Glossary · Term

LangMem

← all terms

Definition

Plain language

An open-source memory tool that summarizes an AI assistant's conversation history.

As stated in the literature

A summarization node for LLM agent frameworks; in governance-decay experiments its summarization step dropped a standing safety rule in 95% of cases.

Why it matters: It keeps long conversations manageable, but its summarizing can silently discard important standing instructions, which matters for any agent meant to follow safety rules.

For example, after a long chat, LangMem condenses the history into a shorter summary so the assistant can keep going without exceeding its memory limit.

Heard on the show

“… long-horizon agent on any of the mainstream frameworks — and the paper names real ones, LangGraph, AutoGen, LangMem — you are very likely shipping this failure today, without knowing it. …”
Episode 164 — The Summarizer That Quietly Deletes Your Agent's Safety Rules

Mentioned in 1 episode

  1. 164
    The Summarizer That Quietly Deletes Your Agent's Safety Rules

Related terms