RunAnywhere Files

RunAnywhere SDKs v0.18.0

Privacy-first, on-device AI SDKs for iOS, Android, Flutter, and React Native.

What's New

Tool Calling (Function Calling)

• Register custom tools (weather, calculator, etc.) with typed parameters
• Auto-execute loop: generate → parse → execute → loop
• Multi-format support: default JSON + LFM2 (Liquid AI models)
• Actor-based thread-safe tool registry
• C++ parsing via <tool_call> tag detection (single source of truth)

VLM (Vision Language Model)

• Describe and analyze images with on-device AI
• Multiple image input formats: file path, RGB pixels, base64, UIImage, CVPixelBuffer
• Streaming support with real-time token output
• Camera and photo library integration in example app
• llama.cpp backend with mmproj (multimodal projection) support

Diffusion (Image Generation)

• On-device image generation via CoreML + Apple Neural Engine
• 6 model variants: SD 1.5, SD 2.1, SDXL, SDXL Turbo, SDXS (1-step), LCM
• 3 modes: text-to-image, image-to-image, inpainting
• 8 schedulers: DPM++ 2M Karras, DDIM, Euler, Euler Ancestral, PNDM, LMS, and more
• Progress streaming with intermediate image previews
• Automatic tokenizer download from HuggingFace

Structured Output

• Type-safe JSON generation via Generatable protocol
• Streaming support with token-by-token display + final typed result
• C++ JSON extraction for reliability

Apple Foundation Models (iOS 26+)

• Integration with Apple Intelligence built-in models
• 4096 token context window
• Automatic device eligibility checking

Installation (Swift Package Manager)

:::swift
dependencies: [
    .package(url: "https://github.com/RunanywhereAI/runanywhere-sdks", from: "0.18.0")
]

Assets