Introduction

The RunAnywhere Swift SDK runs language, speech, and vision models on the device. Prompts, audio, and images stay on the hardware, and everything keeps working with no network. One package supplies the whole surface. You register the backends you need, call RunAnywhere.initialize(), and then reach for a namespace.

import RunAnywhere
import RunAnywhereLlamaCPP

LlamaCPP.register()
try RunAnywhere.initialize(apiKey: "", baseUrl: "", environment: .development)

let result = try await RunAnywhere.llm.generate(prompt: "Name three colours")
print(result.text)

Requirements

Item	Minimum
Platforms	iOS 17.5, macOS 14.5
Xcode	26, with Swift 6.2
Package	`runanywhere-swift` 0.20.24

The namespaces

Every capability hangs off RunAnywhere as a namespace. This is the whole API.

Namespace	What it does
`llm`	Text generation, streaming, tools, structured output
`vlm`	Image plus prompt
`stt`	Speech to text
`tts`	Text to speech
`vad`	Voice activity detection
`voice`	A full spoken conversation
`rag`	Retrieval over your own documents
`models`	Registry, download, load, unload
`embeddings`	Text vectors
`rerank`	Reorder results against a query
`lora`	Adapters on the loaded base model
`images`	Image generation
`diarization`	Who spoke when
`segmentation`	Labelled regions in an image
`solutions`	YAML-defined pipelines
`pluginLoader`	Load an engine plugin at runtime

The flat verbs from earlier releases (RunAnywhere.generate, RunAnywhere.loadModel, RunAnywhere.transcribe, and the rest) are deprecated in favour of these namespaces and will be removed. New code should use the namespaced form throughout.

Products

runanywhere-swift publishes five libraries. RunAnywhere is always required; add the backends that match the models you plan to run.

Product	Role
`RunAnywhere`	Core SDK
`RunAnywhereLlamaCPP`	llama.cpp: LLM and VLM
`RunAnywhereONNX`	Sherpa-ONNX: STT, TTS, VAD
`RunAnywhereMLX`	Apple MLX; physical device or native macOS only
`RunAnywhereNeuRT`	Apple Neural Engine

MLX reports unavailable on the simulator. MLX.register() returns false there, so a simulator build validates packaging and startup but runs no MLX inference.

Where to go next

Installation

Add the package and pick your products

Quick Start

A working chat in one file

Text generation

Prompts, conversations, and options

Voice agent

Speech in, speech out

Getting Started

Swift SDK

Kotlin SDK

React Native SDK

Flutter SDK

Web SDK

Electron SDK

Vibe Coding

Requirements

The namespaces

Products

Where to go next

Installation

Quick Start

Text generation

Voice agent

​Requirements

​The namespaces

​Products

​Where to go next

Installation

Quick Start

Text generation

Voice agent

Requirements

The namespaces

Products

Where to go next