What it does ReGoap is a generic C# implementation of Goal Oriented Action Planning — the AI technique where agents figure out what to do by chaining actions that satisfy goals, rather than following scripted trees. It comes with adapters for both Unity and Godot, plus an engine-agnostic core you could theoretically yank out for something else.

The interesting bit The library treats planning as an A* search through world-state space, which means your NPCs can spontaneously discover action sequences you never explicitly authored. The newer comparator conditions (GreaterOrEqual, etc.) let you express thresholds like “craft swords until chest count ≥ 10” without exploding your state space into boolean flags.

Key highlights

• Engine-agnostic core with Unity and Godot scene-component adapters
• Weighted-random goal selection with optional deterministic seeding for reproducible tests
• Comparator-based state matching for numeric/inventory logic
• Replanning on world changes so agents don’t stand around holding obsolete plans
• Godot runtime debugger support (Unity debugging mentioned less specifically)

Caveats

• The README is thorough on concepts but thin on actual API surface — you’ll be reading the source or examples to get started
• No candidate images provided, so visual learners are out of luck

Verdict Worth a look if you’re building simulation-heavy games with emergent agent behavior and want to escape behavior tree hell. Skip it if your AI needs are simple enough that a switch statement still works, or if you’re not in Unity/Godot and don’t feel like writing your own adapter.