Agentic Voice Control.
Speak an edit, not a prompt. The VoiceOrb parses plain language into a graph operation and applies it to the canvas you're looking at — add a Review node, rewire an edge, set a model, press Run. Hands‑free orchestration.
Your voice is a graph editor.
The VoiceOrb doesn't just transcribe — it resolves your words into a concrete graph operation and applies it in place, the same mutation you'd make by hand.
It mutates the canvas, not a chat log.
Nodes resolve by name. "After Investigate" finds the real node, "wire it" becomes a real edge. Every change is undoable like any edit — and ambiguity is asked back, not guessed.
Speak a command. Watch the graph change.
Press Talk or pick a phrase. The orb runs the full loop — listen, plan, apply, speak — on the live mini-canvas at right.
VoiceOrb state: idle — a gentle breathing sphere.
Press Talk, or pick a phrase, to give the canvas an instruction.
Demo plays a scripted command loop — local voice runs against your real graph. Reduced-motion safe; the orb pauses when off-screen.
Orchestrate without leaving the work.
Building a graph by mouse is fiddly — drag a node, find the handle, drag the edge, open the model picker. Voice collapses that into a sentence. You stay zoomed in on the run while the structure changes around it.
Mid-run edits. Add a Review gate the moment a node looks risky — without pausing to hunt for the toolbar.
Big restructures, fast. "Fan Design and Doc into a Parallel, then Merge" wires four nodes in one breath.
Accessible by default. The whole canvas is drivable without a mouse — every state is announced via aria-live.
Stays private. Transcription can run fully local — speech never has to leave your machine.
Pairs well with
Model Hotswapping
"Run Review on Opus" — voice sets the provider + model on any node, live.
openInbuilt Terminal
Say "run" — or type it. The same commands the orb issues, in a real CLI.
openLocal-network Collaboration
Voice-driven edits stream to every teammate's cursor in real time.
open