We're building autocomplete for canvas

so raw and so good, keep sharing!

zoom out style as hell

@maxbittker Pegasus by Takeo Igarashi might be a good reference

This is so cool! What is the input to the AI? Is it the underlying representation of the tldraw diagram? Is it/will it be open source?

Depending on the settings, the input is an xml representation of the drawing plus an image of the drawing, with a prompt to respond (using the xml format) with the best guess as to what the user will do next

Feels like magic to me 🪄

Im confused @violetto96 will break this down whats happening understand the hood

Amazing stuff thanks for sharing

Love seeing this! Have you considered adding an optional text prompt to provide additional context/guidance. Similarly, you could have it transcribe whatever the users says and use that too

That’s cheating! (but yes)