Skip to main content
The more the agent knows about what you’re working on, the less you have to explain. Most of the time, it already knows.

The agent knows where you are

The agent always sees which folder or entry you have open. If you’re looking at a task, “change the status of this task” works — no need to say which one. The same goes for “summarize this document” or “add a row to this dataset”.

Selecting inside an artifact

When you’re viewing an artifact, click a frame or an element inside it. A context pill appears above the composer showing what you’ve selected, and your next message targets it — “make this bigger”, “change this to blue”, “rewrite this headline”.

Attaching files

To hand the agent a file for the conversation:
1

Open the + menu in the composer

Choose Upload to pick a file from your device, or Choose asset to pick one from your library.
2

Or drag a file onto the composer

Drop it anywhere on the chat — the agent will read it. You can also drag assets straight out of the library.
Then ask what you want: “summarize this”, “pull the totals from this spreadsheet”, “match this style”.

What you can attach

Very long documents are shortened to fit the conversation, so for a long report, point the agent at the part you care about. PDFs are read visually: the agent sees the first page right away and can look at more pages when it needs them. That means scanned documents work too — no selectable text required.
Attachments are just for the conversation — they aren’t imported into your workspace. To keep one, ask the agent to save it to your library, or import the file yourself. Removing an attachment chip (✕) only detaches it from the message; it doesn’t delete anything.