August 26, 2026
Genvi Agent builds on your canvas while you watch
Describe what you want and the agent does the work on whichever surface you already have open, then tells you what it actually did.
We didn't want to bolt a chatbot onto the side of the app. The agent drives the same stores your mouse does, on whatever surface you're looking at, and you watch it happen in front of you.
It works where you are
Ask on the node canvas and it adds and wires nodes. Ask on the whiteboard and it draws. Ask in the video composer and it places clips. The same sentence means different work on different surfaces, which is the whole idea: you say what you want, and you don't have to know how it gets done.
Each step shows up as it happens. When it finishes it re-reads the canvas and reports what it actually did, not what it meant to do.
It tells you the cost before it spends anything
Anything that costs credits gets estimated first and reported after. The agent asks before it starts, and the number it shows you comes out of the same table that does the billing, so it can't drift from what you're charged.
It already knows your references
Your library is part of what the agent can see, so "use Nova for the hero shot" works without typing an @. When it generates, it uses the description and pictures you wrote once, weeks ago, and had forgotten about.