WorldSeed

A YAML-defined world engine for running, observing, and steering emergent multi-agent scenarios.

Stars
★ 822
Last updated
4mo ago
License
MIT
Primary language
Python

At a glance

Works with
Universal · cross-platformCodex · OpenAI API · Claude API
You'll need
Python 3.11+Node.js 18+uvnpmAPI key for a LiteLLM providerShell / CLINetwork accessLocal filesystem
Typical use
A research group wants specialist characters to turn an early idea into hypotheses, experiments, peer review, and an auditable research trail.
Main limitation
The documented setup requires Python 3.11+, Node.js 18+, uv, and an npm frontend build.

What does this agent do, and when should you use it?

WorldSeed is a local world engine for emergent multi-agent scenarios. You define entities, roles, private information, perception filters, actions, rules, and consequences in YAML, then start a scene with the worldseed CLI. Its tick loop gives each character a filtered view of the world and resolves actions through either the YAML DSL or an LLM-based Dungeon Master that returns structured effects. A local dashboard supports observation, private intervention, and playing as a character; runs are preserved for replay. Bundled scenes include AI Layoffs and Teahouse, while documented integration paths include OpenClaw agents, Codex subagents, and models supported by LiteLLM.

Run uv run worldseed validate configs/my_scene.yaml to validate a scene YAML file, then run uv run worldseed play configs/my_scene.yaml to execute it. On every tick, WorldSeed filters state through each character's perception rules, gathers proposed actions, resolves deterministic actions with the in-YAML DSL, and sends uncertain outcomes to the LLM Dungeon Master. The Dungeon Master returns structured effects, which mutate state, trigger consequences, and are logged as the world advances. The local dashboard at http://localhost:8000 provides Watch, Intervene, and Play modes. The /create-world command generates validated YAML scene and UI configuration from a prompt, and authors can also edit files such as configs/teahouse.yaml and configs/ai_layoffs.yaml directly.

  1. A research group wants specialist characters to turn an early idea into hypotheses, experiments, peer review, and an auditable research trail.
  2. A narrative or game designer needs to test an espionage or workplace scenario involving secrets, factions, and asymmetric information.
  3. An AI-tool team wants API researchers, builders, critics, audience agents, and a curator to compete over artifacts in one traceable scene.
  4. A developer wants to experiment with social-simulation rules, perceptions, and consequence chains in YAML instead of rebuilding a domain-specific engine.
  5. An engineering team wants to connect OpenClaw agents or Codex subagents to an observable world where operators can privately influence characters.

How do you install or deploy this agent?

Requirements: Python 3.11+, Node.js 18+, and uv. 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 credentials for a LiteLLM provider in .env; the documented examples include OpenAI, Anthropic, and Ollama.

How do you use this agent?

Validate a bundled scene with uv run worldseed validate configs/ai_layoffs.yaml, then start it with uv run worldseed play configs/ai_layoffs.yaml. Open http://localhost:8000 and choose Watch to observe, Intervene to privately message an agent, or Play to enter a character. For a new world, use /create-world "your world description", or author YAML directly, then validate and launch it with uv run worldseed validate configs/my_scene.yaml and uv run worldseed play configs/my_scene.yaml.

What are this agent's strengths and limitations?

Pros
  • YAML declares characters, rules, private information, and perception filters, while the engine has no hardcoded domain knowledge.
  • It separates deterministic DSL resolution from uncertain LLM judgment, with the Dungeon Master returning structured effects rather than free-form prose.
  • Character-specific perception filtering directly supports secrets, conflicting beliefs, intelligence work, and social simulation.
  • The local dashboard combines observation, private intervention, role-play, and replayable run history.
Limitations
  • The documented setup requires Python 3.11+, Node.js 18+, uv, and an npm frontend build.
  • A LiteLLM provider configuration and credentials are required; Dungeon Master outcomes depend on the selected model.
  • Creating robust worlds requires maintaining YAML scene and UI configuration, perception rules, and DSL rules.
  • The documented path is local startup; production deployment, hosted operation, and cost controls are not specified.

