openai

OpenAI LLM Example

This example demonstrates how to use the OpenAI client from the Agent SDK.

Features

• Direct text generation using the Generate method
• Chat completion using the Chat method
• Configuration options for model parameters

Usage

Prerequisites

• Set the OPENAI_API_KEY environment variable with your OpenAI API key

export OPENAI_API_KEY=your_api_key_here

Running the Example

go run main.go

Code Explanation

Creating the Client

client := openai.NewClient(
    apiKey,
    openai.WithModel("gpt-4o-mini"), // Optional: specify model
)

Text Generation

response, err := client.Generate(
    context.Background(),
    "Write a haiku about programming",
    openai.WithTemperature(0.7),
    openai.WithMaxTokens(50),
)

Chat Completion

messages := []llm.Message{
    {
        Role:    "system",
        Content: "You are a helpful programming assistant.",
    },
    {
        Role:    "user",
        Content: "What's the best way to handle errors in Go?",
    },
}

response, err := client.Chat(context.Background(), messages, nil)

Available Options

The OpenAI client provides several option functions for configuring requests:

• WithTemperature(float64) - Controls randomness (0.0 to 1.0)
• WithMaxTokens(int) - Sets maximum response length
• WithTopP(float64) - Controls diversity via nucleus sampling
• WithFrequencyPenalty(float64) - Reduces repetition of token sequences
• WithPresencePenalty(float64) - Reduces repetition of topics
• WithStopSequences([]string) - Specifies sequences where generation should stop

Tool Integration

The OpenAI client also supports tool calling with the GenerateWithTools method. See the agent examples for demonstrations of tool integration.

Additional Examples

Reasoning Support

This package includes an example demonstrating how to use the reasoning capability:

cd reasoning
go run main.go

The reasoning example shows how to control the verbosity and detail of the model's thought process through the WithReasoning option. See the reasoning example README for more details.