Start here
By product
When it goes wrong
Looking for what a specific tool does? The toolbox lists every one of them.
Start here
How to ask
The assistant can see your document. It cannot see what you meant. Almost every disappointing answer comes down to one of four things — scope, units, verification, or asking it to guess.
Name the scope
“The model”, “this view” and “my selection” are three different jobs. Say which one.
Vague“Set the fire rating on the walls”
Better“Set Fire Rating to 60 on the walls in the current view”
Four walls or four thousand? Editing tools refuse a whole-model default, so vagueness only costs you a round trip.
Vague“Renumber the doors”
Better“Renumber the doors on Level 2 by their position along the corridor, starting at 201”
Give the order and the start number, and you get a result instead of a question.
Say the units
A bare number is ambiguous until it’s wrong. Write the unit — 1200 mm, 0.5 m, 4 ft — and it converts instead of assuming.
Vague“Make the offset 250”
Better“Make the offset 250 mm”
The APIs work in feet internally; the assistant converts once it knows your unit.
Let it look before it acts
Touching more than a handful of elements? Ask for the list first — you catch a wrong filter before it’s applied.
Vague“Delete the unused view filters”
Better“List the view filters nothing uses, then delete the ones I confirm”
One extra message, and no more “that’s not what I meant”.
Ask it to report back, not to reassure you
A good answer says what changed and what didn’t. If you only get “done”, ask for the breakdown — the tools return per-element success and failure.
Vague“Did that work?”
Better“How many did it change, and which ones failed and why?”
The second is answered from what the tool returned. The first invites a guess.
Undo works — use it as your safety net
Every write is one named transaction — one step in the host’s own undo stack. Not what you wanted? Press Ctrl+Z, as for anything else.
Give it a target, not a puzzle
Faced with ambiguity, the assistant picks a sensible default and gets on with it. Usually that’s what you want — but a truly vague request gets a confident answer to the wrong question.
Vague“Clean up the model”
Better“Show me the warnings grouped by type, then let’s work through the top three”
“Clean up” has a dozen readings. Name the first move.
When there is no tool for it
The named tools aren’t a ceiling. Ask for something outside them and the assistant writes C# against the live document and runs it — a custom report, a one-off bulk edit.
“There’s no tool for this: write code that lists every family type used on fewer than three instances, with its category.”
Code is screened first against a blocklist (files, processes, network, registry). Need a fixed tool surface? Switch the feature off — everything else keeps working.
Keep the phrasings that work
Once a request works, save it. The Quick Prompts tab keeps the wording under a name for reuse.
Saved prompts, with the number of times each has been used. The ones marked “fill-in” ask for a value when you run them.
The useful part is fill-ins: write {{category}} or {{value}} and it asks each time — so “Find every {{category}} where Mark is empty” is one prompt, not one per category.
Tell it once, not every time
Repeating the same context — parameter naming, the standard titleblock, Turkish level names? Put it in Memory; it loads into every chat.
Memory has two scopes. General applies everywhere; the project one only loads when that Revit document is the active one.
General is how you work — conventions, preferences, reporting style. Project is facts about one model, loaded only when it’s open, so nothing leaks between jobs.
Write in your own language
Ask in English, Turkish, German, Spanish, Italian and more — replies follow your language. Tool and parameter names stay in English, as they are in the applications.
Next
Worked examples for your product: Revit, AutoCAD & Civil 3D, Navisworks, Inventor, 3ds Max or Robot.