@stewmore/expo-ai-cloud

The cloud fallback client adapter (pure TypeScript). It POSTs to a configurable backend and streams via expo/fetch, giving the router a reliable path when no on-device provider is available — and an explicit one for prompts that aren't marked sensitive.

npm install @stewmore/expo-ai-cloud

Configure the endpoint

import { configureCloud } from '@stewmore/expo-ai-cloud';
import { fetch as expoFetch } from 'expo/fetch';

configureCloud({ endpoint: 'http://localhost:8787', fetch: expoFetch });

export type CloudProviderConfig = {
  endpoint: string;
  headers?: Record<string, string>;
  provider?: 'openai' | 'gemini' | 'anthropic' | 'bedrock' | 'custom';
};

Fallback behavior

await ExpoAI.generate({
  prompt: 'Extract the risks from this proposal.',
  fallback: 'cloud',
});

The router only reaches the cloud when fallback is allowed and the prompt isn't gated by the privacy model. Every result still carries provider and privacy metadata — cloud responses report third-party-cloud.

The repo ships a reference cloud server that implements this wire protocol and runs in mock mode by default, so the cloud examples work offline.