ConversationSummaryMemory

1. What is ConversationSummaryMemory?

ConversationSummaryMemory stores a running summary of the conversation instead of storing raw messages.

After each turn, the LLM updates the summary to include new important information.

---

2. Why does it exist?

Buffer-based memories:

• Grow forever

• Cost tokens

• Include noise

Summary memory solves this by:

• Compressing old conversation

• Keeping only what matters

• Maintaining long-term context cheaply

In short:

Remember the conversation as a summary, not a transcript.

---

3. Real-world analogy

Think of:

• ❌ Chat log → word-by-word recording

• ✅ Summary → meeting minutes

You don’t remember every sentence, only the important points.

---

4. Minimal working example (Gemini)

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

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

memory = ConversationSummaryMemory(
    llm=llm
)

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

conversation.invoke("My name is John")
conversation.invoke("I live in Toronto")
conversation.invoke("I work at Google as a software engineer")
conversation.invoke("What do you know about me?")

---

5. What does it store internally?

print(memory.buffer)

Example summary:

The user is named John, lives in Toronto, and works at Google as a software engineer.

Notice:

• No raw messages

• Just important facts

---

6. How does it work internally?

After each turn:

1. Previous summary is taken

2. New message is added

3. LLM rewrites the summary

So the summary evolves over time.

---

7. Key characteristics

Feature	Summary Memory
Stores	Summary text
Token usage	Low
Long conversations	✅
Exact wording	❌
Fact accuracy	Medium

---

8. Comparison with other memories

Memory Type	Best at
Buffer	Short chats
Window	Recent context
Token buffer	Cost control
Entity	Facts
KG	Relationships
Summary	Long conversations

---

9. Common mistakes

❌ Expecting exact quotes ❌ Using it for precise instructions ❌ Assuming summaries never drift

Summaries can lose detail over time.

---

10. When should you use it?

Use ConversationSummaryMemory when:

• Conversations are long

• You want long-term context

• Exact wording is not important

Avoid when:

• You need step-by-step instructions

• You need recent verbatim context

---

11. One-line mental model

ConversationSummaryMemory = rolling conversation summary