Skip to content

Configuration ​

All settings live in .llmockrc.json in your project root. A scaffolded project has one already; with no file, LLMock uses its built-in presets.

Presets ​

The file holds named presets under models. Each preset says which API to mimic (name), where to serve it (endpoint) and how to reply. defaultModel names the preset used when you don't pass --model:

json
{
  "defaultModel": "chatgpt",
  "models": {
    "chatgpt": {
      "name": "openai",
      "model": "gpt-4o",
      "endpoint": "chatgpt/chat/completions",
      "responseDelay": { "min": 800, "max": 2500 }
    },
    "claude": {
      "name": "claude",
      "model": "claude-opus-5-5",
      "endpoint": "v1/messages"
    }
  },
  "server": { "port": 8001 }
}
bash
llmock start                  # the defaultModel preset
llmock start --model=claude   # another preset

Only name and endpoint are required in a preset; everything else has a default. Every key, with its default, is in the config file reference.

When changes apply ​

  • The config file is read once at startup, so restart the server after changing it.
  • Fixture files and the stored responses file are read on every request, so edits to those apply straight away.
  • CLI flags override the file for one run (see the CLI reference).
  • The admin API changes rules, chaos, delay and reply settings on the running server, with no restart.

Adding custom models ​

Extend the models object with any additional preset, then start with --model=<name>:

json
{
  "models": {
    "my-model": {
      "name": "openai",
      "model": "gpt-3.5-turbo",
      "endpoint": "api/v1/chat/completions",
      "responseType": "stored",
      "validateRequests": true,
      "logRequests": false,
      "debug": true,
      "stream": false,
      "responseDelay": { "min": 1000, "max": 2000 },
      "embeddings": { "enabled": false, "dimensions": 64 }
    }
  }
}
bash
llmock start --model=my-model

To mimic a provider LLMock does not have a template for, see Creating a custom provider template.

Custom API paths ​

Set the endpoint to match any provider's path structure:

endpointURL
chatgpt/chat/completionshttp://localhost:8001/chatgpt/chat/completions
models/gemini-pro:generateContenthttp://localhost:8001/models/gemini-pro:generateContent

Write the path without a leading slash.

Environment variables ​

LLMock does not read any of these itself. They are a suggested convention for your own app, so it can switch between the mock and the real provider (the LangChain example uses them):

bash
TEST_MODE=true
TEST_BASE_URL=http://localhost:8001/chatgpt
TEST_EMBEDDING_URL=http://localhost:8001/v1/embeddings

With TEST_MODE=false your app talks to the real LLM service again.

Released under the MIT License.