Chat Completion API

The chat completion API allows you to interact with GitHub's AI models through a conversational interface.

Simple Example

using OpenGithubModelsApi

# Create a client (using a placeholder token for documentation)
client = GithubModelsClient("ghp_example_token")

# Create a message history
messages = [
    Message(role="system", content="You are a helpful assistant"),
    Message(role="user", content="What is Julia programming language?")
]

# Create an inference request
request = InferenceRequest(
    model="openai/gpt-4.1",  # Example model ID
    messages=messages,
    temperature=0.7
)

# Get a response (returns just the content by default)
response = create_chat_completion(client, request)
println(response)

Request Parameters

Required Parameters

• model: ID of the specific model to use (format: {publisher}/{model_name})
• messages: Array of message objects with role and content

Valid message roles are:

• "system": For system instructions
• "user": For user messages
• "assistant": For assistant responses
• "developer": For developer-specific messages

Advanced Usage

Getting Full Response Details

By default, create_chat_completion returns just the content string. To get the full response object, use verbose=true:

full_response = create_chat_completion(client, request, verbose=true)

Controlling Response Length and Creativity

request = InferenceRequest(
    model="openai/gpt-4.1",
    messages=messages,
    temperature=0.3,      # More deterministic (0.0-1.0)
    max_tokens=100,       # Limit response length
    top_p=0.9             # Alternative to temperature
)

Error Handling

Common errors include:

• Invalid model ID format
• Invalid message roles
• Parameter validation errors (e.g., temperature outside 0-1 range)
• API rate limits

Best Practices

• Always include a system message to guide the model's behavior
• Keep conversation history within the model's context window
• Adjust temperature based on your use case (lower for factual responses, higher for creative tasks)
• Validate inputs before making API calls