World simulation skill

The Sim skill turns Nexma into a 3D agent-based simulation platform. It models agents, environments, physics, and interactions, and runs them at scale to predict emergent behavior.

What it covers

• Entities. Agent, Environment, Object, Force, Collision, PhysicsBody.

Each carries typed properties (mass, velocity, perception range, behavior policy).

• Relationships. AgentMovement, EnvironmentalInteraction, PhysicsEvent.

Events carry timestamp, participants, and outcome state.

• Constraints. Physics determinism per timestep, collision resolution under

configurable solvers, perception-radius limits, agent-count budgets per worker.

• Constants. Standard physics-engine defaults, gravity and friction tables,

behavior-tree primitives.

• Layer config. Agents as 3D models scaled by class, environments as terrain

meshes with material maps, force fields as gradient overlays.

• Toolbar tools. Generate agent population, run simulation, capture frame, export

trajectories.

Typical workflow

1. Scope. Define the environment polygon and bring a terrain mesh.
2. Population. Specify agent classes, counts, and starting policies.
3. Generate. Ask Jax: "Simulate 5,000 pedestrians evacuating this stadium under

alarm conditions. Use exit-prefer behavior with a 12-meter perception radius."

1. Inspect. Scrub the timeline; click agents for state history; overlay

density heatmaps.

1. Refine. Add or remove exits, change behavior policy, adjust agent mix —

re-run.

1. Validate. Conservation checks, deterministic-replay audit, bottleneck report.
2. Export. Trajectory CSV, video capture, or aggregate metrics from `Project →

Export`.

What Jax is good at, in Sim specifically

• Population generation that matches a target distribution (age, behavior class,

capability).

• Scenario branching — runs N variants with parameter sweeps and surfaces the

policy that hits the goal.

• Bottleneck detection in pedestrian, vehicle, or material-flow simulations.
• Causal explanation — explains why an outcome happened by tracing back through

agent state history.

Standards

Conservation laws (mass, momentum) where applicable; standard pedestrian-flow models (Helbing social force, ORCA reciprocal velocity obstacles).

What it does not do (yet)

• Continuum fluid simulation (agent-based only).
• Photoreal rendering (visualization is functional, not cinematic).
• Multi-million-agent simulations on a single browser session — server backend

required at that scale.