Registry · ActiveFiling /MAI-001Issued 29.vii.2026
§ 01The matter at hand

Subject

The documentation gap.

Enterprise AI systems operating without governance record.

Aug 2026 — four months

EU AI Act enforcement.

Nobody knows what their AI is allowed to do.

An architecture design workbench where the diagram you draw is also the governance record — no separate documentation pass, no PDF that goes stale the day it's exported.

§ 02The argument

Position

Governance should be a property of the design, not a layer bolted on afterward.

The design is the deliverable.

Every architecture team already has diagrams — in Lucidchart, Miro, draw.io. They get exported to PDF once, attached to a compliance document, and left there. Six months later the system has changed twice and the diagram hasn't.

What a regulator actually asks for is more specific than a picture: which models are deployed where, what authority each component holds, where a human has to sign off, and what data classification moves across each connection. Those are questions about the system itself, not about a drawing of it.

Treating the design as the record means every component you place declares its authority, the data it touches, and when a human needs to be involved — not just its shape. Once that's true, the diagram and the documentation stop being two separate things.

In the field

Most enterprises keep their AI architecture in three different places — a diagramming tool, a Confluence page, a risk register — and none of them talk to each other.

What this is not

Lucidchart, Visio, and LeanIX all produce documents about your system. ModelAIr produces a record of it.

§ 03The mechanism
/01Typed node schemaComponents as structured objects.Every component carries real properties — model family, hosting environment, data residency, access controls — rather than being a shape you drag onto a canvas.
/02Assertion levelsSignal. Advisory. Authoritative.Signal, Advisory, and Authoritative define what a component is permitted to do, not just what it's technically capable of. That's the distinction auditors care about.
/03Authority boundariesWhere autonomous authority ends.You mark exactly where AI can act without sign-off and where a human has to step in. Both show up on the diagram and carry through to the export.
/04Edge semanticsConnections that carry meaning.Every connection carries its own protocol, authentication method, and data sensitivity classification, so the wiring is governed the same way the components are.
/05Governance outputThe diagram is the artefact.There's no separate export step — the workspace you design in is already the governance record.

Status

Beta · Invite only

Capacity limited


For architects under governance pressure.

Built for architects, not curious observers.

If you're under pressure to produce AI system documentation that doesn't exist yet — for the EU AI Act, an internal audit, or a board that wants answers — get in touch.

Request accessNo commitment · Early access only