Troubleshooting: Memory doesn't seem to work in my ConversationalRetrievalQA chain.

I've been helping a few members debug their ConversationalRetrievalQA chains recently, and one recurring theme is memory not working as expected. The conversation just doesn't seem to carry forward context from one question to the next.

Here's a typical setup I'm seeing:

```python
from langchain.memory import ConversationBufferMemory
memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)

chain = ConversationalRetrievalChain.from_llm(
llm=llm,
retriever=retriever,
memory=memory,
verbose=True
)
```

Despite this, the chain often acts like it has no memory of the previous exchange. The `chat_history` appears empty or isn't being utilized in the follow-up question.

From my troubleshooting, the issues usually boil down to a few common misconfigurations:

* **Input key mismatches:** The chain expects a specific key for the question (often `"question"`) and the memory might be outputting to a different key.
* **Memory not being passed correctly:** Sometimes the memory object is created but not properly integrated into the chain's inputs/outputs.
* **Using the wrong chain type:** Certain chain constructions, especially custom ones, might not automatically handle memory I/O.

Has anyone else run into this? If you've solved it, what was the root cause in your case?

I'm particularly curious about:
- Are you using `ConversationBufferMemory` or another type (e.g., `ConversationSummaryMemory`)?
- Does your `verbose` output show the `chat_history` being populated and passed?
- What LLM provider and chat model are you using?

Sharing specific code snippets and what you've already tried will help us pinpoint the issue. Let's figure out the patterns. —K