Realitykit Agent Skill

RealityKit

Build AR experiences on iOS using RealityKit for rendering and ARKit for world

tracking. Covers RealityView, entity management, raycasting, scene

understanding, and gesture-based interactions. Targets Swift 6.3 / iOS 26+.

Contents

• Setup
• RealityView Basics
• Loading and Creating Entities
• Anchoring and Placement
• Raycasting
• Gestures and Interaction
• Scene Understanding
• Common Mistakes
• Review Checklist
• References

Setup

Project Configuration

• Add NSCameraUsageDescription to Info.plist
• On iOS, RealityViewCameraContent displays an AR camera view by default (iOS 18+, macOS 15+); use .virtual camera mode for explicit non-AR fallback
• No entitlement is required for basic AR. If AR is core to the app, add the arkit required-device capability; otherwise gate AR UI with isSupported.

Device Requirements

Check the exact AR configuration's isSupported value before presenting AR UI.

import ARKit

guard ARWorldTrackingConfiguration.isSupported else {
    showUnsupportedDeviceMessage()
    return
}

Key Types

| Type | Platform | Role |

|---|---|---|

| RealityView | iOS 18+, visionOS 1+ | SwiftUI view that hosts RealityKit content |

| RealityViewCameraContent | iOS 18+, macOS 15+ | Content displayed through an AR camera view on iOS, non-AR on macOS |

| Entity | All | Base class for all scene objects |

| ModelEntity | All | Entity with a visible 3D model |

| AnchorEntity | All | Tethers entities to a real-world anchor |

RealityView Basics

RealityView is the SwiftUI entry point for RealityKit.

RealityViewCameraContent is the iOS/macOS content type. On iOS, it uses an AR

camera view by default and can use content.camera = .virtual for non-AR mode

when requested or when AR/camera access is unavailable.

import ARKit
import SwiftUI
import RealityKit

struct ARExperienceView: View {
    var body: some View {
        RealityView { (content: RealityViewCameraContent) in
            if !ARWorldTrackingConfiguration.isSupported {
                content.camera = .virtual
            }

            let sphere = ModelEntity(
                mesh: .generateSphere(radius: 0.05),
                materials: [SimpleMaterial(
                    color: .blue,
                    isMetallic: true
                )]
            )
            sphere.position = [0, 0, -0.5]  // 50cm in front of camera
            content.add(sphere)
        }
    }
}

Make and Update Pattern

Use the update closure to respond to SwiftUI state changes:

struct PlacementView: View {
    @State private var modelColor: UIColor = .red

    var body: some View {
        RealityView { content in
            let box = ModelEntity(
                mesh: .generateBox(size: 0.1),
                materials: [SimpleMaterial(
                    color: .red,
                    isMetallic: false
                )]
            )
            box.name = "colorBox"
            box.position = [0, 0, -0.5]
            content.add(box)
        } update: { content in
            if let box = content.entities.first(
                where: { $0.name == "colorBox" }
            ) as? ModelEntity {
                box.model?.materials = [SimpleMaterial(
                    color: modelColor,
                    isMetallic: false
                )]
            }
        }

        Button("Change Color") {
            modelColor = modelColor == .red ? .green : .red
        }
    }
}

Loading and Creating Entities

Loading from USDZ Files

Load 3D models asynchronously to avoid blocking the main thread:

RealityView { content in
    if let robot = try? await ModelEntity(named: "robot") {
        robot.position = [0, -0.2, -0.8]
        robot.scale = [0.01, 0.01, 0.01]
        content.add(robot)
    }
}

Adding Components

Entities use an ECS (Entity Component System) architecture. Add components

to give entities behavior:

let box = ModelEntity(
    mesh: .generateBox(size: 0.1),
    materials: [SimpleMaterial(color: .red, isMetallic: false)]
)

// Make it respond to physics
box.components.set(PhysicsBodyComponent(
    massProperties: .default,
    material: .default,
    mode: .dynamic
))

