Skip to main content

Getting Started: Chat Models

info

Conceptual Guide

LangChain provides a standard interface for using chat models. Chat models are a variation on language models. While chat models use language models under the hood, the interface they expose is a bit different. Rather than expose a "text in, text out" API, they expose an interface where "chat messages" are the inputs and outputs.

Chat Messages​

A ChatMessage is what we refer to as the modular unit of information for a chat model. At the moment, this consists of a "text" field, which refers to the content of the chat message.

There are currently four different classes of ChatMessage supported by LangChain:

  • HumanChatMessage: A chat message that is sent as if from a Human's point of view.
  • AIChatMessage: A chat message that is sent from the point of view of the AI system to which the Human is corresponding.
  • SystemChatMessage: A chat message that gives the AI system some information about the conversation. This is usually sent at the beginning of a conversation.
  • ChatMessage: A generic chat message, with not only a "text" field but also an arbitrary "role" field.

These classes model conversation history. When calling a model with GenerateContent, messages are passed as llms.MessageContent, which carries a role plus one or more ContentParts — so a single message can mix text, images, and other modalities on providers that support it (build them with llms.TextParts, llms.TextPart, llms.ImageURLPart, and related helpers).

Dig deeper​