Graybox
Model any company. See where AI agents fit.
Every company is a graph — Graybox makes it queryable.
Graybox is an organization meta-ontology: a single, typed model that captures how a whole company actually works — its structure, its work, the resources that flow between activities, its goals, capabilities, governance, and knowledge — across eight facets at once. Point it at a real business and you get one operating graph instead of a drawer full of slide decks.
The poster shows a model of Voltforge Prototyping, a 45-person quick-turn electronics factory, rendered as 369 typed entities and 57 activities wired together by the resources they hand off down the line.
Where do AI agents actually fit?
That is the question Graybox is built to answer. Every activity in the model is scored on an autonomy ladder — from inform and recommend up to act & report and fully autonomous — together with the one constraint that is keeping it from going further.
The result is uncomfortable and useful: for Voltforge, 7 activities already run autonomously, but for the largest blocked group the binding constraint is access & actuation — no tool surface, scattered data — not whether a model is smart enough. Graybox names that blocker, per activity, so an operator can see exactly where to point automation and what to fix first.
It is local-first and git-native: every organization is a folder of plain JSON, diffable in version control, with no hosted backend to trust.
A self-contained, seeded interactive version (switch lenses between agent readiness, operating flow, and org structure; hover any node to see who owns the work and what is blocking its delegation) is available as a live embed — no login, no customer data.