// Add collision shape for interaction
box.components.set(CollisionComponent(
    shapes: [.generateBox(size: [0.1, 0.1, 0.1])]
))

// Enable input targeting for gestures
box.components.set(InputTargetComponent())

Anchoring and Placement

AnchorEntity

Use AnchorEntity to anchor content to detected surfaces or world positions:

RealityView { content in
    // Anchor to a horizontal surface
    let floorAnchor = AnchorEntity(.plane(
        .horizontal,
        classification: .floor,
        minimumBounds: [0.2, 0.2]
    ))

    let model = ModelEntity(
        mesh: .generateBox(size: 0.1),
        materials: [SimpleMaterial(color: .orange, isMetallic: false)]
    )
    floorAnchor.addChild(model)
    content.add(floorAnchor)
}

Anchor Targets

| Target | Description |

|---|---|

| .plane(.horizontal, ...) | Horizontal surfaces (floors, tables) |

| .plane(.vertical, ...) | Vertical surfaces (walls) |

| .plane(.any, ...) | Any detected plane |

| .world(transform:) | Fixed world-space position |

Raycasting

Keep RealityKit scene queries separate from ARKit real-world raycasts:

• RealityViewCameraContent.ray(through:in:to:) returns a camera ray in

RealityKit coordinate spaces. It projects a screen point into the virtual

scene; it is not proof of a detected physical surface.

• RealityViewCameraContent.hitTest(point:in:query:mask:) hits virtual

entities made hittable by CollisionComponent shapes. Use those shapes for

entity picking and targeted gestures, not ARKit plane detection.

• Use AnchorEntity(.plane(...)) for simple placement on detected planes.
• Use ARKit ARRaycastQuery plus ARSession.raycast(_:) when the task needs

a one-shot intersection with real-world surfaces, then anchor with

AnchorEntity(raycastResult:).

let results = session.raycast(query)
if let result = results.first {
    let anchor = AnchorEntity(raycastResult: result)
    anchor.addChild(model)
    content.add(anchor)
}

Do not treat entity hit tests as substitutes for ARKit surface raycasts.

Gestures and Interaction

For gesture-based entity interaction, add CollisionComponent for the hittable

shape and InputTargetComponent for input targeting.

Drag Gesture on Entities

struct DraggableARView: View {
    var body: some View {
        RealityView { content in
            let box = ModelEntity(
                mesh: .generateBox(size: 0.1),
                materials: [SimpleMaterial(color: .blue, isMetallic: true)]
            )
            box.position = [0, 0, -0.5]
            box.components.set(CollisionComponent(
                shapes: [.generateBox(size: [0.1, 0.1, 0.1])]
            ))
            box.components.set(InputTargetComponent())
            box.name = "draggable"
            content.add(box)
        }
        .gesture(
            DragGesture()
                .targetedToAnyEntity()
                .onChanged { value in
                    let entity = value.entity
                    guard let parent = entity.parent else { return }
                    entity.position = value.convert(
                        value.location3D,
                        from: .local,
                        to: parent
                    )
                }
        )
    }
}

Scene Understanding

Per-Frame Updates

Subscribe to SceneEvents.Update for continuous scene work rather than driving

RealityKit with a SwiftUI timer. Keep the subscription alive and use

event.deltaTime; see Entity Animations.

Platform Boundaries

On visionOS, ARKit provides a different API surface with ARKitSession,

WorldTrackingProvider, and PlaneDetectionProvider. These visionOS-specific

types are not available on iOS. On iOS, RealityKit handles world tracking

automatically through RealityViewCameraContent.

For iOS architecture or migration notes, gate AR with

ARWorldTrackingConfiguration.isSupported, host content with

RealityViewCameraContent, and build scenes from Entity/ModelEntity placed

with AnchorEntity.

Handoffs: CollisionComponent + InputTargetComponent handle RealityKit

interaction; AccessibilityComponent handles entity accessibility metadata;

detailed SwiftUI gestures and VoiceOver/Switch Control policy belong to siblings.

