This tutorial covers how to create a multi-turn Chain that remembers previous conversations, using LangChain.
It includes managing conversation history, defining a ChatPromptTemplate, and utilizing an LLM for chain creation.
The conversation history is managed using chat_history.
Alternatively, environment variables can also be set using a .env file.
[Note]
This is not necessary if you've already set the environment variables in the previous step.
from dotenv import load_dotenv
load_dotenv()
Creating a Chain that Remembers Previous Conversations
MessagesPlaceholder is a class in LangChain used to handle conversation history. It is primarily utilized in chatbots or multi-turn conversation systems to store and reuse previous conversation content.
Key Roles
Inserting Conversation History :
Used to insert prior conversations (e.g., question-and-answer history) into the prompt.
This allows the model to understand the context of the conversation and generate appropriate responses.
Managing Variables :
Manages conversation history within the prompt using a specific key (e.g., chat_history).
It is linked to a user-defined variable name.
Usage
MessagesPlaceholder(variable_name="chat_history")
Here, chat_history is the variable name where conversation history is stored.
As the conversation progresses, chat_history is continually updated with pairs of questions and responses.
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_community.chat_message_histories import ChatMessageHistory
from langchain_core.chat_history import BaseChatMessageHistory
from langchain_core.runnables.history import RunnableWithMessageHistory
from langchain_openai import ChatOpenAI
from langchain_core.output_parsers import StrOutputParser
# Define the prompt.
prompt = ChatPromptTemplate.from_messages(
[
(
"system",
"You are a Question-Answering chatbot. Please provide answers to the given questions.",
),
# Use "chat_history" as the key for conversation history without modifying it if possible.
MessagesPlaceholder(variable_name="chat_history"),
("human", "#Question:\n{question}"), # Use user input as a variable.
]
)
# Create the LLM.
llm = ChatOpenAI(model_name="gpt-4o")
# Create a basic chain.
chain = prompt | llm | StrOutputParser()
Creating a Chain to Record Conversations (chain_with_history)
In this step, we create a system that manages session-based conversation history and generates an executable chain.
Conversation History Management : The store dictionary saves and retrieves conversation history (ChatMessageHistory) by session ID. If a session does not exist, a new one is created.
Chain Execution : RunnableWithMessageHistory combines conversation history and the chain to generate responses based on user questions and conversation history. This structure is designed to effectively manage multi-turn conversations.
# A dictionary to store session history.
store = {}
# A function to retrieve session history based on the session ID.
def get_session_history(session_ids):
print(f"[Conversation session ID]: {session_ids}")
if session_ids not in store: # When the session ID is not in the store.
# Create a new ChatMessageHistory object and save it in the store.
store[session_ids] = ChatMessageHistory()
return store[session_ids] # Return the session history for the given session ID.
chain_with_history = RunnableWithMessageHistory(
chain,
get_session_history, # A function to retrieve session history.
input_messages_key="question", # The key where the user's question will be inserted into the template variable.
history_messages_key="chat_history", # The key for the message in the history.
)
Execute the first question.
chain_with_history.invoke(
# Question input.
{"question": "My name is Teddy."},
# Record conversations based on the session ID.
config={"configurable": {"session_id": "abc123"}},
)
[Conversation session ID]: abc123
"Hello, Teddy! Do you have a question or something specific you'd like to discuss?"
Execute the next question.
chain_with_history.invoke(
# Question input.
{"question": "What's my name?"},
# Record conversations based on the session ID.
config={"configurable": {"session_id": "abc123"}},
)
[Conversation session ID]: abc123
'Your name is Teddy.'
Below is a case where a new session is created when the session_id is different.
chain_with_history.invoke(
# Question input.
{"question": "What's my name?"},
# Record conversations based on the session ID.
config={"configurable": {"session_id": "abc1234"}},
)
[Conversation session ID]: abc1234
"I'm sorry, but I don't have access to personal information about you, including your name."
