Designer assistant

The designer assistant helps whoever builds a project in Studio: it explains a script, fixes an error, writes a first draft, and answers questions about the project. It reads the project and the live gateway as you, so it sees only what your role can see. It cannot change anything: a change it wants becomes a proposal that a person approves.

You need the AI Use capability, a gateway whose Designer assistant switch is on (Turning Iris on), and Studio connected to that gateway. Without these, the Iris dock says Iris is not available and the editor menus show nothing.

The Iris dock

Open View › Iris. Type a question and press Enter (Shift+Enter adds a line). Iris is told what you are looking at in Studio, such as the workspace, the resource open and the text you have selected, as background only. It looks things up with its tools, and the dock shows Iris is looking something up… while it does.

Every reply is labeled Iris (AI-generated), and the dock reminds you that Iris can be wrong. Stop abandons a call, and New chat starts afresh. Replies are shown as plain text: markup and links in them are not interpreted.

When Iris wants to change a project file, the dock shows a card such as Iris proposed a change: projects/Plant/screens/Home.json, with Preview and Open in Proposals. The card says Nothing has changed, and that is true: the change waits in the Iris Proposals dock for a person to approve it. Iris can propose an edit to an existing screen, script or named query, or a new one of those. It cannot propose changes to anything else.

In the editors

In a script, named-query or expression editor, right-click and choose Iris (AI):

Action	Shortcut	What you get
Explain	Ctrl+Alt+E	A plain-language explanation. Nothing to apply.
Fix	Ctrl+Alt+F	A corrected version, using the editor's current error message.
Generate	Ctrl+Alt+G	New text from your comment or selection.
Complete	Ctrl+Alt+C	A continuation at the cursor.

Fix, Generate and Complete show a diff. Nothing in your editor changes until you accept it, and accepting is a local edit like any other: save it the way you save anything.

Every suggestion is checked before you see it. A script is compiled (never run), an expression is parsed by the real expression engine, a query must be a single read-only SELECT, a screen must be valid JSON of known components, and tag paths are looked up. A suggestion that fails goes back to the model once with the errors; if it fails again you see the errors and no suggestion. A warning (an unknown tag that may be created later, for instance) does not block.

What it sends

The text of the editor or message, the resource you have open, and whatever its tools look up for you. API keys, passwords and tokens are removed before anything leaves. Data, limits and records has the full list.

Limits

• An editor's text is limited to 64 KB, a selection to 16 KB, and a message to 64 KB. Over that, the request is refused before any model is called.
• It does not run your scripts, queries or expressions.
• It is a drafting aid. Review what it produces before you rely on it, and test it as you would anything you wrote yourself.