Openai reference

Chat Completions client with JSON and SSE results built on Fetcher and decorators. No other provider product API is exposed by this package.

Install

pnpm add @ahoo-wang/fetcher-openai @ahoo-wang/fetcher @ahoo-wang/fetcher-eventstream @ahoo-wang/fetcher-decorator

The command includes OpenAI → Decorator/EventStream → Fetcher. Prebuilt clients do not require decorator syntax in your application; custom decorated classes do. Streaming additionally needs the EventStream runtime APIs.

Version 5.0.0 declares Node >=18.20.8 for consumers. Repository development requires Node >=20.20.2 and pnpm 10.34.5.

Minimal example

import { OpenAI, type OpenAIOptions } from '@ahoo-wang/fetcher-openai';

export async function answer(options: OpenAIOptions, model: string) {
  const client = new OpenAI(options);
  const result = await client.chat.completions({
    model,
    messages: [{ role: 'user', content: 'Say hello.' }],
    stream: false,
  });
  return result.choices[0]?.message?.content ?? '';
}

Choose an entry point

Use OpenAI when the SDK should create a Bearer-authenticated Fetcher; use ChatClient({ fetcher }) to reuse an application proxy or an existing transport policy. Use stream: false (or omit it) for one JSON response and literal stream: true for SSE events. Use the underlying Fetcher plus CompletionStreamResultExtractor when per-call cancellation is required.

Topics

• Client and chat completions
• Streaming chat completions

Complete public symbol index

Supply a server-side credential and a compatible base URL ending at the API prefix (for example /v1). The request targets /v1/chat/completions; caller code handles rejection. A complete streaming flow is in the Chat guide.