Checkpointing and persistence in LangChain - Step-by-Step Execution

Concept Flow - Checkpointing and persistence

Start: Initialize LangChain

↓

Create Chain with Memory

↓

User Input -> Chain Process

↓

Save State to Persistent Storage

↓

Load State from Storage on Restart

↓

Continue Conversation with Restored State

↓

End

Shows how LangChain saves conversation state (checkpoint) to storage and reloads it to continue later.

Execution Sample

LangChain

from langchain.chains import ConversationChain
from langchain.memory import ConversationBufferMemory
from langchain.llms import OpenAI
import json

llm = OpenAI(temperature=0)  # Requires OPENAI_API_KEY environment variable
memory = ConversationBufferMemory()
chain = ConversationChain(llm=llm, memory=memory)

response = chain.run("Hello!")
# Save state to persistent storage
history = [{"input": human, "output": ai} for human, ai in memory.buffer]
with open("memory.json", "w") as f:
    json.dump({"history": history}, f)

This code runs a conversation chain, then saves the memory state to a file for persistence.

Execution Table

Step	Action	Memory State Before	Memory State After	Output
1	Initialize ConversationBufferMemory	{}	{"history": []}	None
2	Create ConversationChain with memory	{"history": []}	{"history": []}	None
3	Run chain with input 'Hello!'	{"history": []}	{"history": [{"input": "Hello!", "output": "Hi! How can I help you today?"}]}	Hi! How can I help you today?
4	Save memory to disk 'memory.json'	{"history": [{"input": "Hello!", "output": "Hi! How can I help you today?"}]}	{"history": [{"input": "Hello!", "output": "Hi! How can I help you today?"}]}	File saved
5	Load memory from disk 'memory.json'	{}	{"history": [{"input": "Hello!", "output": "Hi! How can I help you today?"}]}	Memory restored
6	Run chain with input 'What is LangChain?'	{"history": [{"input": "Hello!", "output": "Hi! How can I help you today?"}]}	{"history": [{"input": "Hello!", "output": "Hi! How can I help you today?"}, {"input": "What is LangChain?", "output": "LangChain is a framework to build applications with language models."}]}	LangChain is a framework to build applications with language models.
7	End of example	{"history": [...]}	{"history": [...]}	Conversation continues

💡 Execution stops after conversation state is saved and restored, showing persistence.

Variable Tracker

Variable	Start	After Step 3	After Step 4	After Step 5	After Step 6	Final
memory.history	[]	[{"input": "Hello!", "output": "Hi! How can I help you today?"}]	[{"input": "Hello!", "output": "Hi! How can I help you today?"}]	[{"input": "Hello!", "output": "Hi! How can I help you today?"}]	[{"input": "Hello!", "output": "Hi! How can I help you today?"}, {"input": "What is LangChain?", "output": "LangChain is a framework to build applications with language models."}]	[{"input": "Hello!", "output": "Hi! How can I help you today?"}, {"input": "What is LangChain?", "output": "LangChain is a framework to build applications with language models."}]

Key Moments - 3 Insights

Why does the memory state before saving already include the conversation?

Because the memory updates immediately after running the chain (see step 3 and 4 in execution_table), so saving captures the latest conversation.

What happens if we load memory from disk before any conversation?

Loading restores the saved history, so the chain can continue as if the conversation never stopped (see step 5).

Does the chain output change after restoring memory?

Yes, because the chain uses the restored history to generate context-aware responses (see step 6 output).

Visual Quiz - 3 Questions

Look at the execution_table at step 3. What is the memory.history after running the chain with input 'Hello!'?

A []

B [{"input": "Hello!", "output": "Hi! How can I help you today?"}]

C [{"input": "Hi!", "output": "Hello!"}]

D null

💡 Hint

Check the 'Memory State After' column at step 3 in the execution_table.

At which step does the memory get saved to disk?

A Step 2

B Step 5

C Step 4

D Step 6

💡 Hint

Look for the action mentioning 'Save memory to disk' in the execution_table.

If we did not load memory from disk at step 5, what would happen at step 6?

A The chain would continue with empty history.

B The chain would crash.

C The chain would use the saved history anyway.

D The chain would delete the previous conversation.

💡 Hint

Refer to variable_tracker and execution_table steps 5 and 6 about memory restoration.

Concept Snapshot

Checkpointing and persistence in LangChain:
- Use ConversationBufferMemory to store conversation.
- Run chain to update memory with inputs/outputs.
- Save memory.buffer as JSON to disk.
- Load memory.buffer from JSON on restart.
- Restored memory lets chain continue conversation seamlessly.

Full Transcript

Checkpointing and persistence in LangChain means saving the conversation memory to a file and loading it later to continue the chat. We start by creating a ConversationBufferMemory and a ConversationChain using it. When the user inputs text, the chain processes it and updates the memory with the input and output. We then save this memory state to disk. Later, we can load this saved memory back, restoring the conversation history. This allows the chain to respond with context from previous messages, making the conversation feel continuous. The execution table shows each step, including memory states before and after actions, and the outputs. The variable tracker follows how the memory.history list grows with each input/output pair. Key moments clarify why memory updates immediately and how loading restores context. The visual quiz tests understanding of memory states and saving/loading steps. The snapshot summarizes the process in simple steps.