ruvnet/swarm-world

Swarm-world is a plugin for the Unity game engine that lets you create flocking and group-movement behaviors with minimal effort. If you need birds that flock together, fish that school, enemy squads that coordinate, or large crowds that flow naturally through an environment, this project provides the building blocks so you don't have to write the underlying math and coordination logic from scratch. At a high level, the system works around two main concepts: a manager that oversees the whole group and individual agents that follow simple rules. Each agent looks at its nearby neighbors and adjusts its movement based on three classic behaviors, separation (don't crowd others), alignment (move in the same direction as neighbors), and cohesion (stay close to the group). When hundreds or thousands of agents follow those simple rules simultaneously, complex and lifelike group patterns emerge on their own. The plugin adds more advanced capabilities on top of that foundation, including obstacle avoidance, formation control (like V-formations or circles), predator-prey relationships, and target following. The tool is built for game developers, simulation creators, and researchers. An RTS developer could use it for unit formations and group pathfinding. An indie developer making an underwater adventure could populate the ocean with schooling fish that react realistically to the player. Architects and urban planners could simulate pedestrian flow through buildings or public spaces. The plugin includes over 20 ready-to-use example scenes covering scenarios like basic flocking, obstacle navigation, military formations, and predator-prey ecosystems. What stands out is the performance scaling. The system automatically handles anywhere from a handful of agents up to 10,000-plus by shifting between different technical strategies depending on scale, from standard Unity components for small groups, to multi-threaded processing for medium swarms, to GPU acceleration for massive crowds. This means a developer can prototype with a few dozen agents on a laptop and later scale up to thousands without rewriting their code.