API_BASE, WORLD_SLICE_API_KEY, and WORLD_SLICE_MODEL.
• Running: Execute npm run dev to start the development server, accessible at http://localhost:3000.

Highlighted Details

• Interactive Chat: Inject events via chat; the LLM interprets and integrates them into the simulation.
• Tick Lifecycle: Time progresses in discrete ticks, with each agent independently deciding actions based on tailored prompts incorporating state and mechanism system feedback.
• LLM Feedback Loop: Agent actions generate structured system_feedback (e.g., reputation changes) that update mechanism systems, influencing future prompts.
• Wave Execution: Co-located agents act in sequential waves within a tick, enabling natural dialogue and reactive sequences.

Maintenance & Community

The provided README does not detail specific maintenance contributors, community channels (e.g., Discord, Slack), or a public roadmap.

Licensing & Compatibility

The project is licensed under the MIT License, permitting broad use and modification. It is designed to be compatible with any LLM API adhering to OpenAI or Anthropic interfaces.

Limitations & Caveats

The simulation's emergent behavior is heavily dependent on the underlying LLM's capabilities and API costs. The complexity of interpreting and managing highly emergent narratives may pose a challenge. The README does not explicitly detail potential limitations regarding scalability, specific unsupported scenarios, or known bugs.