Messages: how a model reads and writes

Messages are the objects a chat model reads and writes. The four you use most are HumanMessage, AIMessage, SystemMessage and ToolMessage.

Last updated: 27 Sep, 2026 · LangChain 1.4

A conversation is a list of messages. Each one has a type (its role) and content. The model is given the list and returns the next message.

The message-type imports

from langchain.messages import HumanMessage, AIMessage, SystemMessage, ToolMessage

A conversation as a list

Start by importing the two message types you need and building a conversation. A conversation is a plain Python list of message objects.

from langchain.messages import SystemMessage, HumanMessage

conversation = [
    SystemMessage("You are a support assistant."),   # your instructions to the model
    HumanMessage("I was charged twice."),            # the user's turn
]

Reading role and content

Walk the list and print each one. type(m).__name__ is the message's class name, which is its role, and m.content is its text.

for m in conversation:
    print(type(m).__name__, "->", m.content)   # role, then text

Printing the conversation

The two snippets together:

from langchain.messages import SystemMessage, HumanMessage

conversation = [
    SystemMessage("You are a support assistant."),
    HumanMessage("I was charged twice."),
]

for m in conversation:
    print(type(m).__name__, "->", m.content)

What each message carries

• Each message is an object with a role (its class) and content.
• A SystemMessage sets instructions; a HumanMessage is the user's turn; the model replies with an AIMessage.
• A ToolMessage carries a tool's result back to the model and must set tool_call_id to match the call it answers.

The four message types

Where messages appear

• Every model call takes a list of messages and returns one.
• The conversation your agent remembers is a list of these objects.

Related

• Previous: Subgraphs: a graph inside a graph
• Next: MessagesState and add_messages
• Reference: Messages