Bad results on talking with my own documents

[guilhermelia]

Hello guys, I would like to know your experiencies with your own documents and the history of previous interactions being passed to the OpenAIFunction:

messages = [{"role": "system", "content": system_message}]
for message in chat_history:
messages.append({"role": message["role"], "content": message["content"]})
messages.append({"role": "user", "content": user_message})

I realized that some chunks was passed in the context but LLM was unable do answer correctly. When I bypass the history of messages it could respond correctly. Anyone of you experience a similar situation? Now im working without the history of messages being pulled to the openai method.

Im using GPT4-o model.