What is agent memory?

Agent memory is information an AI agent saves outside its context window and loads back into later sessions, so it can reuse facts, preferences, and past work without being told again. The context window only holds the current session, and a new session doesn’t start with what the last one saw; memory is the part that carries over.

As of October 2026, ChatGPT’s memory remembers context from your chats, files, and connected apps and shows it as a memory summary you can edit, per OpenAI’s Memory FAQ. Claude’s memory, per Anthropic’s help center, saves topics from your chats as you talk, and each project gets its own memory. Anthropic’s Claude Code keeps memory in files: CLAUDE.md files you write, plus auto memory notes Claude writes from your corrections, and both load at the start of every session.

Developers building their own agents get the same thing from frameworks and APIs. LangChain’s LangGraph keeps short-term memory per conversation thread and long-term memory in a store shared across threads, and Anthropic’s memory tool lets Claude read and write files under a /memories directory that your own application stores.

Whatever the agent pulls from memory arrives as context, so it shares the same window and the same limits as everything else in it. Claude Code’s docs say Claude treats memory files “as context, not enforced configuration”: a remembered rule can still be ignored, and a stale one keeps loading until it’s edited out.