Worldseed
More is Different. A multi-agent world engine where AI agents live, talk, compete, ally.
Last verified:
What is Worldseed?
WorldSeed is an open-source multi-agent world engine that allows AI agents to live, talk, compete, and form alliances in a persistent simulated world with rules, space, items, and information asymmetry. Unlike mainstream multi-agent frameworks that focus on task collaboration, WorldSeed creates a runtime environment layer where agents truly inhabit a world with constraints, secrets, and game-like dynamics. The tool is developed by the MorphMind team, which includes researchers from Harvard, Tsinghua, and leading tech companies.
Key features include sandboxed agent lifecycles, event-driven interactions, YAML-based scenario definition (no coding required), true information asymmetry where each agent only sees what rules allow, spatial and temporal modeling, resource and goal systems, and tooling to replay and inspect multi-agent traces for analysis. The repository includes two ready-to-run demos: an AI Layoffs scenario and another scenario where users can directly observe how AI agents probe, lie, form alliances, and collapse without any account needed.
WorldSeed is designed for researchers studying emergent agent behavior and social simulation, developers building multi-agent systems who need an environment layer for testing, creators making interactive narratives or AI NPC worlds, teachers conducting classroom demonstrations and human-AI collaboration experiments, and anyone interested in game-like AI simulations such as tabletop RPGs or text game prototypes.
The tool is MIT licensed, model-agnostic (can connect different agent frameworks and LLMs), and serves as a reproducible sandbox for studying agent-to-agent cooperation and competition, generating evaluation traces for reputation systems, and running what-if scenarios with structured, inspectable outputs.
Worldseed pricing
Pricing model: Freemium
WorldSeed is completely free and open-source under the MIT license. The core multi-agent world engine can be self-hosted at no cost. There are no paid tiers, subscription plans, or usage fees for the WorldSeed software itself. The browser demo at worldseed.morphmind.ai/demo is free to use without creating an account. Note that while WorldSeed is free, running simulations will incur costs for the underlying LLM API calls if using paid models like ChatGPT or Claude.
Worldseed pros
- Open-source under MIT license with no vendor lock-in
- No coding required - define scenarios in YAML
- True information asymmetry between agents
- Sandboxed agent lifecycles prevent rule violations
- Event-driven interactions produce emergent behavior
- Built-in replay and trace inspection tooling
- Model-agnostic - works with any LLM or agent framework
- Two ready-to-run demos included out of the box
- Browser demo accessible without creating an account
- Persistent world with space, time, items, and rules
- Reproducible simulations for scientific evaluation
- Structured traces essential for A2A evaluation
- Runtime-level environment layer, not just orchestration
- AI agents act as participants, not just task assistants
- Supports social simulation and game design use cases
Worldseed cons
- Relatively new project with limited documentation
- Requires Python and TypeScript setup for local deployment
- No official cloud hosting - self-hosting required
- Steep learning curve for designing balanced world rules
- Limited community ecosystem compared to mature frameworks
- No visual editor - YAML configuration only
- May require significant compute for long-running simulations
- Debugging emergent behavior can be challenging
Frequently asked questions about Worldseed
What is WorldSeed and how is it different from other multi-agent frameworks?
WorldSeed is a multi-agent world engine where AI agents live, talk, compete, and ally in a persistent world with rules, space, items, and information asymmetry. Unlike most multi-agent frameworks that focus on task collaboration under shared context, WorldSeed creates a true runtime environment where agents are constrained by world rules, have genuine information asymmetry (each agent only sees what rules allow), and cannot逾越 boundaries. The world exists independently and agents enter it, rather than agents pretending to live in a world.
Do I need coding skills to use WorldSeed?
No coding is required for basic usage. You define scenarios in YAML configuration files, specifying roles, goals, secrets, and world rules. The repository includes two ready-to-run demos that can be observed directly in the browser without any account or setup. For advanced customization or integration into your own projects, Python knowledge would be helpful but the core experience is designed for non-programmers.
Can I use WorldSeed with my preferred LLM or agent framework?
Yes, WorldSeed is model-agnostic and runtime-level, meaning it can connect different agents playing roles regardless of which LLM or agent framework they use. You can plug in agents from different systems, making it flexible for testing how various agent configurations behave in the same environment.
What are the main use cases for WorldSeed?
Four main use cases include: (1) Multi-agent social simulation and game theory experiments for researchers studying organization behavior and complex systems; (2) Building runnable AI剧情/game worlds for interactive叙事, tabletop RPGs, text game prototypes, and AI NPC worlds; (3) Serving as an environment layer for multi-agent systems to test agent behavior under complexity; (4) Teaching, demonstrations, and ablation tests where teachers or teams can do classroom demos, role-playing experiments, and human-AI collaboration tests.
Is WorldSeed really open-source and free?
Yes, WorldSeed is MIT licensed open-source software. The complete codebase is available on GitHub at github.com/AIScientists-Dev/WorldSeed. You can self-host it for free without any subscription fees or usage limits. The browser demo is also free to use without creating an account. The only potential costs would be for the underlying LLM API calls if using paid models.
How do I get started with WorldSeed?
You can start immediately by visiting the browser demo at worldseed.morphmind.ai/demo which requires no account - you can directly observe AI agents interacting in scenarios like the AI Layoffs demo. For local development, clone the GitHub repository and follow the quickstart instructions. The official recommendation is to first verify world consistency with few characters, few rules, and clear goals before gradually adding complexity.
What makes information asymmetry important in WorldSeed?
True information asymmetry is one of WorldSeed's most valuable features. Each agent runs independently and can only see information that rules permit - they don't share context like in traditional multi-agent systems. This creates genuine secrets, deception opportunities, alliance dynamics, and emergent behavior that more closely mirrors real-world social interactions. The recommendation is to design information asymmetry as a core mechanic, not an afterthought, carefully considering who knows what, who doesn't know what, and when states change.
Can I analyze and replay what happens in WorldSeed simulations?
Yes, WorldSeed provides tooling to replay and inspect multi-agent traces for analysis. The engine produces structured, reproducible traces that are essential for building agent track records and A2A (agent-to-agent) evaluation. Unlike many ephemeral multi-agent demos, WorldSeed makes behaviors reproducible and inspectable, which is critical for evaluation and trust research.
What scenarios are included in the official demos?
The repository includes at least two ready-to-run demos. The AI Layoffs scenario features a tech company that just cut 30% of staff, with six AI agents navigating a critical bug nobody can fix, a junior who rewrote three years of code with AI in ten minutes, a designer watching her work get replaced by Midjourney, a VP watching fake green metrics from home, and rumors of a second round. Agents run independently, can only see what rules allow, and nothing is scripted.
Who created WorldSeed and what is their background?
WorldSeed is open-source released by the MorphMind team. MorphMind is a controllable AI platform for individual users, built by AI research and engineering teams from Harvard, Tsinghua, and leading tech companies. WorldSeed grew out of the multi-agent engine they built at MorphMind, where domain experts compose AI agents just by talking. The team's goal is to let users build, supervise, and continuously evolve their AI agent teams according to their own ideas.