AI 日报hiw3c.com

显示HN:Stemsson-一种供LLM动态扩展原生移动应用程序的语言

原文标题 · Show HN: StemJSON – a language for LLMs to extend native mobile apps on the fly
Hacker News Top stemjson.com 网页快照
正文为英文,可一键机器翻译(仅首次需要等待)

Mobile apps that build themselves

StemJSON is a new declarative language that turns descriptions into native apps - written by AI, pushed by your backend, or shaped by your users.

The language AI uses to build native apps

StemJSON is a declarative language based on JSON, well-suited for AI to generate comprehensive, performant native app UI and logic - whether building new apps or extending existing ones.

The format is open. Drop it into AI tools, prototyping environments, IDE extensions, or alongside hand-written code to simplify and accelerate native app design, development, and delivery.

Where it fits

Stem AI Area

A sandboxed area inside your app where end users describe a feature in their own words - and the runtime renders a fully native module on the spot. Apps that adapt to each user, on demand.

An LLM emits StemJSON from the user's prompt; the runtime validates it, renders it natively, and keeps it inside the sandbox you defined. No new binary, no backend round-trip for the UI.

Server-driven UI

Push complete functional modules from your backend - state, actions, expressions, navigation, validation - straight to the device.

Prototyping & design tools

From a prompt or a Figma export to a runnable native prototype. Tools emit StemJSON; the runtime renders it on-device - no glue code in between.

Figma / AI tool → StemJSON → runnable app

Alongside hand-written code

Drop StemJSON modules into a native app for the parts that change often - settings, onboarding, content feeds - and keep the rest hand-written.

StemJSON modules + native code → one app

Where to start

Read the specification

Components, actions, expressions, state, navigation and the conformance rules both runtimes enforce - so you know exactly what a module can express before you write one.

Run it in your app

Add the Swift or Kotlin SDK and render StemJSON inside an app you already ship: new screens from your backend without an app update, or a Stem AI Area where your users describe what they need and the runtime renders it natively, inside the sandbox you define. Free tier, no time limit.

Study working apps

Complete example apps you can run today.

Open-source apps with the runtime already integrated. Clone one, drop in your own .stem file, and reuse the patterns that fit what you're building.

Spotted something off in the spec? Open an issue on GitHub ↗ - feedback is the fastest way to make the next version better.