llmsim (Python)

In-process llmsim server for Python tests: a fast LLM API simulator for the OpenAI (Chat Completions, Responses, Images), Anthropic Messages, and OpenResponses APIs. No model, no network, no API key. The server is the same Rust code as the llmsim CLI, shipped as a native wheel.

pip install llmsim

Usage

from openai import OpenAI
from llmsim import Server

with Server(generator="echo", latency="instant") as sim:
    client = OpenAI(base_url=sim.openai_base_url, api_key="unused")
    reply = client.chat.completions.create(model="gpt-5", messages=[{"role": "user", "content": "hello"}])
    assert reply.choices[0].message.content == "Echo: hello"
    print(sim.stats()["total_requests"])  # 1

Anthropic:

import anthropic

with Server(latency="instant") as sim:
    client = anthropic.Anthropic(base_url=sim.anthropic_base_url, api_key="unused")

pytest fixture:

import pytest
from llmsim import Server


@pytest.fixture(scope="session")
def llm():
    with Server(latency="instant") as sim:
        yield sim

Configuration

Server(...) accepts the same settings as the llmsim serve TOML file. Sources merge in this order, later wins: config_file= or toml=, then config= (a dict shaped like the TOML), then the shortcut keywords.

Server({"errors": {"rate_limit_rate": 0.2, "server_error_rate": 0.05}})
Server(config_file="llmsim.toml", port=8080)

Properties: url, port, host, openai_base_url, anthropic_base_url, openresponses_base_url, config, is_running. Methods: start(), stop(), stats(). Server is a sync and async context manager.

See the configuration docs for every option.