Core Architecture

Documentation Unreal Engine AI Architecture

The mental model: how the plugin separates strategy from execution, and how a decision flows from perception to movement.

Two layers: strategy and execution

The plugin splits the AI into a layer that decides and a layer that acts. The decision layer picks what to do; the execution layer carries it out. The two talk through components, never directly, so you can change either side without touching the other.

Architecture

Content>Plugins>SoulslikeEnemyCombat

StateTree: Strategic Layer

Evaluates combat context
Scores available actions
Manages state transitions
Controls movement strategy
Handles target acquisition

Action Execution: Tactical Layer

Option A: Gameplay Ability (simple: play montage, apply damage)
Option B: Behavior Tree Sequence (complex: multi-step, branching)

The strategic layer drives decisions. Actions run through Gameplay Abilities or Behavior Tree sequences.

Two ways to run an action. A Gameplay Ability suits a self-contained move: play a montage, apply damage. A Behavior Tree Sequence suits a multi-step move with branches: close the gap, check stamina, commit to a heavy or light finisher. You pick per action in the ActionSet.

The pieces

An enemy is a pawn and an AI controller, each carrying components. EnemyCharacterBase and EnemyControllerBase create them for you. Getting Started covers setup, including how to reparent your own classes.

On the pawn:

Component	Job
`AbilitySystemComponent`	Runs the gameplay abilities that actions trigger
`SECMeleeTraceComponent`	Melee hit detection
`SECActionSetComponent`	Holds and replicates the action set
`SECReactionSetComponent`	Holds and replicates the reaction set
`SECCombatRoleComponent`	Replicates the assigned combat role

On the controller:

Component	Job
`SECCombatControllerComponent`	The hub: config resolution, role registration, death teardown
`SECBrainComponent`	Runs a StateTree or the native combat loop (see below)
`ActionEvaluationComponent`	Scores and runs actions
`ReactionEvaluationComponent`	Runs event-driven reactions
`MovementEvaluatorComponent`	Tactical positioning
`ThreatDetectionComponent`	Player attention tracking, disabled by default
`AwarenessComponent`	Perception memory: sight, hearing, damage

SECCombatControllerComponent is the one piece every setup needs. Add it to any AAIController and the combat system works; the other controller components are opt-in. Its full API lives in the API Reference.

How a decision flows

Every evaluation runs the same loop:

Perceive. Awareness and Threat Detection update what the AI knows about its target.
Decide. The brain builds a decision context (distance, angle, health, stamina, tags), and ActionEvaluationComponent scores every action in the set against it.
Act. The highest-scoring valid action runs, as a Gameplay Ability or a Behavior Tree sequence.
Move. MovementEvaluatorComponent scores directions in parallel and strafes, approaches, or retreats.

Architecture

Content>Plugins>SoulslikeEnemyCombat

reads AIConfig on possession

DefaultStateTree set, or empty

graph-authored

C++, no StateTree

either path drives the combat loop

scores actions through scorers and gates

executes the chosen action as

multi-step sequences

single GAS action

One brain, chosen from the config. A StateTree adds graph structure; the native loop runs the same combat cycle without one.

The brain: StateTree or native loop

SECBrainComponent reads the AI config on possession and picks how the AI runs:

A StateTree is set (DefaultStateTree): the brain creates a StateTree at runtime and runs it. Use this for graph-authored structure such as patrol, investigate, or boss phases.
No StateTree: the brain runs a native C++ loop: build the decision context, move, then score and run an action each tick. This covers combat on its own.

The scoring and movement components do the same work in both modes. See StateTree Tasks for the graph nodes, and the API Reference for the brain's properties.

Scope

The plugin provides:

Enemy AI decision-making and tactical movement
Melee hit detection, usable on player characters too
Action scoring and execution
Event-driven reaction evaluation and execution
Replicated pawn-side state (SECActionSetComponent, SECReactionSetComponent)
Multi-enemy coordination through combat roles
Multiplayer support on listen and dedicated servers
A modular component set you mix per enemy type
Data-driven configuration with full Blueprint access
A showcase enemy to study and extend

You still build:

UI and HUD: health bars, lock-on indicators, damage numbers
Level design and encounter placement
Progression and difficulty balancing