Langchain CSV_agent

[akabass]

@dosu-bot hello i am building my web app chatbot using the Langchain CSV_AGENT but i have a problem here that i want to make the chatbot be conversational also so the agent could generate responses from the csv search and also from the chat history till now i want to make it locally so i dont handle the chat sessions or anything all i need is just whne i run the app there is a new chat history is created for the conversation and thats it so the agent have access to this conversation and the uploaded csv_file ----- trying doing this but the agent is only answering from the csv file but when i try to ask about anything conversationaly eg. i try to send a question of "hello my name is jhon how are you " and the next question i ask "what is my name" the agent answer me that its dont know this private data ----- here is the code of the flask app ------>>>>

@app.route('/csvupload', methods=['GET', 'POST'])
def upload_file():
    global agent,agent_chain,memory_x, file_uploaded, user_questions, ai_answers
    
    if request.method == 'POST':
        file = request.files['csv_file']
        filename = file.filename
        if filename.endswith('.csv')or filename.endswith('.xlsx'):
            clear_folder('uploads')
            filepath = os.path.join(app.config['UPLOAD_FOLDER'], filename)
            file.save(filepath)
            file_uploaded = True  # Set file_uploaded flag

            if agent is None:
                memory = ConversationBufferMemory(memory_key="memory", return_messages=True)
                agent = create_csv_agent(OpenAI(temperature=0, openai_api_key=os.environ["OPENAI_API_KEY"]), filepath,agent_type=AgentType.ZERO_SHOT_REACT_DESCRIPTION,memory=memory,verbose=True)
                # Create the memory object
                agent_chain = AgentExecutor.from_agent_and_tools(
                agent=agent, tools=[],  memory=memory
                    )

                # Add the memory to the agent
                agent.memory = memory
            user_questions = []
            ai_answers = []
            return redirect(url_for('chat_box'))  # Redirect to the chat box route

    return render_template('csvhome.html')


@app.route('/csvchat', methods=['GET', 'POST'])
def chat_box():
    global agent,agent_chain,memory_x, file_uploaded, user_questions, ai_answers  # Access global variables

    if not file_uploaded:
        return redirect(url_for('upload_file'))  # Redirect to file upload if no file has been uploaded

    if request.method == 'POST':
        question = request.form['question']
        response = agent_chain.run(input=question, language="English")  # Use user input from the web page
        user_questions.append(question)
        ai_answers.append(response)

        print(agent.memory)

[dosubot[bot]]

🤖

Hello,

From your code, it seems like you're trying to use the ConversationBufferMemory to store the conversation history and use it for generating responses. However, it seems like the memory is not being updated with the conversation history.

The ConversationBufferMemory class in LangChain is a buffer for storing conversation memory. It has a buffer property that returns the buffer of memory. If return_messages is True, it returns the buffer as a list of messages (buffer_as_messages). If return_messages is False, it returns the buffer as a string (buffer_as_str). You can find more details about this class in the LangChain repository.

In your code, you're creating the ConversationBufferMemory and the agent, and you're adding the memory to the agent. However, it seems like you're not updating the memory with the conversation history. You should update the memory after each interaction with the user. You can do this by calling the update method of the ConversationBufferMemory class with the user's input and the agent's response.

Here's how you can modify your code to update the memory:

@app.route('/csvchat', methods=['GET', 'POST'])
def chat_box():
    global agent,agent_chain,memory_x, file_uploaded, user_questions, ai_answers  # Access global variables

    if not file_uploaded:
        return redirect(url_for('upload_file'))  # Redirect to file upload if no file has been uploaded

    if request.method == 'POST':
        question = request.form['question']
        response = agent_chain.run(input=question, language="English")  # Use user input from the web page
        user_questions.append(question)
        ai_answers.append(response)

        # Update the memory with the user's input and the agent's response
        agent.memory.update(human_input=question, ai_output=response)

        print(agent.memory)

This should update the memory with the conversation history and allow the agent to generate responses based on the conversation history.

I hope this helps! Let me know if you have any other questions.

Sources

• Need help to continue same chain over API Request?
• libs/langchain/langchain/memory/buffer.py
• libs/langchain/langchain/agents/agent_toolkits/csv/base.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.