LLM-agnostic
OpenAI, Anthropic proxies, Ollama — anything OpenAI-compatible.
WorldSimMulti-agent community simulation
Simulate how communities react to new rules, events, or policies — in TypeScript, in 5 minutes.
Simulate how communities react to new rules, events, or policies — in TypeScript, in 5 minutes.
WorldSim is an embeddable multi-agent simulation engine for Node.js. You define a world, add agents with personalities and goals, optionally load rules or a crisis trigger, then let the engine advance tick by tick while agents reason, talk, use tools, build relationships and produce a report.
Use it when you want to test questions like:
At runtime WorldSim is a loop around five ideas:
| Concept | What it means |
|---|---|
| World | The container for time, agents, rules, plugins, stores and reports. |
| Ticks | Discrete simulation steps. A tick can represent a minute, an hour, a day or any turn in your scenario. |
| Agents | LLM-driven actors with personality, mood, energy, memory, goals and optional tools. |
| Rules | JSON/PDF instructions evaluated by governance agents to warn, block or allow actions. |
| Plugins | Extension points for logging, reports, Studio, phones, movement, perception, custom tools and domain logic. |
The default interaction model is simple and backward-compatible: agent speech is routed through conversations, neighborhoods, proximity and finally broadcast. For more realistic worlds you can opt into the perception layer, where speech and entity events become stimuli that agents must physically perceive before they can react.
| Mode | Use it when | Behavior |
|---|---|---|
legacy (default) | You want classic social simulations and maximum compatibility. | Messages flow through the legacy router and can fall back to broadcast. |
perception | Location, senses, attention and causal threading matter. | Agents only react to perceived stimuli; strict mode can drop unheard speech. |
| Goal | Start here |
|---|---|
| Watch a ready-made scenario | npx worldsim demo or npx worldsim studio |
| Build your own policy simulation | Copy evaluation/scenarios/water-rationing/ |
| Test realistic location/senses behavior | Read Perception Layer and copy evaluation/scenarios/village-realistic/ |
| Compare legacy vs perception | npm run eval:compare-perception |
| Integrate WorldSim in an app | Use WorldEngine, stores, plugins and the generated SimulationReport |
| Feature | Description |
|---|---|
| LLM-agnostic | OpenAI, Anthropic proxies, Ollama — anything OpenAI-compatible |
| Personality system | Mood, energy, goals, beliefs, knowledge per agent |
| Realistic perception | Opt-in stimulus/perception/attention/topic/needs stack for physics-aware agent interactions |
| Social dynamics | Relationship tracking with strength decay, neighborhoods |
| Rule enforcement | Hard/soft rules, governance agent with autonomous control |
| Scalability | 1000+ agents via concurrency caps, activity scheduling, token budgets |
| Zero-config persistence | In-memory by default; plug in Redis, Neo4j, PostgreSQL for production |
| Real-time streaming | Socket.IO events for live dashboards |
| Simulation reports | Auto-generated analysis with mood heatmaps and action metrics |