Creating Your First Agent

Set up your first Agent in Unity from scratch in a few minutes

This guide walks you through creating a working Agent step by step. By the end you will have an Agent in your scene that can receive text messages and stream responses back to your code.

Make sure you have already set up your API key before continuing.

---

Step 1 - Create an AgentProfile

The AgentProfile defines your agent's identity and personality.

1. In the Project window, right-click a folder and choose: Create > AI Dev Kit > Agent > Agent Profile

2. Name the asset (e.g. MyAgentProfile)

3. Fill in the Inspector fields:

Field	Description
Name	The agent's display name (e.g. Alex)
Instructions	System prompt - tell the agent who it is and how to behave
Conversation Starter	(Optional) First message the agent sends when a conversation begins
Traits	(Optional) Personality toggles: Witty, Encouraging, Direct, etc.

Example instructions:

You are Alex, a friendly game companion. Help the player with quests,
answer lore questions, and keep your responses concise.

---

Step 2 - Create an AgentSettings

The AgentSettings defines the technical configuration.

1. Right-click and choose: Create > AI Dev Kit > Agent > Agent Settings

2. Name the asset (e.g. MyAgentSettings)

3. In the Inspector, configure:

Field	Description
Chat API Type	ChatCompletions for most use cases
Model	Select a model (e.g. gpt-4o-mini)
Instructions section	Assign your AgentProfile asset here

Leave all other settings at their defaults for now.

---

Step 3 - Add AgentBehaviour to Your Scene

1. Create an empty GameObject (e.g. name it Agent)

2. Click Add Component and search for AgentBehaviour

3. Assign your AgentSettings asset to the Settings field

You don't need to assign an audio recorder or player unless you want voice I/O.

---

Step 4 - Send a Message from Code

Create a MonoBehaviour that holds a reference to AgentBehaviour:

using Glitch9.AIDevKit.Agents;
using Cysharp.Threading.Tasks;
using UnityEngine;

public class ChatExample : MonoBehaviour
{
    [SerializeField] private AgentBehaviour agentBehaviour;

    private async void Start()
    {
        // Wait for the agent to finish initializing
        await agentBehaviour.Agent.InitializeAsync();

        // Send a message and await the full response
        var result = await agentBehaviour.Agent.SendAsync("Hello! Who are you?");

        // Print the response text
        Debug.Log(result.Value.GetText());
    }
}

Attach this script to any GameObject and wire up the AgentBehaviour reference in the Inspector.

---

Step 5 - Handle Streaming Responses (optional)

For a better user experience, show text as it streams rather than waiting for the complete response:

using Glitch9.AIDevKit.Agents;
using Glitch9.AIDevKit;
using Cysharp.Threading.Tasks;
using UnityEngine;
using UnityEngine.UI;

public class StreamingChatExample : MonoBehaviour
{
    [SerializeField] private AgentBehaviour agentBehaviour;
    [SerializeField] private Text responseText;

    private void OnEnable()
    {
        agentBehaviour.Agent.RegisterCallback<Delta<ITextChunk>>(OnTextDelta);
    }

    private void OnDisable()
    {
        agentBehaviour.Agent.UnregisterCallback<Delta<ITextChunk>>(OnTextDelta);
    }

    private void OnTextDelta(Delta<ITextChunk> delta)
    {
        // Append each token chunk to your UI text
        responseText.text += delta.Value.Text;
    }

    public async void OnSendButtonClicked(string userMessage)
    {
        responseText.text = "";
        await agentBehaviour.Agent.SendAsync(userMessage);
    }
}

Streaming must be enabled in AgentSettings > Model Parameters > Stream for OnTextDelta events to fire.

---

AutoInit vs Manual Init

AgentBehaviour exposes an AutoInit option in the Inspector:

Mode	When to use
AutoInit = On (default)	Agent initializes when the scene starts. Safe for most projects.
AutoInit = Off	You call Agent.InitializeAsync() yourself. Useful when you need to configure dependencies at runtime before the agent starts.

---

What's Next

• Learn what happens internally when a message is sent -> How Agent Works

• Persist conversation history across sessions -> Saving & Loading

• Add function calling (tools) -> Tools section

• Enable voice input/output -> Audio section