RunanywhereAI/runanywhere-sdks ? reverse-engineered prompt

Build me a production ready SDK for running AI completely on device, with one simple API that works across mobile, desktop, web, and server apps.

I want it to handle chat with local models, vision questions from images and camera input, speech to text, text to speech, voice conversations, embeddings for search, document Q&A, and image generation. It should choose the best available engine automatically based on the device, and clearly tell me what capabilities are actually available before I try to use them. Make the main experience feel private, offline first, and fast.

Please include working examples for iOS, Android, Swift, Kotlin, web, and a small CLI style demo if that fits. Add model download and loading flow, streaming responses, structured JSON output, and tool calling where supported. If you need current platform docs or model docs, look them up online. Also make sure the project builds cleanly, has tests, and includes simple README examples that show how to get started quickly.