09. Save chat to SQLite

SQL (SQLAlchemy)

Copy

Structured Query Language (SQL) Domain-specific language used for silver programming, designed for managing data in relational database management systems (RDBMS) or for stream processing in relational data stream management systems (RDSMS). It is especially useful for dealing with structured data, including the relationship between entities and variables.

SQLAlchemy Is an open source for Python programming languages distributed according to MIT licenses SQL Toolkit and object relationship mapper (ORM).

In this laptop SQLAlchemy You can store chat records in any database supported by SQLChatMessageHistory Describe the class.

SQLite To use with a database other than, you need to install that database driver.

Copy

# API KEY A configuration file for managing environment variables
from dotenv import load_dotenv

# API KEY Load information
load_dotenv()

Copy

True

Usage

To use storage, only the following 2 are provided:

1. session_id -A unique identifier for the session, such as username, email, chat ID, etc.

2. connection -A string that specifies a database connection. This string is passed to SQLAlchemy's create_engine function.

Copy

from langchain_community.chat_message_histories import SQLChatMessageHistory

# SQLChatMessageHistory Create an object and set the session ID and database connection file.
chat_message_history = SQLChatMessageHistory(
    session_id="sql_history", connection="sqlite:///sqlite.db"
)

Copy

# Add a user message.
chat_message_history.add_user_message(
    "Hello? Nice to meet you. My name is Teddy. I am a Langchain developer. Please take care of me in the future!"
)
# AI add a message.
chat_message_history.add_ai_message("Hi Teddy, nice to meet you. Please take care of me too!")

• Check for saved conversation. chat_message_history.messages

Copy

# 채팅 메시지 기록의 메시지들
chat_message_history.messages

Copy

[HumanMessage (content='Hi? Nice to meet you. My name is Teddy. I am a Langstein developer. Please do it in the future!'), AIMessage (content=' Hi Teddy, nice to meet you. Please do me too!')] 

We have this message recording class LCEL Runnables It can be easily combined with.

Apply to Chain

Copy

from langchain_core.prompts import (
    ChatPromptTemplate,
    MessagesPlaceholder,
)
from langchain_core.runnables.history import RunnableWithMessageHistory
from langchain_openai import ChatOpenAI
from langchain_core.output_parsers import StrOutputParser

Copy

prompt = ChatPromptTemplate.from_messages(
    [
        # system messages
        ("system", "You are a helpful assistant."),
        # For conversation records Placeholder
        MessagesPlaceholder(variable_name="chat_history"),
        ("human", "{question}"),  question
    ]
)

# chain System Messages.
chain = prompt | ChatOpenAI(model_name="gpt-4o") | StrOutputParser()

sqlite.db Create a function that brings in-conversation.

Copy

def get_chat_history(user_id, conversation_id):
    return SQLChatMessageHistory(
        table_name=user_id,
        session_id=conversation_id,
        connection="sqlite:///sqlite.db",
    )

config_fields Set. This is used as a reference when viewing conversational information.

• user_id : User ID

• conversation_id : Conversation ID

Copy

from langchain_core.runnables.utils import ConfigurableFieldSpec

config_fields = [
    ConfigurableFieldSpec(
        id="user_id",
        annotation=str,
        name="User ID",
        description="Unique identifier for a user.",
        default="",
        is_shared=True,
    ),
    ConfigurableFieldSpec(
        id="conversation_id",
        annotation=str,
        name="Conversation ID",
        description="Unique identifier for a conversation.",
        default="",
        is_shared=True,
    ),
]

Copy

chain_with_history = RunnableWithMessageHistory(
    chain,
    get_chat_history,  # Sets a function to retrieve conversation history.
    input_messages_key="question",  # Set the key of the input message to "question"
    history_messages_key="chat_history",  # Set the key for the conversation history message to "history"
    history_factory_config=config_fields,  # Set parameters to be referenced when viewing conversation records.
)

• "configurable" Under the key "user_id" , "conversation_id" Set the key-value pair.

Copy

# config setting
config = {"configurable": {"user_id": "user1", "conversation_id": "conversation1"}}

Let's ask a question that asks for a name. If you have a conversation you saved earlier, you will answer it correctly.

• chain_with_history Object invoke Call the method to generate an answer to the question.

• invoke The method has a question dictionary config Settings are passed.

Copy

# questions and config Pass it and execute it.
chain_with_history.invoke({"question": "Hello, nice to meet you, my name is Teddy."}, config)

Copy

 'You said your name was Teddy. Is that correct?' 

Same this time user_id I have conversion_id Set to have different values.

Copy

# config settings
config = {"configurable": {"user_id": "user1", "conversation_id": "conversation2"}}

# questions and config Execute by passing.
chain_with_history.invoke({"question": "what is my name?"}, config)

Copy

'Sorry, but I can't know your name. How can I help you?'