Memory
Your AI team is only as good as what it remembers. So memory is the deepest part of how TagTeam works. Here is the whole model, in plain terms.
What memory holds
Your Clone (your personal AI worker, built to act as you) needs to know things to act for you. It needs to know who you are, what you are trying to do, and what is true. All of that lives in memory.
Memory is not one big pile. It is organized so the right thing is easy to find, easy to keep current, and safe to change. Two ideas do most of the work: three stores and, inside each store, three levels.
The three stores: who the memory is about
A store is a separate box of memory, scoped to who it is about. There are three.
| Store | What it holds |
|---|---|
| Your Clone scope | Personal: who you are, your goals, how you work, what you are doing now |
| Your company scope | Company-specific: what the company sells, believes, sounds like, and is aiming at |
| The system's own scope | How TagTeam itself works and what state it is in |
Keeping them separate matters. Your personal notes do not leak into the company picture, and the company picture does not get tangled with how the system runs itself.
The three levels: how settled the memory is
Inside every store, the same three levels sort memory by how settled it is.
| Level | What it is |
|---|---|
| Active | The current operating picture: your profile, your goals, your plan. Read first, every time. |
| Adaptive | The capture layer where new signals land and get sorted. |
| Archival | Durable, confirmed truth. Edited rarely, and never without your sign-off. |
Active is the working picture your Clone reads before it does anything. It is kept short and sharp on purpose. If it gets bloated, it is trimmed, and stable parts move down to Archival.
Adaptive is the inbox for new information. A signal shows up, lands here, and gets sorted: is it a fact, a decision you need to make, a change, or something that conflicts with what we believed? Items here are temporary. If nothing confirms them, they fade.
Archival is the long-term record. Things only reach it after they have been confirmed, and nothing lands there without your confirmation.
Where it came from, and whether it is still true
Every piece of captured memory is tagged with where it came from: which scope, which source, and how confident we are. That tagging is what lets the system sort and check memory without re-reading everything from scratch.
Memory is also swept for staleness. Old signals that go nowhere are summarized if they hold durable truth, or dropped if they do not. The capture layer stays a moving window, not a junk drawer.
Most important: memory is checked against reality every cycle. This is the alignment check. It asks a simple question of each thing in your Active picture: does this still match what the signals are showing? When something has drifted, the system does not quietly fix it. It proposes an update and waits for your approval. Your foundational memory is never silently overwritten.
Built to run continuously, running daily today
The capture and the cycle that processes it run daily today, triggered. They are built to go continuous when live connectors and scheduling arrive. The engine is already designed for always-on operation. It simply switches on when those two pieces land, with no change to how it works.
So to be candid: today the system reviews your world on a daily rhythm, not in real time. That is the design choice for now, not a permanent limit.
You stay in control
Two safeguards keep you in charge of your own memory.
First, ordinary background writes are reversible. Routine notes, captures, and small plan updates can be rolled back for about a week, which is seven daily cycles. If the system records something it should not have, there is time to undo it.
Second, nothing foundational changes without you. This is the Review Gate, the checkpoint every meaningful change passes through. It runs on three approval tiers: small reversible work just happens in the background, slightly bigger work happens and gets reported to you, and anything foundational or outward-facing waits for your yes before it moves.
Changes to your profile, your goals, and your confirmed truths always sit in the third tier. They arrive as a clear before-and-after for you to approve. That is why you are never locked in: your Clone gets sharper every cycle, and you stay the one who signs off.
TagTeam is built on Anthropic's platform.