Dipson

A chief of staff. Not a chatbot.

The difference is not the model. It is everything around it.

Head Coach asks How do I see player health?
Authority checked · cleared for Head Coach
Three players are limited for Thursday, pulled from the trainer's records.
Guard, starterLimited · ankle, day-to-day
Wing, rotationOut · concussion protocol
Big, depthLimited · load management
Want the full injury report or the practice-plan adjustment?
ReasonedSourcedLogged: who, what, pulled, answered, when, role
Every request runs the same four steps

The order is the point: authority is checked before a single file is pulled, so a restricted question is refused before it is ever answered.

01
Understand the question
Plain language, in any mode. "Who fits our system at the four" is a question, not a search through menus.
02
Check authority
Before pulling a single file. This is the step that makes the same question return a different answer, or a refusal, per role.
03
Pull and run the engines
Only the files and engines the role is cleared for, against the right data: the roster, the film, the recruiting pool, the cap sheet.
04
Answer in the form that fits
Text, a table, a film clip, a scouting report, a recruiting list, a simulation, or a staged action to confirm.
It runs the engine

It does not make up the answer. It runs the engine.

Ask a language model whether a recruit is worth the number, and it will write you a paragraph that sounds like an answer. It generated that. It has no idea.

Ask Dipson, and it does not write anything. It runs the valuation engine, against the roster you actually have, the system you actually run, and the hole you are actually trying to fill, and it hands you a number with a confidence stamp on it and the reasoning underneath.

"Is this recruit worth the NIL number?"It runs the valuation engine.
"What is holding my team back?"It runs the team read.
"Who fits our system at the four?"It runs the fit and the market.
"What happened on that possession?"It goes to the film.
"Who wins on Saturday?"It runs the simulation.

And the outputs are governed, not written. Deterministic scoring, fixed archetypes, and a confidence stamp on every number. No fabricated data. Ever. If it does not know, it says so, and it tells you what it would need in order to know.

Every other assistant in this market is a language model that has read your files. This one is a front door to an engine that reasons.

It compounds

When the coach leaves, the answers stay.

Every coaching change in the history of this business has been a reset. The scheme walks out of the door. The reasons behind a hundred decisions walk out with it. The next staff spends a year rebuilding what the last one already knew, and the players in the middle pay for it.

A question gets askedA player asks why the coverage rotates that way.
It routesIf it is above your authority, or not in the system yet, it goes to whoever can actually answer. A blocking question to the position coach. A rotation call to the coordinator. A judgment call to the head coach. You never have to know who to ask.
The answer is installedThe coach answers it once. Every player who asks it after that gets it straight from Dipson.
And it staysWhen the staff turns over, the scheme does not leave with them. When a player transfers in, he inherits it on day one.

The knowledge belongs to the institution. Not to the last man who happened to know it, and not to a vendor.

Everywhere you are

It is already open, wherever you are.

Mid-filmPull it up on the possession you are watching and ask what happened.
On the touchlineHalftime, in your hand, with the game in front of you.
In the appEvery surface, every tile, every mode. A parent asks about a payment. A player asks what he should work on. An operator asks what the day is forcing him to decide.
And in the OSEvery panel is wired to the same engine. Change one thing and every screen rethinks itself.

Not a window you open. A thing that is already there, in the context you are already in, that knows who you are.

It acts, not just answers

Within your role, Dipson carries out real work across every layer of the OS. The brake is set by what the action costs to undo.

Comms and ops
Send a message, change a schedule, post to Social, update the roster, file a report.
Money
Move funds through KPay for NIL, dues, and settlements, every dollar traceable, every move behind the wallet confirm.
Media
Publish or pull film through KTV, on your authority.
Content
Assign KPlay courses, install plays, or push the GM game.
The confirm brake
Reversible, in-role work executes directly. Anything that moves money routes through the KPay confirm, the same double-press authorization the OS uses everywhere, so the wallet's own gate fires regardless of amount. Other irreversible actions, a roster change or signing an agreement, stage for an explicit confirm before they commit. The same authority that gates what Dipson can answer gates what it can do.
Every action is logged

Two immutable trails: one for the intelligence, one for the dollars. Most conversational AI has neither.

One trail
The intelligence trail
Every conversation logged: who asked, what was pulled, what was answered, when, and under what role.
The other trail
The money trail
Every KPay transaction ends in an immutable audit record. One discipline for the reasoning, one for the dollars.
Not a black box
Governed by design
Outputs are governed by deterministic scoring, archetypes, and legend specs, not free-form generation, and every output carries a confidence stamp with no fabricated data.
The principle

The model underneath is interchangeable. The governance is not.

Swap in any model and the product does not change. What does not swap out is the authority enforcement and the audit trail on every answer and every dollar. That is the line between Dipson and a language model with a login.