Reference cloud server

A tiny Hono backend implementing the wire protocol the @stewmore/expo-ai-cloud adapter speaks. Runs in mock mode by default (no API keys), so the cloud-fallback and structured-output examples work offline.

Run

npm run server                       # from the repo root
# or
npm run dev --workspace @stewmore/expo-ai-server

Listens on http://localhost:8787 (override with PORT).

Modes

Env	Mode
(none)	mock — canned replies; schema requests return a schema-shaped sample
OPENAI_API_KEY	proxy OpenAI-compatible /chat/completions (OPENAI_BASE_URL, OPENAI_MODEL)
ANTHROPIC_API_KEY	proxy Anthropic /v1/messages (ANTHROPIC_MODEL)
EXPO_AI_FORCE_MOCK=1	force mock even if keys are set

OPENAI_BASE_URL lets you point at any OpenAI-compatible server (OpenRouter, Ollama, LM Studio, a local proxy, …).

Endpoints

GET  /health      -> { ok: true, mode }
POST /v1/generate -> { text, finishReason, usage }
POST /v1/stream   -> SSE: data: {"type":"delta","text":"…"} … data: {"type":"done",…} … data: [DONE]

Request body: { prompt, instructions?, temperature?, maxOutputTokens?, schema?, schemaName? }.

Point the app at it

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

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

note

On a physical device use your machine's LAN IP instead of localhost.