Core runtime

• UE5-native C++ SDK — drop-in Unreal plugin; modular systems you can turn on/off per project.

• Multithreaded by design — agent updates, queries, and tasks parallelized for modern CPUs.

• Data-oriented architecture — cache-friendly containers, predictable memory, low allocator churn.

• Deterministic step option — fixed-tick simulation for reproducible gameplay and networking.

Scale & performance

• Large-crowd simulation — thousands to tens of thousands of agents at stable frame rates (exact numbers depend on content; reproducible benchmarks available on request).

• Projection-based rendering — swap per-agent render modes at runtime (full Skeletal Mesh ↔ Instanced Skeletal Mesh ↔ lightweight impostors) based on LOD or budget.

• Work budgeting — per-frame time budgets for AI/animation queries to keep frame times flat.

AI decision & behavior

• Playbook/Blackboard runtime — lightweight decision layer with conditions, scoring, and gating.

• Composable actions — define choices, sequences, and voting without tying logic to animation.

• Dynamic evaluation rates — hot agents think more often; distant/idle agents cheap to maintain.

• Perception hooks — LOS checks, proximity/area queries, and event-driven stimuli at scale.

Navigation & movement

• Async pathfinding — background path requests with cancellation and smart replanning.

• 3D/free-form locomotion — ground, air, and volume movement (XYZ), with accel/decel, pause/resume.

• Crowd-aware steering — separation/alignment/cohesion (boids-style) and obstacle avoidance.

• Navmesh integration — walkability checks, path corridor following, dynamic obstacle support.

Animation system

• Instanced Skeletal Mesh (ISKM) — batch thousands of similar rigs with shared pose data.

• Leader-pose sharing — followers derive transforms from a leader to minimize animation cost.

• Blend & start-at controls — programmatic clips, loop/fade, and precise time offsets.

• State-agnostic — animation is decoupled from decision logic (behaviors choose, anim plays).

Rendering & LOD

• Runtime LOD policy — per-agent rules to switch render backends and animation detail.

• Crowd culling — distance/angle/occlusion signals to thin out work before it hits the GPU.

• Material/slot mapping — consistent skinning/material setups across render modes.

Spatial systems

• Fast spatial hashing — constant-time neighborhood lookups for perception and steering.

• Zone & interest management — subscribe agents to local regions to cut broadcast noise.