memory
your agent writes down what it learns about you, on its own machine, and reads it back in every conversation. you can look at what it has written, and you can erase it. it starts knowing nothing and gets better at being yours the longer you use it.
an assistant that forgets you after every conversation is a search box with manners. yours keeps notes. when you tell it something that matters, it writes it down on its own machine, and the next time you talk, in any channel, it reads those notes before it answers. this page is what it keeps, where, how you see it, and how you take it back.
what does it remember?
things you tell it and things it learns while working for you: who you are, how you like things done, the people and places that keep coming up, what it tried last time and whether that worked. it writes these to a memory file and a memory folder on its own computer, and it reads them at the start of every conversation.
it does not remember by magic and it does not remember everything. it remembers what it decided was worth writing down, which is why the first week matters: the more you tell it, the more it has to work from.
where does that live?
on your agent's own machine, in files it owns. not in a shared model, not in another customer's environment, and not in our training data. privacy and your data covers the whole picture; the short version is that what your agent knows about you stays with your agent.
how do i see what it knows?
two ways, and they are worth using together.
ask it. the fastest way, and it answers from the same notes it works from:
what do you know about me so far?
what did you write down after our conversation yesterday?
look at the memory page. the dashboard has a page where you can read what your agent has recorded, at instaclaw.io/memory. it starts nearly empty, because a new agent has not learned anything yet, and you can watch it grow from there.
is it the same memory everywhere?
yes. one agent, one machine, one set of notes. tell it something on Telegram and ask about it on the web dashboard; it answers from the same memory, because there is nothing to sync. channels has the two-message test if you want to see it for yourself.
how do i correct something?
tell it. a correction is just another thing to write down:
that's out of date, i moved to porto in june. update what you have on me.
it rewrites the note rather than adding a second one that contradicts the first. if you want to check, ask it to read the entry back to you.
how do i erase it?
from the dashboard: reset agent memory. it removes every conversation, everything it has learned about you, the identity it built for itself and its memory folder, and it keeps your wallet, your connected channels, your plan and its skills. there is no undo. if there is something you want to keep, ask it to write that to a file first; privacy and your data has the exact list of what goes and what stays.
what memory is not, yet
it is not a graph you can browse by topic and date on day one, and it does not carry a long history from before you arrived, because there is none. it is a set of notes that starts thin and thickens. we would rather say that plainly than describe a mind palace and hand you an empty room.
what next
- talking to your agent: how to give it things worth remembering
- privacy and your data: where it lives, who can see it, what erasing does
- your agent's computer: the machine those notes live on
- command center: the page your agent builds from its memory of you
Reading this as a machine? Get the raw markdown.