LangMem SDK

What it does

LangMem SDK is a Python library for adding long-term memory to AI agents. It takes in conversation messages, existing stored memories and, for prompt optimization, scored interaction trajectories with feedback comments. It produces extracted facts and knowledge, consolidated memory updates, and revised system prompts. Memories are namespaced, for example per user or team, so they can be kept separate or shared.

Developers use it in two ways: give an agent memory-management and memory-search tools that it calls during a conversation, or run a background memory manager that extracts and updates knowledge afterward. It covers semantic memory (facts, user preferences) and procedural memory (prompt rules that evolve from feedback). Episodic memory has no opinionated utilities yet. Its core API works with any storage system and agent framework, with native integration into LangGraph's long-term memory store.

It suits developers building personalized or adaptive agents who are comfortable with Python and have an API key for an LLM provider. Install is via pip; the library is MIT-licensed, but LLM and embedding calls are paid separately. The example in-memory store loses data on restart, so production use needs a database-backed store. It handles recall across conversations, not within-thread context.

What it can do

• Extract facts and preferences from conversations into semantic memory
• Memory tools an agent calls to store and search memories during a chat
• Background memory manager that extracts, consolidates and updates knowledge
• Prompt optimizers that rewrite system prompts from scored interaction trajectories
• Namespaced memories scoped to users, teams or an agent's shared procedures

Works with

• LangGraph

Example workflows

Remember user preferences in a chat agent

Input: a LangGraph agent, a store and user messages. Steps: add the manage-memory and search-memory tools with a namespace, tell the agent a preference, then ask about it later. Output: the agent answers from the stored memory.

Refine a system prompt from feedback

Input: conversation trajectories with scores and comments, plus the current prompt. Steps: create a prompt optimizer, pass in the trajectories and prompt. Output: a revised prompt reflecting patterns in good and poor interactions.

Before you choose

• The in-memory store example loses memories on restart; production needs a persistent store such as Postgres.
• Episodic memory has no opinionated utilities yet, per the launch post.
• Memory extraction and prompt optimization call an LLM, so provider costs apply separately.

Listed under

• Coding
• Agent memory
• AI agents
• Developer library
• Developers
• LLM agents
• Long-term memory

Get started

Docs and quickstarts at the official docs site

1. Run pip install -U langmem in your Python environment.
2. Export an API key for your LLM provider, such as ANTHROPIC_API_KEY.
3. Choose a memory store (in-memory for testing, a database-backed store for production) and add the memory tools to your agent.

The MIT-licensed SDK is free to install. Model, embedding and infrastructure charges are separate. Managed services, where used, have their own terms.

LangMem SDK is free to use. See their pricing page ↗

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• LangMem SDK for agent long-term memory ↗www.langchain.com
• Introduction ↗langchain-ai.github.io
• GitHub - langchain-ai/langmem · GitHub ↗github.com
• LangSmith Plans and Pricing ↗www.langchain.com
• Home - Docs by LangChain ↗docs.langchain.com

Company website

langchain-ai.github.io: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.