Demo Adding Short Term Memory

In this tutorial, you’ll learn how to build a simple memory (conversation history) feature into a chatbot using LangChain. We’ll start with a stateless prompt, observe why follow-up questions break, then enhance our chain to carry context across messages for a more natural dialogue.

1. Stateless Prompt: No Memory

First, let’s define a minimal chat chain without any history support:

from langchain_core.prompts import ChatPromptTemplate
from langchain_openai.chat_models import ChatOpenAI

model = ChatOpenAI()
prompt = ChatPromptTemplate.from_messages([
    ("system", "You’re an assistant with expertise in {ability}."),
    ("human", "{input}"),
])
base_chain = prompt | model

Invoke it twice—once to ask about right-angled triangles, then a follow-up without context:

# First call: establishing the topic
response1 = base_chain.invoke({
    "ability": "math",
    "input":   "What’s a right-angled triangle?"
})
print(response1.content)
# Second call: follow-up without history
response2 = base_chain.invoke({
    "ability": "math",
    "input":   "What are the other types?"
})
print(response2.content)
# ➜ "Could you please clarify what you’re asking about?"

2. Adding a History Placeholder

To maintain context, inject a MessagesPlaceholder into your prompt template:

from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_openai.chat_models import ChatOpenAI

model = ChatOpenAI()
prompt = ChatPromptTemplate.from_messages([
    ("system", "You’re an assistant with expertise in {ability}."),
    MessagesPlaceholder(variable_name="history"),
    ("human", "{input}"),
])
base_chain = prompt | model

2.1 Constructing the History

Build a simple list of (role, content) tuples that tracks the dialogue so far:

history = [
    ("human", "What’s a right-angled triangle?"),
    ("ai",    "A right-angled triangle has one angle of 90 degrees, "
              "with the other two angles summing to 90 degrees."),
]

2.2 Invoking with History

Pass the history on each request to provide context:

response = base_chain.invoke({
    "ability": "math",
    "input":   "What are the other types?",
    "history": history
})
print(response.content)
# ➜ "Other types of triangles include equilateral (all sides equal), "
#     "isosceles (two sides equal), and scalene (no sides equal)."

By supplying the history key, the chain can reference previous turns and answer accurately.

3. Putting It All Together

Here’s the complete, memory-enabled chatbot example:

from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_openai.chat_models import ChatOpenAI

model = ChatOpenAI()
prompt = ChatPromptTemplate.from_messages([
    ("system", "You’re an assistant who’s good at {ability}."),
    MessagesPlaceholder(variable_name="history"),
    ("human", "{input}"),
])
base_chain = prompt | model

history = [
    ("human", "What’s a right-angled triangle?"),
    ("ai",    "A right-angled triangle has one angle of 90 degrees, "
              "with the other two angles summing to 90 degrees.")
]

result = base_chain.invoke({
    "ability": "math",
    "input":   "What are the other types?",
    "history": history
})
print(result.content)

4. Sending Configurable Parameters

You can also include adjustable parameters—like response length limits—directly in your system prompt:

from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_openai.chat_models import ChatOpenAI

model = ChatOpenAI()
prompt = ChatPromptTemplate.from_messages([
    ("system", "You’re an assistant who’s good at {ability}. Respond in 20 words or fewer."),
    MessagesPlaceholder(variable_name="history"),
    ("human", "{input}"),
])
base_chain = prompt | model

history = [
    ("human", "What’s a right-angled triangle?"),
    ("ai",    "A right-angled triangle has one angle of 90 degrees, "
              "with the other two angles summing to 90 degrees.")
]

response = base_chain.invoke({
    "ability": "math",
    "input":   "What are the other types?",
    "history": history
})
print(response.content)

Comparison: Stateless vs. Memory-Enabled

Links and References

• LangChain Documentation
• OpenAI Chat Models
• MessagesPlaceholder API