Chat History and Memory

The model remembers only what you resend

Chat models are stateless. A chat message history stores the messages of a conversation (human, AI, tool). On each turn you put the stored messages into the prompt with a MessagesPlaceholder, then append the new question and answer. Modern LangChain recommends keeping this state explicitly, for example with a RunnableWithMessageHistory wrapper or, for richer needs, LangGraph persistence/checkpointers. Control growth: keep the last N messages, summarise old ones, or store key facts separately, and key histories by session ID so users never see each other's conversations.

An in-memory chat history, run

I ran this offline in a Python virtual environment with langchain-core 1.6.6, langchain-text-splitters 1.1.2 and llama-index-core 0.14.25. No API key or network call is needed because a fake model or a toy embedding stands in for the real one. The history stores three messages in order, tagged human, ai, human. Passing them back with the next prompt is what lets the model answer "What is my name?".

from langchain_core.chat_history import InMemoryChatMessageHistory

history = InMemoryChatMessageHistory()
history.add_user_message("My name is Asha.")
history.add_ai_message("Nice to meet you, Asha.")
history.add_user_message("What is my name?")
for m in history.messages:
    print(m.type, "|", m.content)
print("messages kept:", len(history.messages))

Output:

human | My name is Asha.
ai | Nice to meet you, Asha.
human | What is my name?
messages kept: 3

Quick check: Why key chat histories by session ID?

• To shorten prompts only
• To speed up tokens
• To train the model
• So users never see each other's conversations