Install (SwiftPM)¶

Add the package¶

Package.swiftXcodexcodegen

// swift-tools-version: 6.0
dependencies: [
    .package(url: "https://github.com/cagataycali/loopl.git", from: "0.1.0"),
],
targets: [
    .target(name: "MyApp", dependencies: [
        .product(name: "Loopl",         package: "loopl"),  // core: Agent, Tool, Model, Message, events
        .product(name: "LooplRuntimes", package: "loopl"),  // MLXModel, AppleFoundationModel, GGUFModel
        .product(name: "LooplUI",       package: "loopl"),  // SwiftUI components (optional)
    ]),
]

File → Add Package Dependencies… → paste https://github.com/cagataycali/loopl.git → add the Loopl, LooplRuntimes and (optionally) LooplUI products to your app target.

packages:
  Loopl:
    url: https://github.com/cagataycali/loopl.git
    from: 0.1.0
targets:
  MyApp:
    dependencies:
      - package: Loopl
        product: Loopl
      - package: Loopl
        product: LooplRuntimes

Loopl is pure Swift (Foundation only) and builds everywhere — macOS, Linux, the simulator — which is what makes the agent loop unit-testable with FakeModel. LooplRuntimes pulls in mlx-swift-lm 3.32.3 (products MLXLLM + MLXLMCommon only) and swift-transformers 1.3.4 (Tokenizers + Hub), pinned exact: in Package.swift. Platforms: iOS 18, macOS 14.

Entitlement: Increased Memory Limit¶

Add com.apple.developer.kernel.increased-memory-limit to your app's entitlements. Without it iOS jetsams an iPhone 16 Pro while loading a 1.7B 4-bit model (mlx-swift-lm #492).

MyApp.entitlements

<key>com.apple.developer.kernel.increased-memory-limit</key>
<true/>

No other capability is required. AppleFoundationModel needs nothing extra; the system asks the user nothing because the model is already on the device.

Versions that matter¶

Dependency	Pin	Why
`mlx-swift-lm`	`3.32.3` exact; link only `MLXLLM` + `MLXLMCommon`, never `MLXHuggingFace`	its macros pull `MLXFoundationModels`, which needs iOS 27 types under the default trait
`mlx-swift`	≥ 0.30.6 (transitive)	#462: gibberish output on iPhone 16 Pro before that
`swift-transformers`	`1.3.4` exact	`applyChatTemplate(messages:tools:additionalContext:)` for tool specs + `enable_thinking`

Package.swift pins these; the table is here so you know what to expect if you override them.