Telegram Bot mixing chat hisotry using aiogram lib and openai api

i am making telegram bot which will interact openai api and i am using aiogram for telegram bot.
Also for the history maintaining i am using lanchain memory, just memory part not anything else from langchain

but i am seeing that if two user from differnt device , chatting with telegram bot their history is mixing so when more user will chat thier chat hisotry will aslo mix,

I am using ‘/start’ command from the telegram to clear the user chat history, but this will also clear chat histroy from all the user devices,

what am i doing wrong?

When creating a Telegram bot that maintains separate conversation histories with individual users, especially when interfacing with an AI model such as OpenAI’s API, it’s important to isolate each user’s context. If chat histories are getting mixed up across different users, it means there is an issue with how the bot is handling and storing session state.

Here are a few steps and considerations to ensure that each user has a separate chat history:

User-Specific Sessions:

Ensure that you are creating and maintaining separate sessions for each user. In aiogram, this often means using the user_id provided by the Telegram API as a key to store and retrieve conversation history.

Database or In-Memory Store:

Use a database or an in-memory data store like Redis to maintain conversation states. Each user’s chat history should be stored under a unique key, typically their user_id. If you’re using LangChain’s memory module, you’ll need to make sure that each user’s messages are stored and retrieved using their unique identifier.

Scoped Memory within the Bot:

If the LangChain memory module isn’t properly scoped for each user, the histories will mix. Make sure you’re not using a single global instance of memory for all chats. Instead, instantiate a separate memory object for each user session.

Clearing History with /start:

When using the /start command to clear chat history, you need to ensure that it only clears the history for the user who issued the command. This should be implemented in such a way that it only affects the conversation state associated with the user_id that called /start.

Concurrency Handling:

If your bot is expected to handle multiple users at the same time, ensure that the code handling the session state is thread-safe or that you’re using async patterns properly to avoid race conditions that could lead to mixed histories.


Add logging to your bot to track conversation states and histories. This will help you debug issues where histories might be getting mixed. You should be able to see exactly when and where a conversation history is being accessed or modified.

Check Bot Logic:

Ensure that your bot logic is correctly associating incoming messages with the correct user session. Each interaction should be tagged with the user’s unique identifier and handled within the context of their specific session.

Here is a rough outline of steps in code for handling user-specific sessions in aiogram:

from aiogram import Bot, Dispatcher, types
from aiogram.contrib.middlewares.logging import LoggingMiddleware
from aiogram.utils import executor

# Initialize bot and dispatcher
bot = Bot(token='YOUR_BOT_TOKEN')
dp = Dispatcher(bot)

# Assuming a dictionary to hold user sessions
user_sessions = {}

async def start_command(message: types.Message):
    user_id =
    user_sessions[user_id] = {}  # Initializes or clears the session for the specific user
    await message.reply("Welcome! Your chat history has been cleared.")

async def handle_message(message: types.Message):
    user_id =
    # Retrieve user session, creating a new one if necessary
    session = user_sessions.get(user_id, {})
    # Here you would interact with OpenAI API, using the session for context
    # And then update the session with the new message
    user_sessions[user_id] = session  # Save the updated session

# More handlers and bot logic here...

if __name__ == '__main__':

Remember to replace 'YOUR_BOT_TOKEN' with your actual bot token.

If the problem persists, you may want to examine your LangChain memory integration and ensure that you’re passing the correct user identifier for each interaction.

1 Like