Back
productivitymediaOpen source

jev-canvas

Draw and edit a canvas with spoken commands and a pointing finger.

What Jev decides

Whether to act and which action, object, appearance, and position to use.

jev-canvas combines browser speech recognition, webcam hand tracking, and Jev decisions to create or change shapes on a tldraw canvas.

CHOICESCORENOUL
Jev role
Interpret spoken canvas instructions as bounded edit decisions.
Input
Recognized speech, available shapes, canvas state, and pointer position.
Output
Noul, Choice, and size Score answers used to edit the canvas.
Decision frequency
For spoken or typed canvas commands.
Fallback
Noncommands are ignored; mouse and typed input can replace camera and speech input.
Author reported · Source reviewed

How it works

Input
Recognized speech, available shapes, canvas state, and pointer position.
Jev
CHOICE + SCORE + NOUL
Output
Noul, Choice, and size Score answers used to edit the canvas.

Overview

The browser turns speech into text and a webcam image into a fingertip position. Jev decides whether the utterance is a canvas command, then selects the action, shape, color, target, placement, and size from bounded options. Code applies the edit to tldraw; direct pinch-and-drag movement does not call Jev. The project requires an OpenRouter key for Jev decisions, while typed commands and mouse placement can replace microphone and webcam input.

Language
JavaScript
License
MIT
Author
gaborishka
Sources checked Sep 26, 2026

Sources

Related projects