Ask the model — the AI assistant, step by step

The AI assistant answers questions about the currently open model in natural language, instead of you manually browsing elements and properties.

You ask in plain language (for example "how many walls are in the model?") and the assistant answers based on the model's data. In its answer it can reference specific elements that you then click to find in 3D.

It works read-only — it never changes or exports the model. It's a helper, not a source of truth; verify important figures directly in the model.

• Model from the server: the assistant works only with a model loaded from the server (authorized). It's not available for local files; the panel says so in the "Query context" area.
• Must be enabled: the panel header shows the state AI: active, disabled or unavailable. If it's not active, it isn't enabled in this environment.
• Index preparation: after a model loads, its index may take a moment to prepare. Until then the panel reports that the index is being prepared; then queries unlock.

• Header: the name, AI state, the language model (LLM) in use, and the Clear chat and Close icons.
• Query context: which models are included and whether the AI is available for them.
• Include selection: a toggle that attaches the elements currently selected in 3D to your query.
• Saved context: your saved text files; those marked auto are attached to every query and persist after closing.
• Chat window with history and a query field with the Send and Stop buttons.
• Attachment (📎): attaches a text file (txt, md, csv, json) to a single query, for the current chat only.
• Complete result: for long listings and tables a download card (CSV/JSON) appears under the answer.
• Model interaction: a toggle that lets the assistant temporarily show or color elements directly in 3D on request. A control card then appears under the answer.

On request the assistant can temporarily and reversibly show elements directly in 3D — you don't have to find them by hand. This is governed by the Model interaction toggle; when it is off, the 3D does not change by itself (even if the answer says “I showed…”) and the visualization card under the answer says so and offers a Show in viewer button. The display never changes the model; it is only a temporary visualization.

• Show only: e.g. "show only the spaces" — the assistant hides everything else and keeps just the matching elements.
• Highlight: matching elements are highlighted in color, the rest stays visible.
• Color by group: e.g. "color the spaces by storey" or "color the doors by fire rating" — the assistant assigns each group a color and shows a legend under the answer (color → value → count).

Controls under the answer

A visualization card appears under the answer. In the color legend each group is interactive:

• The checkbox next to a group hides or shows it in the scene.
• Clicking the group name isolates it (shows only that group).
• The zoom icon moves the camera to that group.

The card also has Show again (restores the full display), Zoom (to the whole set) and Clear AI highlight (returns the model to its original state).

Besides the model, you can give the assistant your own text files as context — for example a list of rooms or company abbreviations. Supported formats: txt, md, csv, json.

One-off attachment (this chat only)

The 📎 button next to the query field attaches a file to the current conversation. It shows as a chip with the name and size; the cross removes it. Attachments disappear with Clear chat.

Saved context (tied to your account)

In the Saved context section you add a file with the ＋ button. It persists after closing the window and is tied to your account. Each file has an auto toggle: when on, the assistant attaches it automatically to every query. The cross removes it permanently.

When a result is long or tabular, the assistant offers a download of the complete data under the answer — only a summary is printed in the chat.

• Component list (CSV/JSON): for large element listings you download the full list even though only part is shown in the chat.
• Table (CSV): for an overview table you download its complete image; the card is named after the table.
• Edited file (CSV): when you ask for a data change (for example adding a column from an attachment), the assistant builds a new file and offers it via the "Download edited file" card.

The download link is temporary and valid for a limited time (shown on the card). Numbers are in a format suitable for Excel.

In the Models on the server dialog you can find out what each model contains before opening or federating it. The assistant only reads their AI indexes; no geometry is downloaded.

• Where it is: the assistant is always shown in the right-hand column of the dialog (on a phone below the model list). Only signed-in users with access to projects see it, and only when AI is enabled on the server. It never covers the model list or the Open buttons.
• Scope: the assistant always works with the models of one project, which you select in the library on the left (on narrower screens in the selector above the list). The current scope is always shown in the panel as Authorized scope; access to it is verified by the server. Each project has its own conversation.
• Completeness: when a model of the project has no usable AI index, the panel says so and the answer is not complete for the whole project.
• Model labels: element count, IFC schema and an estimated discipline (indicative only, based on IFC classes), or "no AI index" / "indexing".
• Recommendations: when you ask which models to open (e.g. for MEP coordination), the assistant lists them with a reason. Select recommended ticks them in the list; you open them yourself with Open next to the model or in the recommendation.

The assistant opens and changes nothing. It cannot see models without an AI index and can be wrong, so verify important facts after opening the model.

• Reads only: it never changes or exports a modified IFC.
• Only from the model's data: it answers based on what's in the model (its index). What isn't in the data, it doesn't know.
• Files are data, not commands: the assistant reads the content of attached and saved files as context and does not follow instructions hidden inside.
• Privacy: the query, context and file contents are sent to the server and the language model. Saved context is stored with your account until you delete it. Don't send sensitive data.
• Not for local files: the assistant is available only over an authorized model from the server.