An operator that remembers, on your terms
OperatorNest keeps a memory so you don’t have to repeat yourself. Tell it once that you prefer window seats, that your Friday reports go to a specific list, or that a certain client always gets a formal tone, and it applies that on the next task without you restating it. You can open that memory at any time, read exactly what’s in it, and edit or delete any part of it.
What editable memory changes
An assistant that remembers things with no way to check or correct them asks you to trust it blindly. One with no memory makes you repeat yourself and never gets more useful. Editable memory gets more useful the longer you use it, and you can check what it knows at any point.
What ends up in memory
Memory holds what makes repeat work faster: preferences (“always book aisle seats”), standing instructions (“cc my assistant on client emails”) and context (“Acme’s renewal is in March”). It builds up as you work with your operator, and you can add or correct entries yourself.
Your operator keeps what’s useful for future tasks, with no raw log of every message and document. Anything sensitive it comes across during a task stays with that task and isn’t folded automatically into long-term memory.
Reading and correcting what it knows
Your memory is a list. Open it and you’ll see entries like:
“Prefers replies kept under 150 words.” Added from an edit you made on March 3.
If an entry is wrong or outdated, correct it and your operator uses the corrected version from then on. Fixing one entry is faster than talking an assistant out of a bad assumption over several messages.
Deleting what you don’t want kept
Delete a single entry, or clear memory for a task or a whole operator, at any point. A deleted entry is gone from what your operator uses from then on. If you’re winding down an operator, or want it to forget a finished project, you delete that memory yourself, with no request to file and wait on.
Exporting your memory
You can export what your operator remembers as a plain file: for your own records, to document how an operator is set up for a team, or to move what one operator knows to another. Exporting is a copy and leaves your operator’s memory as it was.
Memory and model choice
Memory belongs to your operator and to you, not to whichever AI model runs a task. Switch a task from one model to another, covered in model choice, and the memory attached to that operator carries over unchanged.
A concrete example
Say you tell your operator once, “our refund policy is 30 days, no exceptions without my sign-off.” Weeks later, a customer follow-up task needs that policy. Your operator applies it from memory and still holds the refund itself for your approval, since that’s a consequential action whatever it remembers. If the policy changes to 45 days, you open that entry, edit the number, and every future task uses 45.
Separate memory for separate operators
If you run more than one operator, for example one for research and one for scheduling, each keeps its own memory by default; see multiple operators. Your scheduling operator doesn’t see the details of a research project unless you set up sharing.