Account memory
The difference between a tool and someone who knows the account.
Everything that makes your account yours lives in someone’s head: the client who will not run video, the market that looks unprofitable and is not, the suggestion you have now rejected four times. None of it is in the ad platforms.
It is why a new hire takes a quarter to become useful, and why every generic AI tool still sounds like a stranger on day ninety.
Adgent picks these up as you say them — and every verdict after that is written by something that already knows them.
There is no setup screen, because nobody fills one in.
Every tool that asks you to declare your rules up front gets a half-finished form and then quietly ignores it. You do not know your exceptions in the abstract — you know them when something violates one. So that is when Adgent picks them up: from what you say in chat, and from what you reject.
It shows you what it learned and asks before keeping it, and you can override any of it. Memory you cannot see is not memory, it is drift.
The rule outlived the session that set it.
Memory is what turns an audit into a judgment.
An audit applies general rules to your account. A judgment applies your rules to your account — and that gap is the whole reason a good freelancer is worth more than a good checklist. This is the least visible thing on the site and the one that decides whether people are still using the product in month three.
- “Consider consolidating these campaigns” — they are split by client, deliberately
- “This market is underperforming” — it is the one with the highest margin
- “ROAS is below target” — that is not the number you judge on
- Every finding needs the same context typed underneath it, again
Technically correct and wrong about your account. On week eight it is saying the same four things.
- The consolidation suggestion is gone — you rejected it in June, and said why.
- That market is read on margin, because you told it once that it should be.
- Reports reconcile to store revenue first, with platform ROAS beside it as context.
- The brief is shorter, because what it already knows is not being re-argued every morning.
Same product, same account. The difference is eight weeks of having been told things.
The fastest way to make someone stop reading a daily brief is to put the same dismissed recommendation in it every morning. Rejecting something twice is information, and it gets treated as information.
Tell it once that this brand is judged against store revenue rather than platform ROAS, and every report after that reconciles to the store first. Not a per-report setting — a fact about the account.
"Nothing scales past this without flagging me" is a rule that has to survive to the moment a change is proposed, six weeks later, on a campaign that did not exist when you said it.
Memory is the one thing here that only accrues by starting. A quarter of deferring is a quarter of re-explaining the account every session — the same exceptions retyped, the same rejected suggestion arriving again on Tuesday, the €500/day limit living in someone's head instead of in the thing proposing the change.
The context you would have by March is not something you can buy in March. It is only worth what it has been running.
The context you would have by March cannot be bought in March.
It is only worth what it has been running. A rule you set in July is still holding in September, on every account under the brand.
- 1sentence to set a limit, once
- 0setup screens, because nobody fills them in honestly
- bothplatforms — the rule crosses Meta and Google without restating
Yours, and it follows the brand.
The context belongs to the account, not to one ad platform's view of it. Your rules travel with the brand across every platform you connect, so a decision made about a campaign on one platform is not forgotten when the same question comes up on another.
Each company's data sits in its own database file rather than a shared table with a customer column in it. The separation is structural, so what one account taught Adgent cannot leak into another's verdict.
Memory is the layer the verdict stands on.
Without it, every morning read starts from zero and reads like a stranger's opinion of your account. With it, the brief already knows what you told it last month.
Say it once. See it come back next month.
Give us the rule you are tired of repeating — the scaling limit, the metric you actually judge on, the suggestion you keep rejecting — and watch where it surfaces.
- Learned from how you talk to it, no setup screen
- Shown to you and confirmed before it is kept
- Follows the brand across Meta and Google
What Adgent would pick up from your account without being told, and what it would ask you to confirm before remembering.
Thanks — we’ve got it.
We’ll be in touch within a couple of working days.