Advanced9 min

LangMem SDK & Store API

The LangMem SDK for automatic memory extraction and the LangGraph Store API for persistent cross-thread storage.

Quick Reference

→LangMem provides create_memory_manager() that automatically extracts and updates user memories from conversations
→Store API: store.put(namespace, key, value) and store.search(namespace, query) for semantic retrieval
→Namespaces follow the pattern ('memories', user_id, memory_type) for organized multi-tenant storage
→Memory managers can run as background tasks on conversation end or as inline nodes in the graph
→Use store.search() with embedding-based similarity to retrieve relevant memories without exact key matches

LangMem Overview

What LangMem does

LangMem automates memory extraction: it reads conversation history, identifies important facts about the user, and stores them as structured memories. You configure the extraction policy; LangMem handles the rest.

Automatic memory extraction with LangMem

Store API Deep Dive

Method	Purpose	When to use
store.put(ns, key, value)	Create or overwrite a memory	Saving a known fact with a specific key
store.get(ns, key)	Retrieve by exact key	Looking up a specific preference ('language')
store.search(ns, query, limit)	Semantic similarity search	Finding relevant memories from natural language
store.delete(ns, key)	Remove a memory	User requests data deletion
store.list(ns)	List all memories in a namespace	Admin views, debugging, data export

Namespace Design Patterns

Namespace design is your memory schema. Get it right early, because restructuring stored memories is painful. The tuple format supports hierarchical organization:

Sign in to read this article

This is a premium article. Sign in with your Google account to continue.