Overview

AgentMark provides extensive configuration options that give application developers full control over model selection, configuration, and behavior. All configuration is managed through the agentmark.json file in your project root.

These options allow you to customize your AgentMark dashboard to your needs.

​

Initializing the agentmark.json file

You can initialize the agentmark.json file by running the following command:

npx @agentmark/cli@latest init

​

The agentmark.json File

The agentmark.json file is the central configuration file for your AgentMark implementation. It consists of two main components:

{
  "agentmarkPath": "/",
  "version": "2.0.0",
  "mdxVersion": "1.0",
  "builtInModels": ["gpt-4"],
  "modelSchemas": {
    // Model configurations go here
  }
}

​

agentmarkPath

The agentmarkPath property defines the base directory where AgentMark will look for your application’s resources including:

• Prompt templates
• Evaluation configurations
• Datasets
• Other AgentMark-specific resources

By default, this is set to ”/”, meaning AgentMark will look in your project root. You can modify this path if you want to organize these resources in a different directory structure.

​

version

The version property is used to specify the version of the AgentMark configuration. This is useful for tracking changes and ensuring compatibility between different versions of your application.

​

mdxVersion

The mdxVersion property is used to specify the version of the MDX format. This is useful for tracking changes and ensuring compatibility between different versions of the MDX format.

​

builtInModels

The builtInModels property specifies which models are supported by AgentMark Cloud and available for use in your application. These models are pre-configured and ready to use without additional setup.

​

modelSchemas

The modelSchemas object is where you define custom models that are not already built-in to AgentMark. This is useful when you want to:

• Add support for custom or proprietary models
• Configure models from providers not natively supported
• Customize the UI and settings for specific model implementations

Each model entry contains:

• A unique identifier (e.g., “gpt-4”, “claude-3-5-sonnet-latest”)
• Display configuration (labels, descriptions)
• Model-specific settings and their UI controls
• Parameter constraints and defaults