Chat Completion

Overview

The Chat Completion API processes a sequence of messages and generates a context-aware response from the selected language model. Unlike traditional text generation, chat models understand conversation history, system instructions, and user intent, allowing them to produce more accurate and natural responses.

Every request consists of a model and a messages array, while optional parameters allow you to control creativity, response length, streaming, and more.

How Chat Completion Works

Create a Chat Completion

1import { NovaAI } from "@novaai/sdk";
2 
3const client = new NovaAI({
4  apiKey: process.env.NOVA_API_KEY,
5});
6 
7const response = await client.chat.create({
8  model: "nova-chat-4",
9  messages: [
10    {
11      role: "system",
12      content: "You are a helpful technical assistant."
13    },
14    {
15      role: "user",
16      content: "Explain what APIs are in simple terms."
17    }
18  ]
19});
20 
21console.log(response.output);

Request Properties

The language model used to generate the response.

Conversation history sent to the model.

Controls response creativity.

Maximum number of generated tokens.

Enables streaming responses.

Controls response diversity.

Response Example

{
  "id": "chat_9f8ab1",
  "model": "nova-chat-4",
  "output": "An API allows different software applications to communicate with each other by exchanging information.",
  "usage": {
    "prompt_tokens": 24,
    "completion_tokens": 18,
    "total_tokens": 42
  }
}

Supported Use Cases

Parameter Guide

Response Lifecycle

Best Practices

• Start every conversation with a clear system instruction.

• Keep conversation history relevant.

• Choose the appropriate model for your workload.

• Limit response length using max_tokens.

• Enable streaming for real-time conversational experiences.

• Monitor token usage to optimize performance and cost.

Common Errors

Invalid Messages Format

The messages array is missing or contains invalid objects.

Solution

Ensure every message includes both a role and content property.

Unsupported Model

The selected model is unavailable or does not support chat completion.

Solution

Verify the model identifier before sending the request.

Context Window Exceeded

The combined conversation exceeds the model's maximum context window.

Solution

Remove unnecessary messages or summarize earlier parts of the conversation.

Frequently Asked Questions

Next Step

Now that you've learned how to generate conversational responses, continue to Image Generation to explore how NovaAI creates high-quality images from natural language prompts.