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

Agent Source review Stars Updated Language Full support on
WorldSeed This agent 37 · Major gaps ★ 822 4mo ago Python Codex · OpenAI API · Claude API
MassGen 51 · Major gaps ★ 1.1k 3mo ago Python Codex · Claude Code · OpenAI API · Claude API
IR-SIM: Lightweight Robot Simulator 43 · Major gaps ★ 1.1k 5d ago Python
RAI — Embodied AI Agent Framework for Robotics 55 · Major gaps ★ 592 13d ago Python

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
37/ 100 5-point scale 1.9 / 5
Trust 0/29
Reliability 8/14
Adaptability 10/18
Convention 9/18
Effectiveness 7/13
Verifiability 3/8
Why each dimension lost points
Trust0 / 29 · 0.0/5

Evidence shows no permission model, user confirmation, data flow transparency, sensitive data handling, dependency security audit, external effect control, rollback mechanism, or source attribution. All trust criteria are absent, hence 0.

Reliability8 / 14 · 2.9/5

Self-consistency: README, pyproject.toml, and test files are consistent in architecture description, but no runtime verification. Dependency availability: dependency list is clear, but no lock file or integrity checks. Failure messages: no error handling or user-actionable failure hints in tests or docs.

Adaptability10 / 18 · 2.8/5

Audience and scenarios: README describes multiple scenarios but not target users. Capability boundaries: docs explain engine generality but not limitations. Trigger precision: no explicit trigger conditions or input validation. Environment fit: installation and run instructions provided but not all platforms covered.

Convention9 / 18 · 2.5/5

Information architecture: clear doc structure with architecture. Install notes: installation steps provided. Naming stability: no API stability promise. Examples and FAQ: examples provided, no FAQ. Known limitations: not explicitly listed. License: MIT license clear. Versioning and changelog: version number but no changelog. Maintenance responsibility: maintainer not specified.

Effectiveness7 / 13 · 2.7/5

Output usability: demos and examples provided but output quality not verified. Marginal value: unique features but no comparison with alternatives. Cost-benefit: no performance or cost data.

Verifiability3 / 8 · 1.9/5

Claim traceability: README claims lack evidence. Cross-source corroboration: no external validation. Fact-inference separation: no clear distinction between facts and inferences.

Risks and how to mitigate them
  • Not found in source: least-privilege scopingGrant only what the task needs: a dedicated account or read-only token, scoped to specific directories and repos.
  • Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
  • Not found in source: data-flow disclosureWatch which external services it contacts (proxy or firewall logs) and keep sensitive data out until you know where it goes.
  • Not found in source: sensitive-data handlingUse dedicated, low-privilege, revocable API keys — never production credentials — and keep secrets out of logs.
  • Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
  • Not found in source: disclosed external effectsEstablish which external systems it writes to, sends to or changes, and verify with test accounts or repos before production.
  • Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
  • Not found in source: verifiable attributionInstall from the official repo or registry and check the publisher and URL to avoid look-alike packages.
  • The repository provides no security or permission control mechanisms; running agents may have unrestricted permissions.
  • Dependencies are not pinned, posing supply chain risks.
  • No rollback or recovery mechanism is provided; failures may lead to state loss.
Evidence confidence: Low Reviewed Aug 09, 2026 Reviewed revision 4ffeabfdee6d
See the full review method →

FAQ

Is it tied to one model provider?
No. The Dungeon Master is described as compatible with any LiteLLM-supported model; documented examples include OpenAI, Anthropic, and Ollama.
Which actions are deterministic?
Predictable actions are resolved by the YAML DSL. Uncertain actions are judged by the LLM Dungeon Master and returned as structured effects.
What can an operator do during a run?
At the local dashboard, an operator can watch characters and their inner state, privately intervene with an agent, or play as a character.
What does it cost to run?
A LiteLLM provider API key is required, but the repository does not specify call volume, pricing, or budget-control features.
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related agents