Your own AI application — what it learns about your model

When you connect your own AI application (Claude, ChatGPT or another) to roleALPHA over OAuth, it does not only read data and make proposals — it also receives the knowledge it needs to model sensibly. This page explains exactly what that is and how to look it up.

Why this is needed

A relationship in roleALPHA is two things at once: a statement about reality and what the interface draws. If an application picks the wrong relationship type, the data is correct — and the view still stays empty. No error message appears when this happens.

Without this knowledge a connected application would only get the list of available tools. It would know neither which entity types your tenant keeps, nor which relationship types exist, nor which way round they are meant.

What the application receives when it connects

Three things, right as the connection is established:

The modelling rules. Specific relationship types before generic ones; which way round hierarchies run; what effect an edge has on the picture. The same rules rAlph works by.
Your entity types. Which ones exist, which module each belongs to, and which fields they have — including required fields and allowed choice values. A field that is not listed does not exist.
Your relationship types. With their allowed source and target types, their reading direction, a marker for generic placeholders — and the rendering effect: whether the edge orders the layout, opens a box, fills a card row or carries the path.

What it can re-read at any time

The same two overviews are permanently available as resources, at rolealpha://modell/entitaetstypen and rolealpha://modell/beziehungstypen. The difference: what arrives at connection time arrives once, whereas a resource can be fetched again hours later without rebuilding the connection.

There are also three tools the application can call on its own:

Explain relationship types — what a type means, which way round it is meant, which entity types it connects.
Explain rendering effects — what an edge does for the picture and what would stay empty without it.
Check rendering — after modelling: which entities stay empty, and which relationship they are missing.

When the application creates a relationship it is also told directly what is still missing in the picture afterwards. That happens even if it never read any of the above.

What it does NOT receive

Nothing you have not released. The same module release applies as for rAlph (Settings → AI). A module that is not released appears neither in the catalogue nor in a resource.
No data from another tenant. The tenant follows from your grant, not from what the application claims.
No write permissions you do not have. And in any case writing only creates a draft — a human has to approve it.

If something is missing

If the application cannot see an entity type or relationship type, check the module release under Settings → AI first. After that, reconnecting helps: the catalogue is built when the connection is established, so a newly created entity type only appears in the next session — or immediately if the application re-reads the resource.