Every large language model can only consider a limited amount of text at a time, measured in tokens. That limit is the context window. It includes your messages, the model's replies, and any files or instructions in the conversation.
When a conversation runs past the window, older parts drop out and the model 'forgets' them. That is why long chats drift, repeat themselves, or contradict earlier answers. Models with larger windows handle long documents better, but still benefit from focused, well-organized input.