The Agentic Builders / New to Agents · 4 of 14
What is Memory (for an Agent)
· 3 min read
New to Agents · Beginner Path · 4 of 14
If the context window gets wiped at the end of every session - and it does - then where does an agent's memory actually live?
The trick is to split "memory" into two kinds, because they behave completely differently.
Working memory is the context window: everything the agent can see right now. It's fast, and it's where all the thinking happens. But it's small, it fills up as the conversation grows, and when the session ends it's wiped blank. (That's the previous article.) Working memory is the desk - useful while you're working, cleared every night.
Durable memory is everything written down outside the window: files on disk that survive the session. Here's the key thing - an agent has no durable memory of its own. You give it one, by writing things down where it can find them again. At the start of each session, the agent reads those files back in, and the brilliant stranger suddenly knows who you are, what you decided last time, and what's still open.
So an agent's real memory isn't inside the model at all. It's in the files. The model supplies the thinking. The files supply the remembering.
Why this matters for Tropo
This is the hinge the whole of Tropo turns on.
Tropo gives an agent its durable memory as plain markdown files in a navigable vault - notes, decisions, status, history - and reloads the relevant ones at the start of every session. The model forgets; the vault remembers. That single move is what lets a crew of agents stay coherent across thousands of sessions, hand work to each other, and pick up exactly where the last one left off. Memory you can read, edit, and trust is the difference between a clever tool and a team you can build on.
Beginner Path · New to Agents ← Before: What is a Context Window · Next: What is a Session → Related: Why markdown files · See the whole path: New to Agents - Beginner Path