Working through this module?0 of 17 done
Module 08 · Page 9 of 20 · 8 sections

Memory in LangChain

Why memory matters, the memory types and their use cases, how ConversationBufferMemory works and examples for buffer, window and summary memory.

06Chains in LangChain07Agents in LangChain15Embeddings and Semantic Search
8.1

What is Memory?

  • Memory allows LLMs to remember previous interactions.
  • It helps maintain conversation context across multiple turns.
  • Without memory, each request is treated independently and the model forgets previous messages.
  • LangChain provides different types of memory based on the use case.
8.2

Why Use Memory?

  • Maintain conversation history.
  • Provide contextual responses.
  • Useful for chatbots, virtual assistants and agents.
  • Improves user experience.
  • Supports short-term and long-term memory.
  • Helps in building more human-like conversations.
8.3

Types of Memory in LangChain

Memory TypeDescriptionUse Case
ConversationBufferMemoryStores full conversation history in memory.Simple chat applications.
ConversationBufferWindowMemoryStores last k messages.Limit memory size.
ConversationSummaryMemorySummarizes old messages.Long conversations.
EntityMemoryRemembers important entities (e.g., names, facts).Personalized assistants.
VectorStoreRetrieverMemoryStores conversation history in a vector store.Long-term memory with retrieval.
8.4

ConversationBufferMemory

python
from langchain_openai import ChatOpenAI
from langchain.memory import ConversationBufferMemory
from langchain.chains import ConversationChain

llm = ChatOpenAI(model="gpt-3.5-turbo", temperature=0.7)

memory = ConversationBufferMemory()

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

# First message
response1 = conversation.predict(input="Hi, my name is Rahul.")
print(response1)
# Second message (model remembers previous message)
response2 = conversation.predict(input="What is my name?")
print(response2)
8.5

How ConversationBufferMemory Works

User InputNew message
Load Previous MessagesFrom memory
Pass Full ConversationAs context to LLM
Get Response
Save Current MessageTo memory
8.6

ConversationBufferWindowMemory

python
from langchain.memory import ConversationBufferWindowMemory
from langchain.chains import ConversationChain

memory = ConversationBufferWindowMemory(k=3)

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

# This will keep only last 3 messages in memory
8.7

ConversationSummaryMemory

python
from langchain.memory import ConversationSummaryMemory
memory = ConversationSummaryMemory(llm=llm)

conversation = ConversationChain(
    llm=llm,
    memory=memory,
    verbose=True
)
8.8

Key Takeaways

  • Memory helps in maintaining context in conversations.
  • Choose the right memory type based on your use case.
  • Use summary memory for long conversations.
  • Vector store memory is useful for long-term storage.
  • Memory can also be used with agents and chains.