Agent memory is information saved outside the immediate prompt. It can include user preferences, task state, past decisions, files, summaries, or long-term notes.

In practice

Memory helps agents continue work over time. For example, a coding agent may remember project conventions, while a support agent may remember customer context.

What to watch

Memory creates privacy and accuracy risks. Stored information should be inspectable, editable, permissioned, and easy to delete when needed.