Agent Settings

AgentSettings is a ScriptableObject that defines all persistent configuration for your agent.

Creating Agent Settings

1. Right-click in Project window

2. Select Create > AI Dev Kit > Agent Settings

3. Name it descriptively (e.g., "CustomerSupportAgent")

Core Settings

Identification

// Unique identifier (auto-generated)
public string Id { get; }

// Human-readable name
public string Name { get; }

Chat Service API

Choose which API type to use:

public ChatService ChatServiceApi { get; set; }

Options:

• Default - Auto-selects best API

• AssistantsApi - OpenAI Assistants

• RealtimeApi - Low-latency voice

Model Configuration

public string Model { get; set; }
public Temperature Temperature { get; set; }
public TopP TopP { get; set; }
public TokenCount MaxTokens { get; set; }
public Seed Seed { get; set; }

Example:

settings.Model = "gpt-4o";
settings.Temperature = 0.7f;
settings.MaxTokens = 2000;

Instructions (System Prompt)

public string Instructions { get; }

Define agent's behavior and personality:

You are a helpful customer support assistant.
Be friendly, professional, and concise.
Always ask for clarification if needed.

Agent Profile

Configure the agent's identity:

[SerializeField] private AgentProfile agentProfile;

public string Name { get; }
public string Description { get; }
public string StartingMessage { get; }

Create via Create > AI Dev Kit > Agent Profile.

User Profile

Configure user information:

[SerializeField] private UserProfile userProfile;

public string UserName { get; }
public string UserOccupation { get; }

Audio Settings

Input Audio

public bool EnableInputAudio { get; set; }
public TranscriptionParameters InputAudioParameters { get; }
public SystemLanguage InputAudioLanguage { get; }

Output Audio

public bool EnableOutputAudio { get; set; }
public SpeechParameters OutputAudioParameters { get; }
public string Voice { get; }

Example:

settings.EnableOutputAudio = true;
settings.OutputAudioParameters.Voice = "alloy";
settings.OutputAudioParameters.Speed = 1.0f;

Tool Configuration

Tool Definitions

public List<ToolDefinitionBase> ToolDefinitions { get; }

Add tools in Inspector or code:

var imageTool = ScriptableObject.CreateInstance<ImageGenerationToolDefinition>();
imageTool.Model = "dall-e-3";
settings.ToolDefinitions.Add(imageTool);

Tool Parameters

public int? MaxToolCalls { get; }
public bool ParallelToolCalls { get; set; }

Advanced Settings

Reasoning

public bool EnableReasoning { get; set; }
public ReasoningOptions ReasoningOptions { get; }

For models that support reasoning (o1, o3):

settings.EnableReasoning = true;
settings.ReasoningOptions.Effort = ReasoningEffort.High;

Moderation

public bool EnableModeration { get; set; }

Enable content moderation:

settings.EnableModeration = true;

Memory

public AgentMemorySettings Memory { get; }

Configure conversation memory:

settings.Memory.MaxContextMessages = 50;
settings.Memory.EnableSummarization = true;

Safety (Google)

public List<SafetySetting> SafetySettings { get; }

For Google Gemini models:

settings.SafetySettings.Add(new SafetySetting
{
    Category = HarmCategory.HateSpeech,
    Threshold = HarmBlockThreshold.BlockMediumAndAbove
});

Assistants API Options

public AssistantsApiSettings AssistantsApiOptions { get; }

Configure OpenAI Assistants:

settings.AssistantsApiOptions.AssistantId = "asst_xxxxx";
settings.AssistantsApiOptions.RunPollingIntervalMs = 2000;

Complete Example

using UnityEngine;
using Glitch9.AIDevKit.Agents;

[CreateAssetMenu(fileName = "NewAgentSettings", menuName = "Examples/Agent Settings")]
public class CustomAgentSettings : ScriptableObject
{
    public static AgentSettings CreateDefault()
    {
        var settings = CreateInstance<AgentSettings>();
        
        // Core settings
        settings.ChatServiceApi = ChatService.Default;
        settings.Model = "gpt-4o";
        settings.Temperature = 0.7f;
        settings.MaxTokens = 2000;
        
        // Audio
        settings.EnableOutputAudio = true;
        settings.OutputAudioParameters = new SpeechParameters
        {
            Voice = "alloy",
            Speed = 1.0f
        };
        
        // Tools
        settings.ParallelToolCalls = true;
        
        // Memory
        settings.Memory.MaxContextMessages = 50;
        settings.Memory.EnableSummarization = true;
        
        return settings;
    }
}

Runtime Modification

Some settings can be changed at runtime:

// Can change
agent.Model = "gpt-4o-mini";
agent.Temperature = 0.9f;
agent.Voice = "nova";

// Cannot change (require reinitialize)
// settings.ChatServiceApi
// settings.EnableInputAudio
// settings.ToolDefinitions

Validation

Settings are validated on initialization:

try
{
    await agent.InitializeAsync();
}
catch (ArgumentException ex)
{
    Debug.LogError($"Invalid settings: {ex.Message}");
}

Common validation errors:

• Missing or invalid model ID

• Incompatible feature combinations

• Invalid API key

• Missing required tool definitions

Best Practices

1. Create Presets

Create different settings for different purposes:

• FastResponse.asset - GPT-3.5, low temperature

• Creative.asset - GPT-4, high temperature

• VoiceAssistant.asset - Realtime API, audio enabled

2. Use Profiles

Separate personality (AgentProfile) from functionality (AgentSettings):

Settings/
├─ SupportAgent.asset (settings)
└─ Profiles/
   ├─ FriendlyHelper.asset
   └─ ProfessionalAdvisor.asset

3. Version Control

Keep settings in version control but exclude API keys.

4. Environment-Specific

Use different settings for dev/prod:

Settings/
├─ Dev/
│  └─ TestAgent.asset
└─ Prod/
   └─ ProductionAgent.asset

Next Steps

• Chat Service API Types - Choose API

• Audio Setup - Configure voice I/O

• Memory Settings - Manage context