Skip to main content

Ask the Assistant

Goal​

Type what you want in plain words and have it answered or done: "which tables does this environment have?", "add a user for Sara with the Maker role", or "build a solution for tracking equipment loans — items, loans, and an app".

Prerequisites​

  • An AI provider is connected — see Connect an AI provider. Until it is, the box says so and points you at Settings → AI.
  • An environment is selected in the top bar. The assistant works inside that environment.

Steps​

  1. Open the home screen (the Home entry at the top of the left navigation).
  2. Type your request into the large box under Ask the assistant. Several lines are fine — describe the tables you want, what each one records, how they relate. Press Ask (or Ctrl+Enter).
  3. Watch the transcript. The assistant says what it is doing, and every look-up or change it makes appears as a line naming the tool it used — list_tables, create_table, add_field — with what it was given.
  4. Read its summary at the end. It names what it found or made, with each thing's logical name in parentheses so you can find it in the solution afterwards.
  5. Ask a follow-up in the same box; the conversation continues. New conversation starts over.

Allowing changes​

The assistant starts in read-only mode: it can list environments, solutions, tables, fields, forms, views, option sets, apps, users, roles and teams, and explain them — but it cannot create or change anything, and it will tell you what it would do instead.

Tick Allow changes before asking when you want it to act. With changes allowed it can:

  • create publishers, solutions and environments;
  • create tables, add fields (including lookups, which also create the relationship), create option sets, and lay out a table's default form and view so new fields show;
  • create model-driven apps listing tables;
  • create users, roles and teams, and give a user a role.

It cannot delete anything, and it cannot do anything you could not do by hand: every change goes through the same platform calls the screens use, as you, so your own permissions apply.

Building a whole solution​

Say what the application is for and let the assistant choose the details. A good request names the things the application records and how they relate:

Build a solution for equipment loans. Items have a name, a serial number and a category (Laptop, Camera, Projector). A loan is for one item, borrowed by a contact, with a start date, a due date and a status (Requested, Out, Returned, Overdue). Make an app that opens on loans.

The assistant creates the solution, the tables, the option sets, the fields, lays out each table's form and view, and creates the app — then tells you what it named everything. Open the solution afterwards to adjust anything: what it built is ordinary components, editable in the ordinary designers.

What happens behind the scenes​

The model never touches the platform directly. It asks for a tool by name; your browser runs it as you, through the same API a screen uses, and sends the result back. That is why the transcript can list every step, and why nothing happens that you could not have done yourself.

It stops after forty steps. A whole solution is twenty-odd; a request that is still going at forty is stopped and you are asked whether to continue. Stop ends a run sooner; nothing already done is undone.

It answers in your language. The reply arrives in the language the Admin Control Center is set to.