llmao

How to mock the OpenAI SDK in Vitest

Replace the openai package with llmao, a fake LLM that implements the same client, and control it from the test. No API key, no network, no changes to your app.

Install

npm install --save-dev llmao

Mock the module

import { beforeEach, expect, test, vi } from 'vitest';
import * as llmao from 'llmao/testing';
import { classify } from './support'; // calls new OpenAI().chat.completions.create()

vi.mock('openai', () => import('llmao/openai'));

beforeEach(() => llmao.reset());

test('classifies shipping tickets', async () => {
  llmao.configure({ script: [{ when: /never arrived/i, text: 'shipping' }] });

  await expect(classify('My order never arrived')).resolves.toBe('shipping');
  expect(llmao.calls).toHaveLength(1);
  expect(llmao.lastCall()?.prompt).toBe('My order never arrived');
});

vi.mock is hoisted, so it applies before your app imports openai. Importing llmao/testing turns on test mode: every client answers instantly, never hallucinates, uses what you pass to configure(), and records its calls.

Streaming

Streaming works like the real API, so you can test code that renders tokens as they arrive:

const stream = await client.chat.completions.create({ model: 'gpt-4o', stream: true, messages });
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta.content ?? '');
}

Fake timers and latency

Test mode creates no timers, so vi.useFakeTimers() works as is. To test a loading state or your own timeout, turn the latency back on and move the clock:

vi.useFakeTimers();
llmao.configure({ speed: 'realistic', script: [{ text: 'shipping' }] });

const result = classify('where is my order');
await vi.advanceTimersByTimeAsync(100);
// the model is still "thinking" here
await vi.advanceTimersByTimeAsync(60_000);
await expect(result).resolves.toBe('shipping');

Related

← llmao · Markdown