Agent Configuration Guide

Learn how to configure your AI agents in AI Studio Plus. This guide covers all the essential fields and options for creating powerful, customized agents.

Overview

Agent configuration files are JSON documents that define your agent's behavior, capabilities, and interface. They tell AI Studio Plus:

What your agent does
What inputs it needs
What outputs it produces
How it should execute tasks
How users interact with it

Basic Structure

Every agent configuration has three main sections:

{
  "metadata": { /* Agent info visible to users */ },
  "execution": { /* How your agent runs */ },
  "schemas": { /* Input/output structure */ }
}

Metadata Section

The metadata section contains information displayed to users in the marketplace and agent details page.

Required Fields

name (string) The display name of your agent. Keep it clear and descriptive.

"name": "Daily Motivation Generator"

description (string) A brief description of what your agent does. This appears in search results.

"description": "Get personalized motivational quotes every morning"

category (string) Choose from available categories to help users discover your agent.

Available categories:

productivity - Task management, time tracking, scheduling
health_fitness - Wellness, exercise, nutrition
image - Image generation and processing
video - Video creation and editing
audio - Music and sound generation
code - Programming and development
data - Analytics and data processing
research - Information gathering and analysis
marketing - Content creation and promotion
writing - Text generation and editing
assistant - General purpose helpers
education - Learning and teaching
finance - Money management and tracking
entertainment - Games and fun
other - Everything else

"category": "productivity"

Optional Fields

logo (string, URL) A square image (recommended 512x512px) displayed as your agent's avatar.

"logo": "https://example.com/agent-logo.png"

tags (array of strings) Keywords to improve discoverability.

"tags": ["motivation", "productivity", "daily", "wellness"]

version (string) Track your agent's version for updates.

"version": "1.0.0"

accessLevel (string) Control who can use your agent:

public - Anyone can discover and use (default)
private - Only you can use
unlisted - Anyone with the link can use, but not in marketplace

"accessLevel": "public"

author (object) Information about the creator (automatically populated).

"author": {
  "name": "John Doe",
  "email": "john@example.com"
}

Execution Section

This section defines how your agent runs and processes requests.

Required Fields

model (string) The AI model your agent uses.

Available models:

gpt-4o - Most capable, best for complex tasks
gpt-4o-mini - Fast and efficient, good for simple tasks
claude-3-5-sonnet - Excellent for creative writing
claude-3-5-haiku - Quick responses, good for chat
gemini-2.0-flash-exp - Experimental, fast inference

"model": "gpt-4o-mini"

systemPrompt (string) The core instructions that define your agent's behavior. This is the most important field!

Tips for writing system prompts:

Be specific about the agent's role
Define the tone and style
Include any rules or constraints
Provide examples of desired output

"systemPrompt": "You are a motivational coach who provides uplifting, personalized quotes based on the user's current mood and goals. Always be positive, encouraging, and specific to their situation."

Optional Fields

temperature (number, 0-2) Controls randomness in responses:

0 - Deterministic, same input = same output
0.7 - Balanced (default)
1.5 - Very creative and varied

"temperature": 0.7

maxTokens (number) Maximum length of generated responses.

"maxTokens": 500

tools (array) External tools your agent can use:

"tools": [
  {
    "type": "web_search",
    "enabled": true
  },
  {
    "type": "image_generation",
    "enabled": true,
    "provider": "dall-e-3"
  }
]

Available tools:

web_search - Search the internet
image_generation - Create images
code_execution - Run code
file_upload - Accept file uploads

responseFormat (string) How the agent should format responses:

text - Plain text (default)
markdown - Formatted text with markdown
json - Structured JSON output

"responseFormat": "markdown"

Schemas Section

Define the structure of inputs and outputs.

Input Schema

Describes what information users need to provide.

"inputSchema": {
  "type": "object",
  "properties": {
    "mood": {
      "type": "string",
      "description": "Your current mood",
      "enum": ["happy", "stressed", "motivated", "tired"]
    },
    "goal": {
      "type": "string",
      "description": "What you want to achieve today"
    }
  },
  "required": ["mood"]
}

Output Schema

Describes what your agent returns.

