Chat Completion

Chat completion is the standard interface for interacting with modern language models through an API. Rather than sending a single text prompt, the chat completion format structures the interaction as a list of messages, each with a designated role: system (setting model behaviour), user (human input), and assistant (model responses). This message-based format allows the model to maintain conversational context across multiple turns. Each API call includes the full conversation history, enabling the model to reference earlier messages and maintain coherence. The model generates a completion — the assistant's next response — based on the entire message history provided. Chat completion APIs typically expose parameters like temperature, max tokens, top-k, and stop sequences that control the generation behaviour. They may also support streaming (receiving the response token by token), tool use (allowing the model to call external functions), and structured output formats like JSON mode.

The chat completion API pattern is how most businesses integrate LLMs into their applications. Understanding this interface is essential for development teams building AI-powered products and services. It determines how conversations are managed, how context is maintained, and how costs are controlled. Each chat completion request includes the full conversation history, which means longer conversations consume more tokens and therefore cost more. This has practical implications for application design — teams must decide how much history to include, when to summarise or truncate conversations, and how to manage context window limits. The chat completion format also enables powerful patterns like few-shot prompting (including example conversations in the message history), system prompt engineering (configuring behaviour through the system message), and multi-turn tool use (where the model calls functions and receives results across multiple turns). Mastering these patterns is key to building effective AI applications.