The list. The rulebook. The paper trail.
graf finds every AI agent in your estate, names who is accountable for each one, and states — at every answer — how much of the estate that answer actually covers.
The board above is an illustration of the interface, not a screenshot of a customer estate. The figures are invented. graf does not publish a number it cannot show you the provenance of, and that rule applies to its own marketing.
The second number is the one that matters.
Most inventory tools lead with what they found. graf leads with the agents nobody is answerable for — and then tells you that even that number is a floor, because some of the sources that would have found more are not reading.
A finding, never a blank. An agent with no accountable owner gets a redline, a row tint, and the words no owner identified. There are no empty cells anywhere in graf.
Both numbers, always. Last sync 41 days · source retains 30 days. Never “sync error”. Any count resting on that source carries the note.
Removed from the denominator. If no tool gateway is deployed in your estate, that is not a gap someone failed to close — and it does not drag your score down forever.
Three visuals. Nothing else.
The whole estate is never rendered as one graph, because a hairball is not an answer. Everything in graf is one of three things.
The schedule
Every agent in scope, with ownership that expands in place. The daily view — and the surface you land on when you open the product.
The figure
One finding, drawn as three to seven nodes with the inference basis stated on the drawing. Hard-capped at seven, because a picture you have to trace with a finger is not evidence.
The survey
Sources against org units — what is observed, by what method, and what remains dark. This is the honesty, rendered as a grid you can hand to an auditor.
The shared slice
A live link scoped to one stakeholder's org unit. They see their agents, their coverage, and an explicit list of what they cannot see.
“Not surveyed” and “surveyed, none found” never share a string.
One is a coverage gap. The other is a finding of fact. Render both as 0 and you have built a tool that produces confident wrong answers — which is how they get into board decks.
Hatched ground, the reason named, never a zero. It tells you where to point your next connector.
Plain, unhatched, unremarkable. It is an answer, and graf renders it as one.
Six capabilities. Two of them are deliberately empty.
A control plane claiming all six on day one would be claiming enforcement authority over systems it does not own. We would rather tell you which two are empty than have you find out in the pilot.
We are opening graf to a small group at a time.
Design partners get the connector set built against their estate, and their findings shape what v1 reaches.
Thank you. This form is not yet connected — please write to hello@getgraf.ai and we will reply.
No newsletter, no drip sequence. One human reply.