llm-mock

A mock server for Large Language Model (LLM) APIs, supporting OpenAI, Ollama, Gemini and AWS Bedrock endpoints. Useful for testing and development without relying on actual LLM services.

Installation

1. Clone this repository:

   git clone rancher-sandbox/rancher-ai-llm-mock.git
   cd rancher-ai-llm-mock

2. Build and run with Go:

   go run ./cmd/main.go

   Or use Docker:

   docker build -t llm-mock .
   docker run -p 8083:8083 llm-mock

   Or use Helm:

   helm install llm-mock chart/llm-mock \
       --namespace your-namespace \
       --create-namespace

The server will start on http://localhost:8083

Controlling Responses

You can control the mock responses using the /v1/control endpoints:

• POST /v1/control/push: Push a mock response to the queue. Example body:

   {
   	"agent": "Rancher",
   	"text": {
   		"chunks": ["Hello", " world!"]
   	},
   	"tool": {
   		"name": "example_tool",
   		"args": [
   			{
   				"key": "value"
   			}
   		]
   	}
   }

  • The args field accepts either a single object or an array of objects (i.e. confirmation request payload for multiple resources).
  • The next model API call will stream text chunks as response and use tool for MCP invocation.
  • The MCP tool must be one of the supported tools of the agent in request.
  • If there are less than two agents configured in Rancher, the agent must not be provided.

• POST /v1/control/clear: Clear the mock response queue.

If the response queue is empty, default mock responses will be used.

See the OpenAPI spec for full API details.