AI / ML
Agent Memory
A shared store that lets an agent system retain facts about a brand between sessions: constraints such as a CAC ceiling, stated preferences such as reporting cadence, past decisions and their outcomes, and learned performance patterns. The distinction from chat history is durability and scope. Chat history is one conversation; agent memory is meant to be read by any agent at any later point, so a limit stated once does not have to be restated. Whether a given system actually does this, and for how long, is worth checking before relying on it.
Why it matters
Without it every session restarts from zero and the operator becomes the memory, restating the same constraints indefinitely.
In practice
Test it by stating a constraint once, then asking an unrelated question a week later that should honour it. Whether it does is the whole feature.
Related terms