What it does npx rn-ai generates a full-stack project: a React Native mobile app plus an Express server that proxies requests to OpenAI, Anthropic, and Gemini. It handles streaming text responses, image generation via Gemini’s “Nano Banana” models, and comes with five baked-in themes including one called “hackerNews” that probably renders everything in orange.

The interesting bit The server isn’t decorative. It’s a deliberate proxy layer so you can add authentication and authorization without exposing API keys in mobile binaries—a problem many AI demos ignore. The README even walks you through adding new LLM providers by copy-pasting streaming handlers, which is either pragmatic documentation or a quiet admission that most of these APIs work the same way now.

Key highlights

• Supports GPT-5.2, Claude Opus/Sonnet/Haiku 4.5, and Gemini (per README; model versions are as stated)
• Streaming responses from all providers, including image generation
• Express server with auth hook points, not just a client-side SDK wrapper
• Theming system: five themes included, extensible in ~10 lines of TypeScript
• Video tutorial linked for the “watch instead of read” crowd

Caveats

• Adding a new LLM requires touching five files across app and server; the framework helps but doesn’t abstract away provider differences
• “GPT-5.2” and “Claude 4.5” are unusual version strings; unclear if these are typos, future-dated, or specific API designations
• Image generation is Gemini-only out of the box; other providers need manual server wiring

Verdict Worth a spin if you’re building a cross-platform AI app and want the server scaffolding done for you. Skip it if you need deep native module customization or were hoping for a fully provider-agnostic abstraction.