Testing without a provider - swarms

mock_response makes completion and acompletion return a real ChatCompletion without sending a request. No API key is needed, so your test suite stays offline.

import routehub

messages = [{"role": "user", "content": "Approve this expense?"}]

reply = routehub.completion(
    model="gpt-5.4-mini",
    messages=messages,
    mock_response="Approved.",
)
assert reply.choices[0].message.content == "Approved."

Mock a stream

With stream=True, the mock returns a stream of chunks split on spaces. Async streams work too.

stream = routehub.completion(
    model="claude-sonnet-4-6",
    messages=messages,
    mock_response="Approved after review.",
    stream=True,
)
text = "".join(
    chunk.choices[0].delta.content or ""
    for chunk in stream
    if chunk.choices
)
assert text == "Approved after review."

Rehearse failures

Pass an exception to have it raised as given. Use this to test retry and fallback logic:

try:
    routehub.completion(
        model="gpt-5.4-mini",
        messages=messages,
        mock_response=routehub.RateLimitError("simulated"),
    )
except routehub.RateLimitError as error:
    assert error.status_code == 429

import pytest


def test_timeout_is_handled():
    with pytest.raises(TimeoutError):
        routehub.completion(
            model="gpt-5.4-mini",
            messages=messages,
            mock_response=TimeoutError("simulated outage"),
        )

Limits of mocking

The model string is not resolved, so a mock does not catch a wrong model name. Use routehub.get_llm_provider(model) to check one.
Usage counts are rough estimates.
embedding and responses have no mock_response. For embeddings, pass client= a fake OpenAI client instead.