Function mockModel

Defines a programmable mock model for tests. Drive each call's response with respond, and inspect what the model was called with via lastRequest / requests / requestCount.

const model = mockModel(ai, { respond: () => ({ text: 'a summary' }) });
const out = (await ai.generate({ model, prompt: 'Summarize: ...' })).text;
assert.match(model.lastRequest!.messages.at(-1)!.content[0].text!, /Summarize/);
Copy

Streaming and tool calls are first-class:

mockModel(ai, {
  respond: (req, { sendChunk }) => {
    sendChunk('hel');
    sendChunk('lo');
    return { text: 'hello' };
  },
});

mockModel(ai, {
  respond: () => ({ toolRequests: [{ name: 'lookup', input: { id: 1 } }] }),
});
Copy

For structured output, the mock defaults to native constrained generation (supports.constrained: 'all'), so respond sees request.output.schema and no schema blob is injected into the prompt. Override with supports: { constrained: 'none' } to test the simulated path.