Wardenmind

Overview

Unreal provides the AI mechanisms (controllers, Behavior Trees, perception, EQS, StateTree, Smart Objects, Mass) but not the behavior on top, so every project rewrites a guard that patrols, remembers, chases and investigates. Wardenmind is that reusable layer: one profile data asset per enemy type drives ready-made nodes for both Behavior Tree and StateTree. It owns no combat, health or animation, so it plugs into a game’s existing systems, including GAS.

Key Features

• Profile-driven tuning: one data asset per enemy type holds perception, memory, attack, flee, patrol, spawn and group-aggro values, so designers retune without code
• Behavior Tree and StateTree nodes for patrol, chase, attack, flee and investigate, both calling the same decision rules so the two paths cannot drift apart
• Shared threat memory that merges sight and hearing into one state per agent, plus group aggro that relays alerts along allies as investigate leads rather than hard target locks
• Pooled wave spawner with leash, waypoint-graph coarse routing (A*), Smart Object use and EQS additions
• Mass Entity crowds that sleep as lightweight entities and awaken into full AI pawns near a player
• Per-level save/load, server-authoritative multiplayer with a small replicated presentation state, and an optional provider-agnostic LLM hook

Technical Decisions

Wardenmind layers on the engine’s own AI types instead of forking them, and keeps all decision rules in a header-only, world-free namespace shared by both authoring paths, which also makes them unit-testable (47 automation tests). Pooling, A* and graph search ship as embedded private modules so the plugin installs as a single folder, and all engine-version differences (UE 5.3–5.8) are isolated in one macro header.