The AI model of an implementation
The AI model is a structured description of a specific NET Genium implementation – its application groups, applications, edit forms, view pages, controls, database queries, scripts, permissions and data model. It is derived directly from the unified application model of the platform, so it always corresponds to how the solution is actually built.
The model contains no operational or user data. It describes the structure of the solution, not the contents of the stored records – and that is exactly why it can be safely handed to an AI assistant: Claude or its developer variant Claude Code, ChatGPT or its developer variant Codex, an agent in Microsoft Copilot Studio or any other tool supporting the MCP protocol.
The model reaches the assistant through the integrated MCP server of the instance – the assistant asks for it itself, always in its current form, and with the write permission it also writes its own changes into it. Nothing is downloaded, exported or pasted by hand.
What the model contains
- the structure of the instance – application groups, applications, categories, edit forms and view pages,
- the controls of edit forms and view pages, including their properties, database queries and scripts,
- permissions – user groups and rights groups,
- the data model – tables, columns, data types and required fields in the form of CREATE TABLE statements,
- the materials for performance analysis – the usage statistics of the last 30 days, the record counts of the tables and the index analysis.
The model contains no records or values stored in the database. An assistant that knows only the model therefore understands how the solution is built, but it cannot answer “how many records exist” or find a specific document – for that it needs the permission to read data (see The integrated MCP server).
How the model reaches the assistant
The model is served by the integrated MCP server of every NET Genium instance. The assistant reads it in parts – first the overview of application groups, applications and edit forms, then the detailed inventory of one application or a single edit form – so it works only with what the task at hand needs. Alongside the model it is given the knowledge of the framework NET Genium from the same server, that is the documentation and the conventions of the platform.
The knowledge of the framework is given to every token whose owner has administrator rights – it is not turned on by a checkbox and is unrelated to the permissions for reading and writing data. It is the general documentation of the platform, not the content of the instance, so no permission is being extended. The model of the implementation, on the other hand, describes the particular solution of a customer, and therefore only a token with the permission “Reading all data and the model via the MCP server” is given it; the permission “Writing all data and the model via the MCP server” alone is enough as well, so that the model can be read before it is written.
Because the assistant reads the model directly from the running instance, it always reflects its current state. When the application changes – a form is added, a control is modified, a relationship changes – the assistant sees it with the next request, without any manual step.
Not only reading the model, but changing it
With the write permission the assistant also changes the model: it creates application groups, applications, categories, edit forms and view pages, adds and modifies controls, database queries and scripts, creates user groups and rights groups, and renames building blocks without touching their identifiers. The change is written directly into the instance and shows in the portal immediately.
The model is thus the workbench of agentic development – the agent reads the actual shape of the application, proposes a modification and, once approved, writes it back. How writing the model works technically is described in the text The integrated MCP server.
Typical use
- finding your way around complex implementations – how an application is built and which tables and columns belong to it,
- explaining the architecture of the solution and onboarding new developers or administrators,
- proposing modifications in line with the actual structure of the application rather than with a guess,
- consulting on the extension of the application and analysing an assignment,
- performance analysis – which pages, queries and scripts take the longest and where an index is missing,
- agentic development – building and maintaining applications with an AI agent that both reads and writes the model.