Type @ in any agent message to reference a canvas element. The element becomes part of the prompt context — the agent sees its image, video, or note content alongside your text.
How it works
- In the agent chat input, type
@. - The popover lists every element on the current board, filtered as you type.
- Click one (or press Enter on the highlighted match). The element appears as a chip in the input.
- Send the message.
The agent receives the chip as a reference and treats it as:
- Image — attached as a vision input for the LLM, or as a reference image for image-to-image generation.
- Video — first frame attached as vision; the agent can also call frame-extraction tools to read additional frames.
- Note — pasted as text content into the prompt.
- Frame or group — every child element attached, in canvas order.
Multiple mentions
Stack several mentions in one message:
Build a moodboard collage from @hero, @cta-button, @background-render and place it in the top-left frame.
Each mention becomes part of the agent’s context.
Selection as implicit mention
If one or more elements are selected on the canvas when you send a message, the agent sees them too — even if you did not @ them. Useful for “rewrite this note” when the note is already selected.
When to use mention vs select
- Mention when you want a specific element and there might be multiple selected.
- Select when you want everything currently selected (useful with multi-selection).
Limits
- ~10 mentions per message before token cost gets steep.
- Mentioned videos cost more than mentioned images (more frames in context).
- Mentioning a large note (multi-thousand words) eats context window the same as the same words pasted.
Where to go next
- Giving Commands
- Agent Types
- Mentioning Canvas Media (generation panel) — same feature in the image / video generation panels.
Leave a Reply