Conversation Buffer Memory

1. What is Conversation Buffer Memory?

ConversationBufferMemory is LangChain’s simplest memory type. It stores the entire chat history (user + AI messages) and sends it back to the model on every new request.

In short:

It lets the AI remember what was said earlier in the conversation.

---

2. Why does it exist?

LLMs are stateless by default.

Without memory:

User: My name is John
User: What is my name?
AI: I don’t know

With buffer memory:

AI: Your name is John

So memory solves context loss.

---

3. Real-world analogy

Think of:

• ❌ No memory → talking to a person with amnesia

• ✅ Buffer memory → a person who remembers everything you said so far

But note:

• It remembers everything, even irrelevant parts

• That can become expensive and noisy

---

4. Minimal working example (Gemini)

from langchain_google_genai import ChatGoogleGenerativeAI
from langchain.memory import ConversationBufferMemory
from langchain.chains import ConversationChain
import os

llm = ChatGoogleGenerativeAI(
    model="gemini-1.5-flash",
    api_key=os.getenv("GEMINI_API_KEY")
)

memory = ConversationBufferMemory()

conversation = ConversationChain(
    llm=llm,
    memory=memory,
    verbose=True
)

conversation.invoke("My name is John")
conversation.invoke("What is my name?")

What happens internally?

• First call → memory stores: “My name is John”

• Second call → full history is injected into the prompt

• Model answers correctly

---

5. What does the memory actually store?

print(memory.buffer)

Output (roughly):

Human: My name is John
AI: Nice to meet you John
Human: What is my name?

It’s plain text, not embeddings or summaries.

---

6. Key characteristics (important)

Feature	ConversationBufferMemory
Stores	Full conversation
Grows over time	Yes
Token usage	High
Summarization	❌ No
Best for	Short conversations

---

7. Common beginner mistakes

❌ Using it for long chats → token explosion ❌ Assuming it’s smart memory (it’s just text) ❌ Using it in production without limits

---

8. When NOT to use it

Avoid ConversationBufferMemory if:

• Chat is long-running

• Cost/token usage matters

• Only recent context is needed

Use instead:

• ConversationBufferWindowMemory

• ConversationSummaryMemory

---

9. One-line mental model

ConversationBufferMemory = append entire chat history to every prompt