LangMemAn SDK from LangChain for giving AI agents long-term memory that persists and adapts across conversations

LangMem is a software development kit produced by the LangChain team that focuses on equipping AI agents with long-term memory. Where most LLM applications are limited to the context window of a single session, LangMem addresses the problem of persistence: how an agent can retain useful information across many interactions and use it to behave more consistently and personally over time. The SDK provides tools for extracting, storing, and retrieving memories from agent conversations. Rather than simply logging raw transcripts, it is designed to distill interactions into structured or semantic memories that can be searched and reused later. This lets an agent recall facts about a user, accumulated preferences, or prior decisions, and incorporate them into future responses. LangMem distinguishes between different kinds of memory, conceptually borrowing from cognitive ideas such as semantic memory (facts and knowledge), episodic memory (past events and interactions), and procedural memory (learned behaviors or instructions). It exposes utilities for forming these memories and for updating them as new information arrives, so an agent's understanding can evolve instead of remaining static. It is built to work within the broader LangChain and LangGraph ecosystem, and integrates with persistent storage backends so memories survive beyond a single process. This makes it a natural fit for teams already building agents with those frameworks who want to add a memory layer without assembling the retrieval and consolidation logic from scratch. As with most emerging agent-memory tooling, LangMem is aimed primarily at developers comfortable with Python and the LangChain stack rather than no-code users, and the field of long-term agent memory is still maturing, so patterns and APIs around it continue to evolve.