Module 12: LangChain Memory – Making LLMs Remember Context

In real-world applications, like chatbots and assistants, we often want the system to remember past conversations. This is where LangChain Memory comes into play.

What is Memory?

By default, LLMs are stateless. This means:

  • They don’t remember anything from earlier user inputs.
  • Each call to the LLM is treated like a blank slate.

Memory allows us to maintain context between multiple inputs.

Types of Memory in LangChain

LangChain provides several built-in memory classes for different use-cases.

1. ConversationBufferMemory

Stores the entire conversation in memory as a buffer (long text history). Best for simple dialogs.

from langchain.memory import ConversationBufferMemory

memory = ConversationBufferMemory()

2. ConversationBufferWindowMemory

Same as buffer memory but only keeps the last k messages.

from langchain.memory import ConversationBufferWindowMemory

memory = ConversationBufferWindowMemory(k=3)

3. ConversationSummaryMemory

Summarizes the conversation so far to save tokens.

from langchain.chat_models import ChatOpenAI
from langchain.memory import ConversationSummaryMemory

llm = ChatOpenAI()
memory = ConversationSummaryMemory(llm=llm)

4. EntityMemory

Tracks information about entities (like names, places) throughout the conversation.

Adding Memory to Chains

You can add memory to any chain. Most commonly used with ConversationChain.

from langchain.chains import ConversationChain
from langchain.chat_models import ChatOpenAI
from langchain.memory import ConversationBufferMemory

llm = ChatOpenAI()
memory = ConversationBufferMemory()
conversation = ConversationChain(llm=llm, memory=memory)

response = conversation.run("Hello, my name is Nitin.")
print(response)

response = conversation.run("What is my name?")
print(response)  # Should recall 'Nitin'

Customizing Memory

You can customize memory parameters:

  • k in buffer window memory (controls how many turns to remember)
  • LLM model for summarization in summary memory
  • Prefix/suffix prompts for memory formatting

Full Stateful Chat Example

from langchain.chains import ConversationChain
from langchain.chat_models import ChatOpenAI
from langchain.memory import ConversationBufferWindowMemory

llm = ChatOpenAI()
memory = ConversationBufferWindowMemory(k=2)

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

conversation.run("Hi, I'm Anjali.")
conversation.run("I live in Delhi.")
conversation.run("Where do I live?")
Important
Situation Recommended Memory
Simple chatbot ConversationBufferMemory
Long conversation ConversationSummaryMemory
Entity-aware dialog EntityMemory
Limited token budget Summary or Window Memory
Note

Memory is essential when building applications that require context awareness, like chatbots, customer support agents, or tutors.