How does conversation with memory chain adds the input and response back into the memory module?

[Jason-CKY]

I can't seem to find the exact line of code where running the conversation chain updates the memory module. Inspecting the conversation.predict function which calls self.apply that runs the llm's generate function. Is there some sort of callbacks triggered to update the memory module that i'm not catching?

from langchain.chains import ConversationChain
conversation = ConversationChain(
        llm=CustomLLM(), 
        memory=ConversationBufferWindowMemory(k=7),
        verbose=True,
    )

# how does this predict function adds the input/response pair into the memory?
conversation.predict(input="Hi, what's up?")

[oddrationale]

The memory is updated as follows:

1. __call__() calls .prep_outputs()
2. .prep_outputs() calls memory.save_context()
3. memory.save_context() adds the user and AI messages to the memory.