LangMem

LangMem is a Python SDK for adding long-term memory to LLM agents. It provides two complementary paths:

• Hot path tools (create_manage_memory_tool, create_search_memory_tool) that agents call during conversations to store and retrieve facts, preferences, and other context.
• Background manager that periodically extracts, consolidates, and updates memories outside the request path.

The core API is storage-agnostic and works with LangGraph’s BaseStore implementations (e.g., in-memory for dev, Postgres for production). It supports semantic and episodic memory patterns, prompt optimization/“procedural memory,” and dynamic namespaces for user- or team-scoped memories. Typical setups pair LangMem with LangGraph agents, Postgres/pgvector for persistence, and common model providers (Anthropic, OpenAI).

Install: pip install -U langmem Requires: Python 3.10+

Use it to build agents that remember user preferences across sessions, keep evolving profiles, and refine prompts over time—without hand-rolling memory extraction, deduplication, and retrieval plumbing.