A hosted alternative to local LLM mock servers like mockllm
Packages such as mockllm, llmock, fakellm and stubllm start a local server that imitates the OpenAI and Anthropic APIs, so tests can run without a real key or token bill. mockllm, for example, is an Apache-2.0 Python package that answers from a YAML file of canned responses; llmock focuses on injecting errors to test retries and fallbacks.
At a glance
local LLM mock servers compared with this playground, checked 2026-11-09.
They have to be installed, configured and kept running wherever the test runs — a developer laptop, every CI job, a classroom of machines — and each has its own configuration format. That is the right model for an application's own test suite and an obstacle for a quick experiment, a tutorial or a browser-based exercise.
What is different here
The Mock LLM API here is hosted: point an SDK at https://funapi.dev/api/llm/v1 (or /api/llm for Anthropic) with any key. The model name selects the scenario — mock-429, mock-503, mock-slow, mock-stream-cut, mock-malformed-json, mock-tool-call — so a test chooses its failure in the request instead of in a config file. Streaming uses SSE in both providers' shapes.
What local LLM mock servers does better
Runs offline, inside your network, with nothing leaving the machine.
Fully configurable responses — any text, any rule — matched to your prompts.
No dependency on a third-party service in CI.
What this playground does better
Nothing to install or run: one base URL.
Failure scenarios selected per request by model name.
OpenAI and Anthropic shapes, SSE streaming, tool calls and embeddings in one place.
Usable from a browser exercise or a shared tutorial.
Which to use
Run a local mock for your application's own CI and anything private. Use the hosted mock to learn, demonstrate or quickly test streaming, retries and tool calls without installing anything.
Checked against local LLM mock servers as of 2026-11-09.
Official sources and scope
This compares a training playground with the documented use case of the other tool. Plans, limits and hosted features can change. For current terms, use the official sources below.
JSONPlaceholder — JSONPlaceholder fakes its writes. If you need a POST that a later GET can find, that is the difference.
httpbin — httpbin shows you the request. This shows you an API. They answer different questions.
ReqRes — Choose between a hosted test backend with account-backed collections and a no-signup training playground with themed scenarios.
json-server — json-server turns a JSON file into a REST API in seconds. It does not push back — and that is the difference.
WireMock — WireMock reproduces the API you already know, exactly. This one gives you an API you do not control.
Mockoon — Mockoon is for designing your own mock API. This one is an API already designed, with rules you have to discover.
Beeceptor — Beeceptor gives you a hosted endpoint to inspect and mock. This gives you finished APIs that behave like real ones.
DummyJSON — DummyJSON has rich sample data and a JWT login. Its writes are simulated; here they persist.
Postman mock servers — A Postman mock server replays the examples saved in a collection. This one runs an API with state and rules.
PokeAPI — PokeAPI is the real Pokémon database, read-only. This Pokémon API is invented data you can write to and break.
SWAPI — SWAPI serves Star Wars data, read-only. This Star Wars API is invented practice data you can change.
Restful Booker — Restful Booker is one booking API built for testing practice. This is 55 APIs with the failure paths built in.
Swagger Petstore — Petstore is the reference sample for an OpenAPI document. This is a set of APIs whose behaviour is the thing you test.
GoRest — GoRest needs a token from a social login before you can write. This needs no account at all.
FakeStoreAPI — FakeStoreAPI returns a fake id for your POST and saves nothing. Here the cart, coupon and checkout keep their state.
Practice Software Testing — Toolshop is a full web shop with seeded bugs for UI and API testing. Fun API is many small APIs built around HTTP behaviour.
balldontlie — balldontlie serves real league data behind an API key and a free tier of a few requests a minute. This serves invented leagues you can drive yourself.
the HAPI FHIR public test server — The HAPI public server is a full FHIR implementation shared by everyone and purged regularly. This is a small, isolated, synthetic subset for practising tests.
the National Weather Service API — api.weather.gov is real, free US government weather data. This is a deterministic weather API for tests that need the same forecast every time.
the free mock API landscape — Six free options, what each is genuinely for, and how to tell which one your problem needs.