Thread by @pwang_szn

Memory in @LangChainAI is a big deal.

It was a lot to go through, but I broke it into bite-size chunks.

• 4 Memory Types Explained
• How to use Memory in a Chain
• How to add Memory to an Agent

Save this ↓

ConversationBufferMemory:

• Memory allows for storing of messages
• Extracts the messages in a variable.

Pro: Basic to understand/pickup

ConversationBufferWindowMemory:

• Only uses the last K messages.

Pro: Useful to keep the memory history small

ConversationSummaryMemory

• Creates a summary of the conversation over time.

Pro: Useful for condensing information from the conversation over time.

VectorStore-Backed Memory

• Stores memories in a VectorDB
• Queries the top-K most relevant docs

Pro:
- Doesn’t explicitly track the order of messages
- AI can remember relevant pieces of information that it was told earlier

How to Add Memory to a Chain

1) Setup Prompt and Memory
2) Initialize LLMChain
3) Call the LLMChain

How to Add Memory to an Agent

1) Create an LLMChain with memory.
2) Use the LLMChain to create a custom Agent.

How to Add Memory to Multi-Input Chain:

Last but not least:

It's also possible to combine memory classes in the same chain:

That's it!

Will be making a lot more demo projects with memory in the upcoming weeks! 🫡