Treat existing SCNView/SCNNode work as either a separate SceneKit path or an

explicit migration to RealityKit, not a mixed scene graph.

Common Mistakes

DON'T: Skip AR capability checks

Use the configuration-specific support check from Setup before presenting AR.

Show non-AR content or an explicit unavailable state when it fails.

DON'T: Load heavy models synchronously

Loading large USDZ files on the main thread causes frame drops and hangs.

The make closure of RealityView is async -- use it.

// WRONG -- synchronous load blocks the main thread
RealityView { content in
    let model = try! Entity.load(named: "large-scene")
    content.add(model)
}

// CORRECT -- async load
RealityView { content in
    if let model = try? await ModelEntity(named: "large-scene") {
        content.add(model)
    }
}

DON'T: Forget collision and input target components for interactive entities

Interactive entities need both components shown in

Gestures and Interaction; without them, taps and

drags pass through.

DON'T: Create new entities in the update closure

The update closure runs on every SwiftUI state change. Creating entities

there duplicates content on each render pass.

// WRONG -- duplicates entities on every state change
RealityView { content in
    // empty
} update: { content in
    let sphere = ModelEntity(mesh: .generateSphere(radius: 0.05))
    content.add(sphere)  // Added again on every update
}

// CORRECT -- create in make, modify in update
RealityView { content in
    let sphere = ModelEntity(mesh: .generateSphere(radius: 0.05))
    sphere.name = "mySphere"
    content.add(sphere)
} update: { content in
    if let sphere = content.entities.first(
        where: { $0.name == "mySphere" }
    ) as? ModelEntity {
        // Modify existing entity
        sphere.position.y = newYPosition
    }
}

DON'T: Ignore camera permission

RealityKit on iOS needs camera access. If the user denies permission, the

view shows a black screen with no explanation.

// WRONG -- no permission handling
RealityView { content in
    // Black screen if camera denied
}

// CORRECT -- check and request permission
struct ARContainerView: View {
    @State private var cameraAuthorized = false

    var body: some View {
        Group {
            if cameraAuthorized {
                RealityView { content in
                    // AR content
                }
            } else {
                ContentUnavailableView(
                    "Camera Access Required",
                    systemImage: "camera.fill",
                    description: Text("Enable camera in Settings to use AR.")
                )
            }
        }
        .task {
            let status = AVCaptureDevice.authorizationStatus(for: .video)
            if status == .authorized {
                cameraAuthorized = true
            } else if status == .notDetermined {
                cameraAuthorized = await AVCaptureDevice
                    .requestAccess(for: .video)
            }
        }
    }
}

Review Checklist

• [ ] NSCameraUsageDescription set in Info.plist
• [ ] AR device capability checked before presenting AR views
• [ ] Camera permission requested and denial handled with a fallback UI
• [ ] arkit required-device capability added when AR is the app's core purpose
• [ ] 3D models loaded asynchronously in the make closure
• [ ] Entities created in make, modified in update (not created in update)
• [ ] Interaction prerequisites and sibling handoffs follow the Gestures section
• [ ] Entity hit tests, camera rays, and ARKit surface raycasts follow the Raycasting distinctions
• [ ] SceneEvents.Update subscriptions used for per-frame logic (not SwiftUI timers)
• [ ] Large scenes use ModelEntity(named:) async loading, not Entity.load(named:)
• [ ] Anchor entities target appropriate surface types for the use case
• [ ] Entity names set for lookup in the update closure

References

• Read references/realitykit-patterns.md for physics, animations, lighting, ECS, accessibility, and performance patterns.
• RealityKit framework
• RealityView
• RealityViewCameraContent
• RealityViewCamera
• Entity
• ModelEntity
• AnchorEntity
• ARKit framework
• ARKit in iOS
• Verifying Device Support and User Permission
• ARWorldTrackingConfiguration
• ARRaycastQuery
• ARSession.raycast(_:))
• Loading entities from a file