"outputSchema": {
  "type": "object",
  "properties": {
    "quote": {
      "type": "string",
      "description": "Motivational quote"
    },
    "action": {
      "type": "string",
      "description": "Suggested action to take"
    }
  }
}

Schema UI Hints

Add hints to improve the user interface:

"properties": {
  "description": {
    "type": "string",
    "ui:widget": "textarea",
    "ui:placeholder": "Describe your task..."
  }
}

Available UI widgets:

textarea - Multi-line text input
select - Dropdown menu
checkbox - Boolean toggle
file - File upload
date - Date picker

Complete Example

Here's a complete agent configuration:

{
  "metadata": {
    "name": "Daily Motivation Generator",
    "description": "Get personalized motivational quotes every morning",
    "category": "productivity",
    "logo": "https://example.com/motivation-logo.png",
    "tags": ["motivation", "productivity", "daily", "wellness"],
    "version": "1.0.0",
    "accessLevel": "public"
  },
  "execution": {
    "model": "gpt-4o-mini",
    "systemPrompt": "You are a motivational coach who provides uplifting, personalized quotes based on the user's current mood and goals. Always be positive, encouraging, and specific to their situation. End each response with a simple action they can take right now.",
    "temperature": 0.8,
    "maxTokens": 300,
    "responseFormat": "markdown"
  },
  "schemas": {
    "inputSchema": {
      "type": "object",
      "properties": {
        "mood": {
          "type": "string",
          "description": "How are you feeling right now?",
          "enum": ["happy", "stressed", "motivated", "tired", "anxious"],
          "ui:widget": "select"
        },
        "goal": {
          "type": "string",
          "description": "What do you want to achieve today?",
          "ui:widget": "textarea",
          "ui:placeholder": "e.g., finish my project, exercise, be more patient"
        }
      },
      "required": ["mood"]
    },
    "outputSchema": {
      "type": "object",
      "properties": {
        "quote": {
          "type": "string",
          "description": "Personalized motivational quote"
        },
        "action": {
          "type": "string",
          "description": "One simple action to take"
        },
        "affirmation": {
          "type": "string",
          "description": "Positive affirmation"
        }
      }
    }
  }
}

Best Practices

1. Write Clear System Prompts

Your system prompt is your agent's personality and instructions. Make it:

Specific - Define exactly what the agent should do
Actionable - Give clear guidelines for behavior
Contextual - Include relevant background information

2. Design User-Friendly Inputs

Use descriptive field names
Provide helpful descriptions
Add UI hints for better user experience
Only require essential fields

3. Choose the Right Model

gpt-4o for complex reasoning and analysis
gpt-4o-mini for quick, simple tasks
claude-3-5-sonnet for creative writing
gemini-2.0-flash-exp for experimental features

4. Test Thoroughly

Before publishing:

Test with various inputs
Check edge cases
Verify outputs match expectations
Get feedback from others

5. Iterate and Improve

Monitor user feedback
Update based on usage patterns
Refine prompts for better results
Add new features over time

Validation

AI Studio Plus validates your configuration before allowing publication. Common errors:

Invalid JSON Ensure your JSON is properly formatted with matching brackets and quotes.

Missing Required Fields All required fields in metadata, execution, and schemas must be present.

Invalid Schema Schemas must follow JSON Schema specification (Draft 7).

Model Not Available Use only supported model names.

Next Steps

See Example Configurations for real-world examples
Review JSON Schema Reference for detailed schema options
Create Your First Agent in AI Studio Plus

Need Help?

Join our Discord Community for support
Check Common Issues in troubleshooting
Contact support at support@aisp.app

Overview​

Basic Structure​

Metadata Section​

Required Fields​

Optional Fields​

Execution Section​

Required Fields​

Optional Fields​

Schemas Section​

Input Schema​

Output Schema​

Schema UI Hints​

Complete Example​

Best Practices​

1. Write Clear System Prompts​

2. Design User-Friendly Inputs​

3. Choose the Right Model​

4. Test Thoroughly​

5. Iterate and Improve​

Validation​

Next Steps​

Need Help?​

Overview

Basic Structure

Metadata Section

Required Fields

Optional Fields

Execution Section

Required Fields

Optional Fields

Schemas Section

Input Schema

Output Schema

Schema UI Hints

Complete Example

Best Practices

1. Write Clear System Prompts

2. Design User-Friendly Inputs

3. Choose the Right Model

4. Test Thoroughly

5. Iterate and Improve

Validation

Next Steps

Need Help?