Chat Completions

The Chat Completions API is the most widely supported text generation method across AI providers.

Basic Usage

Simple Text Input

string response = await "Hello, AI!"
    .GENCompletion()
    .ExecuteAsync();

With Model Selection

string response = await "Explain photosynthesis"
    .GENCompletion()
    .SetModel(OpenAIModel.GPT4o)
    .ExecuteAsync();

Input Types

1. String Input

Direct text prompt:

string response = await "What is AI?"
    .GENCompletion()
    .ExecuteAsync();

2. Message Input

Use Message object for more control:

var message = new UserMessage("Explain quantum computing");
string response = await message
    .GENCompletion()
    .ExecuteAsync();

3. Prompt Input

Use Prompt object for reusable prompts:

var prompt = new Prompt("Explain {topic}");
string response = await prompt
    .GENCompletion()
    .SetVariable("topic", "machine learning")
    .ExecuteAsync();

Configuration

Temperature

Controls randomness (0.0 = deterministic, 2.0 = very creative):

// Creative writing
string poem = await "Write a poem about stars"
    .GENCompletion()
    .SetTemperature(1.5f)
    .ExecuteAsync();

// Factual answers
string fact = await "What is 2+2?"
    .GENCompletion()
    .SetTemperature(0.0f)
    .ExecuteAsync();

Max Tokens

Limit response length:

string summary = await "Summarize the history of AI"
    .GENCompletion()
    .SetMaxTokens(100) // Short summary
    .ExecuteAsync();

System Message

Set context and behavior:

string response = await "How do I start?"
    .GENCompletion()
    .SetSystemMessage("You are a helpful Unity game development tutor")
    .ExecuteAsync();

Top P (Nucleus Sampling)

Alternative to temperature (0.0-1.0):

string response = await "Generate a character name"
    .GENCompletion()
    .SetTopP(0.9f)
    .ExecuteAsync();

Frequency Penalty

Reduce word repetition (-2.0 to 2.0):

string response = await "List 10 unique animals"
    .GENCompletion()
    .SetFrequencyPenalty(1.0f) // Avoid repetition
    .ExecuteAsync();

Presence Penalty

Encourage topic diversity (-2.0 to 2.0):

string response = await "Write about technology"
    .GENCompletion()
    .SetPresencePenalty(1.0f) // Diverse topics
    .ExecuteAsync();

Multi-Turn Conversations

Build conversation history:

var conversation = new List<Message>
{
    new SystemMessage("You are a helpful assistant"),
    new UserMessage("Hello!"),
    new AssistantMessage("Hi! How can I help?"),
    new UserMessage("What's the weather?")
};

string response = await new ChatCompletionRequest(conversation)
    .SetModel(OpenAIModel.GPT4o)
    .ExecuteAsync();

Streaming

Get real-time token-by-token responses:

await "Tell me a story"
    .GENCompletion()
    .SetModel(OpenAIModel.GPT4o)
    .StreamAsync(
        onToken: token => Debug.Log(token),
        onComplete: response => Debug.Log("Done!")
    );

Provider Support

Provider	Support	Models
OpenAI	✅ Full	GPT-4o, GPT-4, GPT-3.5
Anthropic	✅ Full	Claude 3.5, Claude 3
Google Gemini	✅ Full	Gemini 1.5, Gemini 1.0
OpenRouter	✅ Full	All models
Groq	✅ Full	Llama 3, Mixtral
xAI	✅ Full	Grok
Perplexity	✅ Full	All models
Azure OpenAI	✅ Full	GPT-4, GPT-3.5
Ollama	✅ Full	All local models

Examples

Example 1: Chatbot Response

string response = await userInput
    .GENCompletion()
    .SetModel(OpenAIModel.GPT4o)
    .SetSystemMessage("You are a friendly game NPC")
    .SetTemperature(0.8f)
    .ExecuteAsync();

npcDialogue.text = response;

Example 2: FAQ System

async UniTask<string> AnswerFAQ(string question)
{
    return await question
        .GENCompletion()
        .SetModel(OpenAIModel.GPT4oMini)
        .SetSystemMessage("Answer questions about our Unity plugin")
        .SetTemperature(0.3f)
        .SetMaxTokens(200)
        .ExecuteAsync();
}

Example 3: Contextual Help

async UniTask<string> GetContextualHelp(string topic)
{
    return await $"Explain {topic} for Unity beginners"
        .GENCompletion()
        .SetModel(OpenAIModel.GPT4o)
        .SetSystemMessage("You are a Unity tutorial assistant")
        .SetTemperature(0.5f)
        .ExecuteAsync();
}

Example 4: Translation

async UniTask<string> TranslateText(string text, string targetLang)
{
    return await $"Translate to {targetLang}: {text}"
        .GENCompletion()
        .SetModel(OpenAIModel.GPT4oMini)
        .SetTemperature(0.0f)
        .ExecuteAsync();
}

Error Handling

try
{
    string response = await "Hello"
        .GENCompletion()
        .ExecuteAsync();
}
catch (AIApiException ex)
{
    Debug.LogError($"API Error: {ex.Message}");
    // Handle: rate limit, invalid API key, model not found, etc.
}
catch (System.Net.Http.HttpRequestException ex)
{
    Debug.LogError($"Network Error: {ex.Message}");
    // Handle: no internet connection, timeout, etc.
}

Best Practices

✅ Do

• Use lower temperature (0.0-0.3) for factual answers

• Use higher temperature (0.7-1.5) for creative content

• Set reasonable max tokens to control costs

• Cache responses when possible

• Handle errors gracefully

❌ Don't

• Set temperature above 2.0 (unstable results)

• Use very high frequency/presence penalties (degrades quality)

• Forget to set system messages for context

• Ignore error handling

• Call API in tight loops without rate limiting

Next Steps

• Responses API - More advanced features

• Code Generation - Specialized for code

• Structured Output - Type-safe JSON