minicoder-model-config(5) File Formats Manual minicoder-model-config(5)

minicoder-model-config - AI model configuration file format for minicoder(1)

~/.config/minicoder/models.json /etc/minicoder/models.json

The minicoder command uses JSON configuration files to define available AI models and their parameters. It supports multiple model providers through OpenAI-compatible APIs, including OpenAI, Anthropic (via proxy), Google, local models via Ollama, and many others through services like OpenRouter.

Configuration files are searched in the following order:

1. User configuration: ~/.config/minicoder/models.json 2. System configuration: /etc/minicoder/models.json 3. Built-in defaults (if no configuration files exist)

The first model defined in the configuration file becomes the default model when no --model option is specified.

The configuration file must contain a JSON object where each key is a model name and the value is a model definition object.

Each model definition must include:

type (string, required)

The model type. Currently only "openai" is supported, which works with any OpenAI-compatible API including OpenAI, Ollama, OpenRouter, and many other providers.

endpoint (string, required for openai type)

The API endpoint URL for the model.

model (string, optional)

The specific model identifier to use in API requests.

api_key (string, optional)

The API key for authentication. Can be specified directly or via api_key_env.

api_key_env (string, optional)

Environment variable name containing the API key. Takes precedence over api_key.

max_context_bytes (number, optional)

Maximum context size in bytes (approximate). Default: 512000 (~128K tokens).

params (object, optional)

Additional parameters to send with API requests. Must be a valid JSON object.

For models of type "openai", the following parameters in the params object are commonly used:

stream (boolean)

Enable streaming responses. Default: true.

reasoning (object)

For models that support reasoning (e.g., o3, gemini):
•
effort: "low", "medium", or "high"

temperature (number)

Sampling temperature (0.0 to 2.0).

max_tokens (number)

Maximum tokens to generate.

Basic configuration with two models:

{
  "gpt4": {
    "type": "openai",
    "endpoint": "https://api.openai.com/v1/chat/completions",
    "model": "gpt-4-turbo",
    "api_key_env": "OPENAI_API_KEY",
    "max_context_bytes": 512000,
    "params": {
      "stream": true,
      "temperature": 0.7
    }
  },
  "local-llama": {
    "type": "openai",
    "endpoint": "http://localhost:8080/v1/chat/completions",
    "model": "llama-3.1-8b",
    "max_context_bytes": 32768,
    "params": {
      "stream": false
    }
  }
}

Configuration using OpenRouter with reasoning models:

{
  "o3": {
    "type": "openai", 
    "endpoint": "https://openrouter.ai/api/v1/chat/completions",
    "model": "openai/o3",
    "api_key_env": "OPENROUTER_API_KEY",
    "max_context_bytes": 512000,
    "params": {
      "reasoning": {
        "effort": "high"
      },
      "stream": true
    }
  },
  "gemini": {
    "type": "openai",
    "endpoint": "https://openrouter.ai/api/v1/chat/completions", 
    "model": "google/gemini-2.5-pro",
    "api_key_env": "OPENROUTER_API_KEY",
    "max_context_bytes": 2097152,
    "params": {
      "reasoning": {
        "effort": "high"
      },
      "stream": true
    }
  }
}

Configuration for Ollama (local model server):

{
  "ollama-llama": {
    "type": "openai",
    "endpoint": "http://localhost:11434/v1/chat/completions",
    "model": "llama3.1:latest",
    "max_context_bytes": 32768,
    "params": {
      "stream": true,
      "temperature": 0.7
    }
  },
  "ollama-mistral": {
    "type": "openai",
    "endpoint": "http://localhost:11434/v1/chat/completions",
    "model": "mistral:latest",
    "max_context_bytes": 32768,
    "params": {
      "stream": true
    }
  }
}

If no configuration file is found, minicoder uses built-in models based on available API keys:

The first model in the list becomes the default when no --model option is specified.

API keys can be specified via environment variables using the api_key_env field. Common variables:

OPENAI_API_KEY

For OpenAI API access

OPENROUTER_API_KEY

For OpenRouter service access

ANTHROPIC_API_KEY

For Anthropic API access

minicoder(1)

Maintained by the assist project contributors.

1980-01-01