Structured Output / JSON Mode

llmock supports matching on response_format so you can return different responses for JSON mode requests versus regular text requests. Use match.responseFormat in fixtures or the onJsonOutput() convenience method.

How It Works

When a request includes response_format: { type: "json_object" }, the router checks match.responseFormat
Fixtures with responseFormat: "json_object" only match JSON mode requests
Regular fixtures (without responseFormat) still match JSON mode requests if no specific JSON fixture matches first

Unit Test: Programmatic API

json-output.test.ts ts

const mock = new LLMock();
await mock.start();

// Register JSON output fixture — accepts object or string
mock.onJsonOutput("json-output", { answer: 42, items: ["a", "b"] });

const res = await fetch(`${mock.url}/v1/chat/completions`, {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({
    model: "gpt-4",
    messages: [{ role: "user", content: "json-output" }],
    stream: false,
    response_format: { type: "json_object" },
  }),
});

const body = await res.json();
const content = JSON.parse(body.choices[0].message.content);
expect(content.answer).toBe(42);
expect(content.items).toEqual(["a", "b"]);

JSON Fixture

fixtures/json-mode.json json

{
  "fixtures": [
    {
      "match": {
        "userMessage": "json-output",
        "responseFormat": "json_object"
      },
      "response": {
        "content": "{\"answer\":42,\"items\":[\"a\",\"b\"]}"
      }
    }
  ]
}

Match Behavior

Request response_format	Fixture responseFormat	Match?
{ type: "json_object" }	"json_object"	Yes
{ type: "json_object" }	(not set)	Yes (fallthrough)
(not set)	"json_object"	No
(not set)	(not set)	Yes

The onJsonOutput() method accepts either a plain object (auto-serialized) or a string. This makes it easy to return structured data without manual JSON.stringify.